The first Space Race was an intercontinental ideological battle that eventually culminated in one of the most complex acts of cross-border cooperation: the ISS. By comparison, the new ‘space race’ is just a billionaire dick-measuring contest, and, like most men who peaked twenty years ago, some are struggling to even get it up (i.e., Starship). Now, Blue Origin is beginning to overtake SpaceX, much like the tortoise to the hare. However, unlike Starship, their New Glenn rocket (which is Starship’s direct competitor) is actually walking the walk. Last Sunday, Blue Origin launched its New Glenn rocket for the third time, and while it experienced some rather major hiccups, it once again proved that it is miles ahead of Starship. But should we celebrate Musk taking an L here?

NG-3

Let’s start with this launch, which was called NG-3, because despite it being called a test launch, it wasn’t really a ‘test launch’. NG-3 carried a paying customer’s payload, a gigantic AST SpaceMobile cellular internet satellite called BlueBird 7, which New Glenn was tasked with taking to Low Earth Orbit (LEO).

Subscribe now

However, NG-3 did have other objectives; namely, this was the first time they reflew a booster. From a conceptual standpoint, New Glenn is essentially a larger, more powerful and more efficient version of SpaceX’s Falcon 9 because it is partially reusable, with the upper stage/fairing being a consumable part and the booster using retro-rockets to safely return to Earth for future use. Blue Origin failed to land the booster from the first launch but successfully landed it on the second. NG-3 used that landed booster, making it the first time a New Glenn booster had been reflown. Proving reusability was a key target for this mission, which is exactly what they did.

For the most part, NG-3 went well. It left the launch pad, the stages separated successfully, the booster flew back and landed safely, the upper stage reached orbit, and the payload was successfully delivered to orbit. It was a nearly perfect mission, except for one giant caveat.

The payload was delivered to the wrong orbit!

As TechCrunch reported, just two hours after launch, Blue Origin announced that it had placed the $30 million, 6.1-metric-ton, 233-square-metre satellite in an “off-nominal orbit”. Then, AST SpaceMobile released a statement that New Glenn had placed the BlueBird 7 satellite into an orbit that was “lower than planned”, that this orbit was too low “to sustain operations”, and that the giant BlueBird 7 satellite would need to be deorbited as soon as possible.

That is… less than ideal and does mean that Blue Origin failed its primary mission.

Since then, the FAA has grounded New Glenn and launched an investigation. That might sound alarming, but it’s just standard procedure. Plus, Blue Origin has already admitted to its mistakes. Blue Origin’s chief executive, Dave Limp, claimed that this error occurred because of a lack of “sufficient thrust” in a single engine — which is, most likely, a fixable issue, given that these engines have worked well enough before.

In Context

Delivering a satellite to the wrong orbit might sound like a biggie, and I’m sure it is to AST SpaceMobile, but in the grand scheme of things, NG-3 was still a sizeable leap forward for Blue Origin.

Let’s not forget that this is just its third launch. New Glenn’s first launch, NG-1, was partially successful, as it placed a dummy satellite in Medium Earth Orbit (MEO) but failed to land its booster. Its second launch, NG-2, was a complete success, launching two NASA missions to Mars and safely landing the booster (read more here). So, the fact that NG-3 was only a partial success is still a significant step in the right direction, as they have demonstrated their booster can be relaunched.

Let’s also not forget the name of New Glenn’s biggest competitor. New Glenn is a reusable heavy-lift vehicle designed to lower the cost of launching satellite constellations and crewed missions to the lunar surface. It has exactly the same mission brief as Starship; it just goes about its role in a very different way.

Additionally, by Starship’s third test flight, it had failed to land a booster (let alone relaunch it), launch with a payload, launch without significant engine issues and make it to orbit. In terms of progress per launch or progress per development spend, New Glenn is still miles ahead of Starship. In fact, NG-3 proves it has leapfrogged Starship, as even after 11 test flights, Starship has yet to reach orbit or deliver a payload to orbit. In fact, the updated second-generation Starship has explosively failed during three out of five launches. This means that not only is New Glenn demonstrably more capable now, but it is also statistically more reliable, even after the failure of the NC-3 launch.

So, yes, NG-3 was technically a failure, but not to the same extent as the Starship launches.

The Moon

This is really important, because it looks like Blue Origin is on track to take Starship’s Artemis III contract!

Thanks to Starship’s glacially slow development pace, NASA has opened up its contract to launch Artemis III to others, mainly Blue Origin, as they are the only other company with a launch vehicle and a lunar lander in the works.

Blue Origin is making rapid progress developing its MK1 Blue Moon lunar lander, which is designed to carry three tons of cargo to the lunar surface. Blue Origin’s Blue Moon MK1 is set to complete a lunar landing demonstrator mission later this year and will be launched by New Glenn.

But Blue Origin is also making rapid progress on their MK2 Blue Moon lander, which can carry either 20 tons or two astronauts to the lunar surface. This lander does require the upgraded super-heavy variant of New Glenn, the 9x4 (read more here); however, with New Glenn’s fast progress, they seem to be on target to launch this variant by early 2027. This is ideal, as Blue Moon MK2 was selected as the lander for NASA’s 2030 Artemis V lunar mission on the caveat that Blue Origin undertakes an uncrewed demonstrator mission before 2029.

With Starship currently totally unusable, New Glenn 9x4 and Blue Origin MK2 might be NASA’s only viable option. There is a growing consensus that unless something drastic changes, NASA will give these lander contracts to Blue Origin and delay all lunar missions, including the upcoming Artemis III, until they are ready.

So again, while NG-3 was a failure, the fact that Blue Origin is still progressing and is, arguably, mostly on target to complete these deadlines is huge! The main question is whether the FAA investigation will cause significant delays. That entirely depends on the severity of the engine problem, which is a pretty big caveat. However, even with these delays, New Glenn is winning the race against Starship.

SpaceX has essentially set itself up assuming it would receive these multi-billion-dollar contracts, so the fact that there is a real chance someone else could take them might leave Musk in the lurch. This is especially true when you consider that SpaceX’s upcoming IPO will put the company’s financials and business practices under the spotlight.

Is This A Win?

This might ‘feel’ like a win for common sense and decency, given that Musk is potentially taking a huge loss here. But let’s not forget that Jeff Bezos is behind Blue Origin, and he isn’t exactly any less oligarchical than Musk. He is just marginally better at protecting his public image.

Firstly, it’s not like Bezos’s plans for New Glenn make significantly more sense than Musk’s. After all, Blue Origin also hopes to launch thousands of satellite data centres into orbit. If you want to know why that is an utterly moronic idea, read the article here.

However, Bezos’s overall plan for New Glenn is functionally identical to Musk’s for Starship — to make NASA dependent on them, potentially shuttling tens of billions of dollars in taxpayer funds into their coffers, and to use the rocket to build a giant satellite internet constellation, giving them incredible power over how information is accessed and shared across the globe.

I trust I do not need to explain why all of that is bad…

Don’t get me wrong, I am glad someone is challenging SpaceX and its insanity. And I do think in the long term, New Glenn will be a far more successful rocket than Starship. In that sense, New Glenn’s development is a win for real-world engineering and common sense.

But if you value ethics, denounce oligarchic structures of power, or believe resources like the internet or space should be democratically owned and operated, this is far from a win for anyone but Bezos, despite the mishap.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

Back in February, OpenAI began stuffing ChatGPT with hyper-targeted adverts that exploited its users’ deeply personal connection to the slop bot, which was a decision Altman originally considered a last resort. Now, according to Axios, OpenAI has already reached $100 billion in annual recurring revenue from these adverts, and they expect to reach $2.6 billion in ad revenue by the end of the year, as well as $100 billion in annual revenue by 2030! So, the big question is, has this last resort paid off? Has transforming ChatGPT into an ad-based data farming slop machine secured OpenAI’s future? Hell no! I don’t think people realise just how big of a hole Altman has dug for himself. This doesn’t even touch the sides!

Earlier this year, OpenAI CFO Sarah Friar stated that OpenAI received $20 billion in annual recurring revenue (ARR) from paying users in December 2025, which means she claims OpenAI made $1.66 billion in December 2025. There has recently been a mass exodus of users from ChatGPT (with more on that in a bit), so let’s assume that their ARR is approximately the same. That would mean the rollout of ads has increased OpenAI’s revenue by only 0.5%. It has barely moved the needle!

Subscribe now

Now, yes, OpenAI does say it plans to make $2.6 billion in annual revenue from ads by the end of the year. But OpenAI’s projections are almost always exaggerated. I previously estimated that if OpenAI’s revenue continues to grow at its current rate, it will generate approximately $30.8 billion in 2026. So, even in a best-case scenario, these ads might only boost revenue by 8.4%. Note that this is revenue, not profit. In that previous article, I calculated that OpenAI will likely post a loss of at least $22.2 billion for 2026. So again, even in a best-case scenario, these ads will only reduce OpenAI’s losses by just over 10%.

This isn’t ideal in the present moment, but the long-term perspective is even worse.

Let’s be generous and accept OpenAI’s predicted future ad revenue. How much would that projected $100 billion help their overall revenue?

Well, previously, OpenAI predicted it would make $125 billion in revenue in 2029. If we use this new prediction of $100 billion in annual ad revenue by 2030, we can assume that OpenAI is now expecting to make $225 billion in revenue in 2029.

That sounds impressive — but it isn’t enough.

OpenAI’s own predictions reveal the company is on course to post annual losses of $115 billion in 2029, using that $125 billion projected revenue. So even with this projected ad revenue boosting those earnings, the company would still post annual losses of $15 billion! I know there are a lot of zeros here, but that level of loss is not conducive to a sustainable business at all.

Not to mention that is a highly optimistic best-case scenario.

OpenAI has reportedly reduced its spending targets, with investors recently told it aims to spend around $600 billion on total compute by 2030. For some perspective, Ed Zitron found that OpenAI spent around $13 billion on compute with Microsoft in 2025. OpenAI’s internal documents expect a $14 billion loss in 2026 from a $20 billion revenue, suggesting a $34 billion compute spend this year. Let’s then assume a fairly linear increase in compute spending of $100 billion in 2027, $150 billion in 2028, and $303 billion in 2029, for a total of $600 billion in compute spend between now and 2030, just as OpenAI is telling investors.

That would mean that in 2029, OpenAI would have $125 billion in revenue from paying users and $100 billion in revenue from adverts, generating an annual compute spend of $303 billion. That would make for an annual loss of $78 billion!

Even this scenario is based on OpenAI’s ludicrously optimistic revenue figures.

Quite simply, this ad revenue is nowhere near enough to even substantially delay OpenAI’s inevitable collapse.

So, was it worth it?

Well, OpenAI has already lost a major researcher. In a New York Times article, Zoë Hitzig detailed why she is leaving Altman’s company, stating, “I have deep reservations about OpenAI’s [advert] strategy.” She went on to explain that OpenAI plans to exploit its user base with hyper-targeted adverts, leveraging the highly sensitive data its chatbots have gathered on them, which leaves users vulnerable to a new level of manipulation.

OpenAI’s userbase has echoed Hitzig’s sentiment. It has been reported that 1.5 million users left OpenAI over their decision to take the Pentagon contract that Anthropic refused. This mass exodus was almost certainly spurred on by the rollout of adverts to paying ChatGPT users, and thanks to the similar timing of these two events (which, yes, makes Altman and OpenAI look insanely desperate), we will never know how much of this exodus to assign to either cause.

In short, these adverts are likely more of a hindrance to OpenAI than you probably realise, making the question “Was it worth it?” hard to answer.

In a previous article, I said, “OpenAI deploying ads is the equivalent of the captain of the Hindenburg waving a dummy fire extinguisher out of the cockpit window while flames begin to lick up the side of the hydrogen-filled balloon. It isn’t a solution at all, but it kind of looks like one, and it might keep the delusion that everything is okay going for just a little longer.” And it seems I was right! This mess only proves how desperate Altman is to keep the charade up for just a bit longer.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

Just over a week ago, Anthropic announced its Claude Mythos model to the world, and boy, did everyone lose their tiny little marbles over it. It’s easy to see why. Mythos is effectively a cybersecurity bot, able to crawl through code, find vulnerabilities, and exploit them. In just a few weeks, Anthropic claim Mythos discovered thousands of zero-day vulnerabilities in every operating system and web browser. Naturally, Anthropic is so worried about the ramifications of such a model being publicly available and other companies creating similarly dangerous models that they are currently only launching it as a ‘preview’ for researchers, government bodies, and major software developers so they can get ahead of the tidal wave of cyberattacks these models will cause. But it’s fair to say that, like all AI bros, Anthropic has a solid track record of being more marketing spin than trousers. So, is this the AI armada we were warned about? Or is this just a marketing stunt?

Why We Should Be Sceptical

It’s fair to say that the timing of Mythos’s launch is so coincidental that it raises some major suspicions.

A few weeks ago, Anthropic’s main product, Claude Code, was being publicly dragged. AMD’s director trashed the coding assistant for getting dumber and lazier. It simply couldn’t do tasks it once excelled at. This could be explained by a recent OpenAI memo, which details how Anthropic is compute-limited, or the recent report that Claude is experiencing regular, damaging disruptions. Quite simply, Anthropic might be dumbing down Claude to cope with the enormous influx of former OpenAI users. Then, because Anthropic has vibe-coded much of Claude’s scaffolding, its source code was leaked, allowing the entire world to see what made it tick! Ironically, for a company that can only exist by stealing more copyrighted works than nearly anyone else, except OpenAI, Anthropic claimed the code was their IP and issued approximately 8,000 copyright strikes to get the leaked code taken down. Not exactly a good look for Anthropic from an ethical, cybersecurity, or competency angle — especially when you consider they are planning an IPO by the end of the year.

Subscribe now

So, the fact that they have suddenly developed an unimaginably powerful AI that is too dangerous for us to even try, let alone verify, is a little coincidental, don’t you think? It’s almost like this is a perfect marketing tool to bury all the major issues and bad press thrown at Anthropic over the past month. And, while this is in no way proof that Mythos is exclusively a swizz, it at the very least is a sign we need a dose of healthy scepticism here.

Remember: it isn’t just cranks on the internet like me who are urging scepticism. Some of the most influential AI researchers, industry experts, and even Big Tech investors have serious doubts about Mythos.

AI and neuroscientist Gary Marcus stated that Mythos’s risk was “overblown” and that “To a certain degree, I feel that we were played.” Yann LeCun, one of the ‘AI Godfathers’, totally dismissed Mythos, calling it “BS from self-delusion”, and said that other smaller models could already finish the same tasks. Dr Heidy Khlaaf, Chief AI Scientist at the AI Now Institute, declared that the announcement of Mythos used vague language and didn’t include the right metrics to verify the claims Anthropic made, leading her to question whether the announcement was really designed to generate investment without too much scrutiny. David Sacks, arguably one of the most egregious members of the PayPal mafia — who is, in my opinion, an utter wanker and, more importantly for our conversation, a Big Tech investor — maintained that Anthropic has a history of “scare tactics” to drive hype, meaning we should take their claims with a grain of salt. When one of these guys warns that tech claims are too overblown, you know it’s bad!

That is a chorus of deeply informed people across the political and ethical spectrum, all warning that things might not be as they seem with Mythos.

Not to mention that independent analysis has supported this perspective. AISI is one of the organisations that was allowed to preview Mythos and test its capabilities. Though they did consider it a step forward in terms of its ability to perform simulated cybersecurity attacks, they also found it wasn’t as significant a leap forward as Anthropic claimed.

But there is one glaring anecdotal hole in Anthropic’s narrative.

If Mythos is so outstanding, why didn’t they use it to stop the Claude leak? You’d think that the first thing you would do if your company were about to become the de facto experts in software vulnerability detection and a tidal wave of AI-powered cyberattacks were swiftly approaching would be to use said AI cybersecurity tool to make sure all your software was fully patched. It doesn’t quite make sense that Anthropic can have a model that is supposedly as powerful as Mythos and experience a giant breach at the same time. There are a few possible explanations here: Mythos might not be able to accomplish what they claim it can; Claude’s heavily vibe-coded nature could make it nearly impossible to solve bugs and vulnerabilities; or even that Anthropic doesn’t trust Mythos not to completely ruin Claude. But no matter what the truth is here, this discrepancy doesn’t look good.

Okay, so we need to be sceptical. We get it. What questions should we be asking?

The Questions I Am Asking

Because I am not an AI scientist — just a deeply burnt-out autistic man with an internet connection — I can’t give you the ‘right’ answer. What I can do is tell you the questions I am asking to try and get to the bottom of this, which you can potentially use as a starting point.

My first question is, “How ‘good’ are the vulnerabilities?”

Just because Mythos claims it has found thousands of zero-day vulnerabilities doesn’t mean they’re actually hackable or exploitable cybersecurity risks. Indeed, Daniel Stenberg, founder and lead developer of cURL, told The Register that similar tools often incorrectly flag unexploitable bugs as vulnerabilities, adding serious slop-pressure for teams. Anthropic does claim that Mythos Preview can find working exploits 72.4% of the time, but that could just be hype. After all, the definition of ‘exploitable’ is a moving target. We need to see that Mythos isn’t just throwing spaghetti at the wall and can actually identify real-world risks in code. When the organisations with access to Mythos Preview move past lab-bench marking (which AI can famously be tweaked to do well in) and into real-world testing, we should hopefully receive our answer.

The next question is “How narrow is the model?”

For a variety of reasons — such as the Floridi Conjecture (read more here) and the Efficient Compute Frontier (read more here) — it is very difficult to make general-purpose models much better than they are today. As such, Anthropic has almost certainly had to narrow Mythos’ scope to improve it, transforming it from a general-purpose chatbot like Claude or ChatGPT into a specialised cybersecurity tool. The fact that Anthropic is currently compute-strapped is even more evidence that Mythos is likely very narrow. So, how much did Anthropic have to narrow down the Mythos model to make it efficient at this task? This question matters. Firstly, for obvious reasons, a super-narrow model isn’t as commercially viable. It can only complete one task, after all. But if Anthropic did have to significantly narrow down Mythos, it means they consciously created a tool that could unlock cyberattacks on a scale never seen before. Again, that undermines Anthropic’s ‘good guy’ ethical marketing and is a damn good example of why this cowboy industry needs to be regulated. If any other industry began purposefully making machines that could potentially collapse the digital infrastructure upon which our entire society depends, it would be regulated into the ground!

The next question is one some of you may have already been asking: “How expensive is it to run?”

If Mythos requires ungodly amounts of computing power and energy to run properly, then it can’t be the wide-scale risk Anthropic claims it is, can’t be a viable commercial tool, and is arguably a step backwards for the AI industry. Now, we do know that Mythos is very expensive to run, with Anthropic’s own documentation stating that it is “very expensive for us to serve, and will be very expensive for our customers to use.” But just how expensive? Anthropic doesn’t have deep enough pockets to subsidise a tool like this, and because investors are growing increasingly worried about the AI bubble, that is likely to remain the case for a while. Furthermore, if the model is heinously expensive to run, it might be significantly less efficient at finding bugs and vulnerabilities, measured in dollars per vulnerability found, making it commercially useless. Also, such a cost would keep the Mythos service small and out of the hands of more nefarious users.

This is a major issue plaguing all current AI systems. To make them even slightly better than they currently are requires exponentially more computing power and, therefore, exponentially more cost, which prices these tools beyond usability.

The next question is more of a pragmatic one: “How will Anthropic find solutions to the bugs and vulnerabilities Mythos finds?”

You see, it’s one thing to find a vulnerability — it’s another thing to fix it. Indeed, Daniel Stenberg from earlier told The Register that tools like Mythos don’t offer solutions when they report vulnerabilities. You might suggest that Anthropic could pair Mythos with Claude Code to find and vibe-fix these issues, but sadly, that isn’t likely to work. AI coding tools are great at generating small code snippets, but they are fucking awful at understanding entire programs and even worse at independently coding tasks. In a previous article, I discussed the University of Waterloo’s research, which found that even the best generative AI coders only have a 75% accuracy rate when tasked with very basic coding tasks. In other words, even basic AI-generated code doesn’t work a quarter of the time! I also analysed research from Veracode, which found that 45% of AI-generated code contained security flaws, and a study from Coderabbit, which found that AI-generated code has 70% more bugs than human-written code. So, no, Claude Code probably can’t join forces with Mythos to create an automated turn-key solution to cybersecurity.

This is a bit of a problem. It essentially means Anthropic has just shat in the pool and made it everyone else’s problem. If their claims about Mythos are even half-true, then the fact that AI coding can’t solve the issues Mythos finds means that the entire software industry, which is already strained past breaking point, will be lumbered with exponentially more work as they are obligated to race to fix thousands of vulnerabilities.

Again, this isn’t exactly the ethical thing to do, is it, Anthropic? Maybe industry regulation would have been the preferred solution here. But no, you guys desperately want to be the ‘ethical’ oligarch…

This leads me to the final question: “What is the long game here?”

These techbros all believe they are playing 4D chess, when in reality, they are playing tiddlywinks. It’s not that hard to see through their master plans. And that is a feature, not a bug, as it often enables them to dog whistle to potential investors. So, what is Anthropic’s end goal with Mythos? What is the narrative they are trying to subtly spin?

Well, I think it could be a terrible amalgamation of a protection racket and a Trojan horse. Let me explain.

At a time when the industry as a whole, especially the critical open-source sector, is already strained past breaking point, Mythos will apparently flood developers with reports, putting them under immense pressure. So, how are these teams going to cope? Well, they are going to be expected to turn to AI coding tools like Claude Code (particularly as it will likely work better with Mythos than other tools) to speed up patches. But, as we know, these tools produce a ton of bugs, so this isn’t a solution at all. Therefore, the Mythos reports will keep flying in, and Claude Code will rewrite a huge amount of code. Eventually, after this cycle has repeated for a while, most code will be written by AI, making it almost impossible for human coders to rewrite. That will effectively force software makers to use generative AI coding tools to maintain their services.

I think this is the narrative Anthropic is trying to push — that developers will likely pay an exorbitant price for Mythos to improve their cybersecurity, but the tsunami of reports will pressure them to adopt generative AI coding tools, effectively forcing the software industry to become deeply dependent on their AI coding tools.

Now, to be clear, I do not think this is what will actually happen. As I have said before, neither Mythos nor Claude Code is good enough to pull this off. But this is the AI bubble; reality doesn’t matter. What does matter is that investors believe you are going to become a dominating monopoly. That is why the name Mythos is so fitting. It appears to sell a myth of AI dominance, right before Anthropic goes public.

This is just a hypothetical. I don’t know if these are Anthropic’s intentions. I’m just trying to read between the lines.

Summary

Until Anthropic unleashes Mythos into the wild, we likely won’t get solid answers to any of these questions or know if Mythos is just a giant PR stunt. We are just going to have to sit with the knowledge that we are unaware of how much of this is marketing BS or genuine risk. All I know is that, either way, we should keep a healthy amount of scepticism about Anthropic.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

An upcoming $25,000 EV, dubbed the ‘Model 2’, was once one of the main justifications for Tesla’s ludicrous speculative value. Musk announced this model in 2020, with it slated to launch in 2023. Then, 2023 rolled around, and rather than launching the Model 2, Musk announced Tesla had one in the works, which he expected to launch in 2025 and predicted would sell well over a million units a year. Investors were practically drooling at the idea of Tesla gaining such a monopolistic chunk of the car market, which catapulted Tesla’s value. But less than a year later, in 2024, Musk scrapped the Model 2 and redirected all development funds toward Tesla’s limp robotaxi and robot projects. Now, two years later, reports are circulating that Tesla is trying to bring the dead-and-buried Model 2 back to life. You could rightly see this decision as yet another example of inept flip-flopping from an out-of-touch billionaire CEO with far too much unchecked power. But in light of Tesla’s rapidly downward-spiralling financial health, the abject failure of Tesla’s Robotaxis, and institutional investors’ sudden realisation that reality exists, this revived Model 2 looks more like a futile, impotent, and desperate attempt to avoid the imminent death blow coming for Tesla. Let me explain.

The Leak

How do we know that Tesla is trying to revive the Model 2? Well, outlets like Reuters have reported that anonymous sources working with companies in Tesla’s supply chain claim Tesla is developing a new small EV that isn’t based on the Model 3 or Y. The scant details that we know so far about this car is that it will be smaller than a Model Y at 4.3 m long, it will be built in China, be a “wide margin” less expensive than the Model 3, and that Tesla has not yet greenlit this new EV for production. That’s about it.

Subscribe now

Compare that to the ‘old’ Model 2, which was slated to be less than 4 m long, have a base price of $25,000, and be produced in Mexico. That means this new, smaller EV might not even pick up where the old Model 2 project left off.

This is also a remarkable U-turn for Musk. Back in 2024, against the advice of almost every executive (read more here), he cancelled the ‘old’ Model 2, calling affordable EVs “silly” and “pointless”, and claimed that autonomous robotaxis would soon replace vehicle ownership for the majority of people on the affordable end of the market.

So, why backtrack? And why now?

Why This Makes Sense

In simple terms, Tesla isn’t exactly doing well right now, given that every plan Musk has proposed is failing, and the Model 2 might be the only thing that can help soften the blow of these failures.

Tesla’s global sales dropped by 8% in 2025 compared to 2024 to 1.64 million vehicles, and recent sales figures suggest that trend is only intensifying. On top of that, one of Tesla’s main sources of income, emissions credits, is beginning to totally disappear (which we will cover in more detail in a minute). All of this might not sound too bad, but as I covered in a previous article, this has caused Tesla’s net profit to drop 46% compared to 2024 to just $3.79 billion, and its 2026 profit is set to shrink even further.

It wasn’t supposed to be like this. Tesla was supposed to deliver far more vehicles, rake in billions of dollars from Robotaxis and sell its Optimus robot like hotcakes by now.

Back in 2022, analysts projected that Tesla would sell 1.366 million vehicles in Q1 of 2026, driven by the Cybertruck and the mythical Model 2. Tesla fell short of this prediction by 71%, missing it by more than one million units!

Why did they miss? Well, since its November 2023 launch, Tesla has only cumulatively sold around 50,000 units of the god-awful Cybertruck That is just 20% of the 250,000-unit-per-year production capacity Tesla was hoping to reach by now, and only 5% of the million reservations that the company reportedly had for the vehicle. Musk then cancelled the $25,000 Model 2 and, upon realising this was a mistake, tried to correct the problem by releasing cheaper, stripped-out versions of the Model 3 and Y. However, because of their $37,000 and $40,000 respective prices, they totally failed to boost sales.

But, remember, Musk cancelled the Model 2 because robotaxis were going to make affordable EVs obsolete. Indeed, Musk blatantly put Tesla’s money where his mouth was, given that Tesla has likely sunk somewhere between $10 billion and $20 billion into developing FSD. Such an investment had to pay off. Cathy Wood of Ark Invest predicted that by 2026, Robotaxis would account for more than half of Tesla’s EBITDA (Earnings Before Interest, Taxes, Depreciation, and Amortisation) and that by 2027, Robotaxis would bring $200 billion to $600 billion in revenue for Tesla. Those figures imply Tesla was set to take a gigantic bite out of the passenger transport industry.

Yet, here we are in 2026, and Tesla’s Robotaxi service is limited to a handful of human-supervised Model Ys bumbling around Austin, Texas. Not only that, but these Robotaxis with a human supervisor in the car currently have a crash rate four times higher than that of an average human driver (read more here). As such, Tesla is currently pulling in functionally zero Robotaxi revenue, and thanks to this abominable crash rate, the service is unlikely to pass this pilot program stage for an awfully long time. Additionally, the FSD (Full Self-Driving) system these Robotaxis are based on isn’t selling well either. As I covered in a previous article, by the end of 2025, only 12% of the Tesla fleet had bought FSD, and FSD revenue in Q3 2025 was actually lower than in Q3 2024.

So, no, autonomous driving of any kind isn’t adding anything meaningful to Tesla’s bottom line.

But maybe Tesla’s other AI program, its Optimus humanoid robot, is doing better?

In 2022, Musk announced he hoped to have Optimus production ready by 2023 and start selling thousands of them by 2025. Musk then revised this timeline, stating that Tesla would have “genuinely useful” humanoid robots in low production for internal purposes (as in factory work) by 2025. But here we are a year later, and despite trying to claim otherwise, Optimus isn’t being used in Tesla’s own factories, is basically useless, and its development is glacially slow.

In other words, Optimus isn’t doing any better than Tesla’s awful Robotaxi program.

With all of these massive bets obviously failing and Tesla’s core business slowly dying away, launching a more conventional, competitive, affordable EV is one of the few options Tesla has left to try and resolve this mess. So, it makes sense for them to pull an old idea off the corkboard.

Why It Doesn’t Make Sense

Or does it?

Ultimately, Tesla isn’t the first company to move into the affordable EV space. In fact, they are way behind the curve. Renault, MG, BYD, Peugeot, Jeep, Citroen, Nissan, Vauxhall, Leapmotor and even Chevrolet have cars with similar prices and specs as the original promised Model 2. Arguably, the $20,000 to $30,000 section of the EV market is by far the most competitive. In this regard, other brands have already eaten Tesla’s lunch. Take the Renault 5; its price and specs were nearly identical to the promised Model 2’s, and it is now one of the best-selling cars in Europe. With competition like this, combined with Tesla’s current reputation, Tesla will struggle to sell the Model 2 at the scale it needs.

But it is also far too late. EVs take years to develop. Again, take the Renault 5, whose development started in 2020 and only made it to market by late 2024. Tesla isn’t exactly known for its speedy development either. For example, the Cybertruck was delayed for years. So, if Tesla has just begun development of a completely new Model 2, it might not reach the market until 2030 or later. That gives the current affordable end of the EV market time to mature, making it substantially harder for Tesla to compete. But that timeframe is also too late to resolve the mounting issues stacking up against Tesla. If Musk wanted to have a chance of competing in this market and attracting any serious revenue, Tesla should have started development of a Model 2 in 2019 and be preparing a second generation by now.

Tesla’s loss of a technological advantage exacerbates this market problem. Their 4680 battery was touted as the future, but it is less energy-dense and slower to charge than its predecessor; has a horrifically limited production capacity; and can now be replaced by significantly cheaper and better batteries on the market. Proprietary EV tech is also miles ahead of Tesla — for example, BYD’s newest generation of batteries can deliver a 10% to 70% charge in five mins in a £30,940 EV (read more here). In other words, Tesla’s tech no longer makes their cars the cheapest, most efficient, or the fastest charging, so to be competitive, their Model 2 will need to use third-party components, which means they have zero tech advantage over the market. Remember, one of the only reasons Tesla has grown so much is because it had a tech advantage over everyone else.

Then there is the question of how the hell Tesla is going to pay for Model 2 development, given that they are facing a serious cash squeeze! We have already discussed how Tesla’s profits have shrunk by 46%, but with emission credits disappearing rapidly and sales continuing to dwindle, they are predicted to fall more than 36% this year! Also, Bloomberg has just reported that Tesla’s free cash flow is expected to drop by $12 billion, going from $6.2 billion at the end of 2025 to -$5.8 billion! So, where is Tesla going to find the money to pay for their insane AI projects, and source a few billion dollars to fund development of a new Model 2?

Still, that isn’t even the real issue! Musk cancelled the old Model 2 because he believed robotaxis would make it obsolete, and because he heavily gambled on the idea that AI would become autonomous. This huge bet on autonomy, and the speculation that surrounds it, is what is currently keeping Tesla’s insane valuation so damn high. But, uncancelling the Model 2 implies that Tesla knows this bet is failing and that it has squandered its EV lead for nothing. It also proves that Tesla isn’t fully committed to its pivot to autonomy. Not only does this mean Tesla has to divide its R&D budget, reducing how much it can spend on AI, but it also reduces how much speculation can be pushed about its AI future. So, how does Musk launch a Model 2 without destroying the speculation that Tesla relies on?

Basically, no matter how you cut this cake, the Model 2 isn’t really a solution for Tesla’s problems.

In reality, Tesla is stuck between a rock and a hard place. The gargantuan and ill-advised bets Musk has made for Tesla are not paying off. Yet, Musk squandered Tesla’s lead to make those bets. The foundations of Tesla are falling apart, and it is too late to fix them. So, what will happen to Tesla’s towering value now that the pillars that prop up the company have been removed?

The Realisation

Slowly, institutional analysts and investors are starting to realise that reality is about to bite Tesla in the arse.

Take the recent report from Ryan Brinkman at JP Morgan. He discovered that Tesla is currently grossly overvalued at $361 per share. He pointed to Tesla’s dire financial situation and underwhelming sales, and calculated that Tesla is actually worth just $145 per share. According to him, Musk’s cult of personality could explain the discrepancy, but even that value is being slowly eroded as Musk’s promises continue to fall flat. As such, Brinkman thinks Tesla’s valuation is due to experience an imminent 60% collapse.

However, even that figure might be a little too rosy, as at that value, Tesla’s P/E ratio (which is the value of a company compared to its profit) is still substantially higher than that of any other automaker. That is why I previously calculated that if Tesla loses its speculative value, it could fall by some 94% (read more here).

Summary

This is why the Model 2 matters. Even institutional analysts and investors, who were previously happy to ride the speculative wave and make millions from it, are now worried that reality will hit Tesla and incite an imminent collapse. The fact that Tesla is potentially making a desperate attempt to wind back the clock and deliver the Model 2 shows that Musk and Tesla are aware of this, too. They know the Model 2 was their only real option to boost sales, and they know it is one of the only ways to quell this investor anxiety. But this is an impotent attempt. A new Model 2 won’t garner the results Tesla needs to justify itself, and Musk surely knows that. This desperate effort is not designed to solve the problem at hand, but to keep the facade alive for a little longer.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

We all know that Musk can’t get enough of automation. From Optimus to self-driving cars, Musk wants to let his robots take over. But, as it turns out, Musk has been doing a similar thing with his role as Tesla CEO, except rather than replacing himself with AI, he has pulled more of a “Jesus take the wheel” move. You see, the Washington Post recently published an interview with one of Musk’s old Tesla executives, who let slip that Musk wanted to spend as little time working with Tesla as possible. That might sound innocuous, but to anyone who understands business management and business culture, that is horrific. It means Musk is effectively asleep at the wheel and allowing Tesla to drive itself off a cliff. But this is not an accident. When you look at this in the context of what Tesla has become, it makes complete sense and, in my opinion, proves that Musk is doing this on purpose.

The Interview

Let’s start with this Washington Post interview with Jon McNeill, who was Tesla’s president of global sales, delivery and service for three years. In the interview, McNeill stated that “when I asked [Musk] what success looked like, he said, ‘Success is getting me down to one day a week at Tesla so I can get back to my first love, which is rockets.’” This wasn’t a one-time claim. McNeill alleged this was Musk’s main motive throughout his time at Tesla, from 2015 to 2018. That is in stark contrast to the hilariously hollow workaholic mythos Musk has built up around himself in recent years.

His attitude likely hasn’t changed either. Recently, angry shareholders demanded that Musk spend more time at Tesla, and Musk has threatened to walk from Tesla multiple times. So, it is safe to assume that this one-day-a-week goal is still very much relevant.

This revelation alone destroys much of the cult-like mythos Musk has created. But it has far deeper implications when you take the reality of Tesla into context.

Tesla’s Micro Reality

Tesla is on a historic downward spiral and shows no signs of letting up.

Subscribe now

In a previous article, I covered how Tesla’s annual profit fell 46% in 2025 compared to 2024, with its Q4 2025 profits down an astronomical 61% compared to Q4 2024. In fact, regulatory emissions credits accounted for $542 million of Tesla’s $840 million profit in Q4 2025; however, this revenue stream has now functionally collapsed for Tesla (read more here). The only “good” news was that between 2024 and 2025, Tesla’s energy storage revenue increased by 26.5%. At the time, I predicted that this growth would end soon, given that far better products were entering the market. And it turns out I was right! Recent reporting found that in Q1 2026, Tesla’s energy storage installations totalled just 8.8 GWh, an astonishing 39% short of its 14.4 GWh goal, meaning installations actually shrank 15% year-over-year. On top of that, it found that Tesla failed to meet its already low vehicle deliveries projection by 4%, heavily suggesting that the catastrophic sales slump has continued.

So yeah, Tesla isn’t doing so well! Its core business, where it generates the vast majority of its revenue and profit, is collapsing. Why? You could point to the understandable backlash to Musk’s disastrous political actions. But there is a reasonable argument that, in fact, this has been caused by Musk taking Tesla in a totally different direction and him functionally betting the entire company on this hard pivot.

Tesla’s Macro Reality

This is the broader picture of Tesla. Musk is turning Tesla from an EV pioneer into a “leader” in humanoid robots and robotaxis. I have covered this topic before, but at Musk’s behest, Tesla has effectively dropped all its EV development projects to focus entirely on the self-driving Cybercab and the robot Optimus. Because of this, Tesla’s ailing EV business cannot be recovered, and the robot and robotaxi ventures are the only viable route to sustainability or growth. In effect, Musk has bet the entire future of Tesla on these projects paying off.

Even if the underlying technology and concepts are solid, this pivot is a gargantuan, highly risky bet that explains much of the granular downward spiral we are seeing with Tesla. But here is the thing: Tesla executives and insiders have known that the concepts and technology underlying this shift are non-starters.

Optimus

Take Chris Walti, the original lead on the Optimus robot project. In a previous article, I covered how Walti has publicly stated that humanoid robots simply don’t make sense and that the humanoid form factor isn’t “a useful form factor.” Walti shares the same view as other robotics experts, such as Brad Porter (former VP of Robotics at Amazon), Gartneranalysis and Ken Goldberg (UC Berkeley roboticist), that the human form factor is highly inefficient. For example, why ask an expensive, slow and difficult-to-set-up general-purpose humanoid robot to use a vacuum when you could use a cheaper and more reliable specialised robot like a Roomba to complete the same task? As such, Walti likely saw Tesla’s Optimus as more of a tech demonstrator than an actual useful product.

Walti would have almost certainly fed his opinion back to Musk. He would have also likely fed back the fact that Optimus is far, far behind the competition from China and Boston Dynamics. Yet Musk has totally ignored Walti and proceeded to make selling millions of Optimus robots integral to Tesla’s future. Indeed, Musk’s ridiculous $1 trillion pay packet includes a condition that Musk must sell a million robots.

Robotaxi

The same is true of Tesla’s Robotaxi. In a previous article, I covered how Rohan Patel, Tesla’s former head of business development and policy, warned Musk that the Robotaxi wouldn’t make any profit and that Tesla should instead focus on more affordable EVs.

We know this because Patel’s internal analysis of Musk’s Robotaxi plans leaked. This analysis found that even if Tesla could get its FSD-powered Robotaxis to work, they wouldn’t make any profit for years to come. Patel concluded that the venture was a dead end and advised Musk against investing heavily in it. Instead, he recommended focusing on developing cheaper, mass-market EVs, such as the proposed Model 2, and keeping Tesla at the forefront of the EV niche.

But again, Musk ignored Patel, scrapped the Model 2, and shifted much of Tesla’s R&D to the Cybercab and Robotaxi projects. Unfortunately, Tesla’s Robotaxi service is so small it is laughable and is certainly not a profit-maker for Tesla, which it won’t be for an awfully long time (read more here). Meanwhile, the Renault 5, which offers basically the same specs and price as the promised Model 2, is now one of the best-selling EVs, proving that Patel was right (read more here).

But remember that Patel’s analysis assumed that the technology powering the Robotaxi, Tesla’s FSD (Full Self-Driving), actually works. And here is the really inconvenient thing: it doesn’t, because, again, Musk ignored and overruled his own experts.

FSD

In yet another previous article, I covered how Musk overruled Tesla engineers and, in the process, kneecapped FSD.

Ultimately, Musk believes that because we humans navigate with vision alone, an AI can do the same. This has caused him to remove all the sensors from Teslas and make them “vision-only” systems. Therefore, the self-driving AI only has a few camera feeds to understand the world around it.

Tesla engineers rightly warned Musk that this is a horrific idea because it leaves the system with no redundancy. When conditions inevitably render the camera feeds useless, the AI can’t see or interpret its environment correctly. Likewise, AI computer vision can’t be 100% accurate, so when it inevitably reads the road wrong, there is nothing in place to catch that error. This method effectively puts all your eggs in an AI basket that has already been proven unreliable. Meanwhile, the industry-standard approach of having an entire suite of different sensors, like lidar, radar and ultrasonic sensors, allows separate systems to check and verify against each other, enabling potentially fatal errors to be caught.

Tesla engineers cautioned Musk that switching to a visual-only system would be problematic and unsafe. But again, Musk ignored them and forced the entire Tesla fleet to switch to vision-only.

The engineers were so concerned about the safety implications of this decision that they whistleblew to the Washington Post. The report interviewed nearly a dozen former employees, test drivers, safety officials, and other experts, who all reported an increase in crashes, near-misses, and other embarrassing mistakes by Tesla vehicles once they switched to vision-only. The report even found that Musk rushed the release of FSD (Full Self-Driving) before it was ready and launched it despite the software not being safe for public road use. In fact, a former test operator went on record saying that the company is “nowhere close” to having a finished product.

Even now, five years later, Tesla’s FSD is still far from being safely autonomous. According to FSD Tracker, which uses customer-reported journeys using FSD to calculate the rate of critical disengagement and disengagement rates per mile, the very latest version of FSD has a critical disengagement rate (where the driver must intervene to stop a crash) of once every 2,035 miles on average. For some context, Waymo’s average distance between critical disengagements is more like 17,000 miles. Even worse, when operating on city streets, as a robotaxi does, FSD’s average distance between critical disengagements is just 867 miles. That means that a fully autonomous Tesla Robotaxi, with no human oversight, would crash roughly once a week if let loose on the street today.

Again, Musk ignored critical feedback, and now Tesla is in a significantly worse position because of his ignorance.

How?

When I covered these previous stories, I believed Musk was just an egotistical idiot. Truth be told, he still is. But the revelation of this new Washington Post interview enabled me to answer the “how” and “why” behind these horrific missteps.

If Musk only spends a single day a week at Tesla, that doesn’t give his executives enough time to provide him with proper feedback. It only gives time for Musk to come in, bleat some orders, and then disappear. I cannot stress how important proper feedback is at this level of business. Not only is it critical in making sure different parts of the company function properly together, but it is also essential to creating a culture of responsibility, which is necessary to ensure detrimental actors and decisions are identified and rooted out. Yet Musk decided to reduce his time at Tesla so much that this feedback quite simply isn’t possible, which will have fostered a deep culture of no responsibility.

So, McNeill’s interview shows us the mechanism behind how all these monumental f**k-ups have occurred.

But why has Musk done this?

Why?

Musk doesn’t get paid based on Tesla’s business performance. Even his giant pay packets aren’t contingent on meaningful metrics, like profit. No, Musk gets paid based on Tesla’s stock price. Musk has made billions of dollars through selling Tesla shares and using Tesla shares as collateral for loans. So, if Musk wants to get paid, he doesn’t need to improve Tesla’s business performance — he needs to pump up the stock price.

Once upon a time, those two things were one and the same. But not anymore.

Tesla is one of the most overvalued stocks in the world. Its price has more to do with speculation than material reality. Indeed, you could argue that his cult of personality influences the company’s stock price more than any realistic future earnings. This hyper-speculation has driven Tesla’s share price so far from reality that the two are now functionally separate.

In other words, if Musk wants a big payday, all he needs to do is make some wild claims, pump the hype, and watch the stock tick steadily upwards. At this point, improving Tesla’s actual business doesn’t even move the stock needle.

Inconveniently for him, wild claims like this get dragged down by feedback loops and executives holding each other accountable. Take the Robotaxi. If Musk actually followed Patel’s feedback and took responsibility for how unrealistic his plans were, then Musk couldn’t pump speculation that Tesla’s Robotaxis would take over the world and so couldn’t increase the speculative value of Tesla. Musk needs to break feedback loops and systems of responsibility to push this speculative hype.

It is my opinion that Musk’s lack of time at Tesla was purposefully designed to do exactly this. Musk doesn’t care about Tesla’s reality; it doesn’t affect him if the business itself is in a downward spiral. All he cares about is the speculative value. So, by spending as little time as possible at Tesla, he can still dictate the wider direction Tesla is heading, allowing him to pump speculation while removing any form of feedback or responsibility, which obfuscates the reality of the situation.

Shouldn’t He Be Ousted?

Any other CEO would be ousted for this behaviour. Gutting a company like this is the job of private equity firms, not the CEO, after all. So, why hasn’t Musk been kicked out for decimating Tesla’s core business?

There are two main ways to oust a CEO: being voted out by the board or being voted out by the shareholders.

Musk was once chair of Tesla’s board but has since been forced out by the SEC. However, that isn’t much of a problem, as Musk used his significant influence to jam Tesla’s board, and he pays them more than any other tech company to keep them loyal. They aren’t going to vote him out any time soon.

The same is true of Tesla shareholders. Without Musk’s speculation, Tesla is worth significantly less than its current value. I have previously estimated that if Tesla were valued in line with other tech-forward automakers, it would lose around 90% of its value. Even Tesla bulls are willing to admit that much of what they invested in is Musk and not Tesla. In a previous article, I covered how Tesla bull Gary Black publicly stated that if Musk stepped down, Tesla’s stock could fall by 20% to 25%, wiping out nearly $220 billion in shareholder value. He wasn’t alone in this statement, with hordes of other major investors echoing his sentiment. On top of all of that, Musk owns roughly 20% of Tesla shares! So, to get a majority vote to oust Musk, you would need to get more than 62.5% of the other shares to vote in a way that will utterly tank the company’s value.

In a way, Musk has effectively blackmailed the only people who can hold him accountable. All so he can gut Tesla, exclusively prioritise its speculative value, ignore critical feedback, and avoid even the slightest hint of accountability.

Summary

This is why the implications of McNeill’s interview go much deeper than most people are making out. It heavily implies that Musk is purposely piloting the company blindfolded. He is asleep at the wheel, pandering to an echo chamber of a stock market that laps up his sci-fi bullshit and pushes his grandiose delusions to dangerous heights. There is just one problem with this plan. Speculation doesn’t last forever. Eventually, reality will come back to bite him, and when it does, Musk and Tesla will feel its teeth. The question is when reality will take a mouthful of Musk and Tesla — not if.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTube channel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

Did you know Starship was meant to land on the Moon two years ago? Yet here we are in 2026, and Starship has not even reached orbit. This rocket is so far behind, it’s beyond a joke. With SpaceX’s IPO rapidly approaching and NASA’s crewed Artemis III mission (which requires Starship) just on the horizon, SpaceX desperately needs to take a monumental leap forward. Well, that is what Starship V3 (version 3) is expected to deliver, and Musk recently announced that the first fully V3 Starship will launch in May. So, the big question is: what does V3 need to accomplish, and by when, to get this car crash of a program back on track?

The Goal Posts

Let’s start with what the actual goal is and how Starship needs to operate to achieve it. Considering how much cash NASA has poured into Starship, Artemis III is arguably Starship’s biggest objective. This mission is supposed to use the Human Landing System variant of Starship to shuttle crew between lunar orbit and the lunar surface. The mission was originally slated to happen last year, but because Starship is so far behind, it has been significantly delayed to late 2028. Indeed, NASA’s Watchdog is now concerned that Starship development is progressing so slowly that this mission will be even further delayed.

So, what does Starship have to do to fulfil this mission?

Well, the crew of Artemis III will take off on board a NASA SLS rocket and travel to lunar orbit in a NASA Orion spacecraft, where they will rendezvous with HLS and use it as a lunar lander shuttle. The HLS is essentially just a modified Starship upper stage, so SpaceX has to be able to launch a Starship upper stage, plus all the HLS gubbins and fuel, to the Moon in two years’ time.

Subscribe now

But Starship can’t go straight to the Moon; it needs refuelling in orbit first. The current idea is that a “tanker” variant of the Starship upper stage will be placed in Low Earth Orbit (LEO). It will be refuelled by multiple Starship launches, which will dock with the tanker and transfer their cryogenic propellant into the tanker before coming back down. Once the tanker is full, the HLS will launch, dock with the tanker, fully refuel, and then head to the Moon.

Now, here is the thing: NASA wants SpaceX to demonstrate all of this, as well as lunar landing and ascent with an uncrewed mission, before they conduct a crewed Artemis III mission using HLS.

For some context, the outgoing Starship V2 never made orbit and only ever carried 16 tons, or 16% of its promised payload, on a “successful” suborbital flight. Because it never made orbit, it never even attempted orbital refuelling. It also never successfully landed its upper stage.

So, what needs to be changed in Starship V3 to make it suitable for Artemis III?

Increased Payload to Orbit

Let’s start with the main factor: increasing the payload. As I mentioned previously, the V2 Starship only ever carried 16 tons on suborbital flights, but SpaceX claims it can carry 35 tons to LEO, which might be a bit of an exaggeration, to say the least. Either way, it falls severely short of the promised 100+ ton payload to LEO that is required to make this orbital refuelling shenanigans possible.

Firstly, it needs to successfully reach orbit with a payload, which might be more challenging than you think. The extra fuel required to take Starship to LEO is roughly 20 tonnes (as calculated by Thunderf00t), which suggests that Starship V2 can reach only the lowest possible orbit if it carries no payload!

So, how is Musk solving this horrific problem?

Well, I covered this topic before. The V3 has been fitted with SpaceX’s new Raptor 3 engines. By removing heat shielding and instead using more ablative cooling (which is when cryogenic fuel is used to cool the engine before it is burned), the Raptor 3 is 105 kg lighter than the previous Raptor 2, saving over four tonnes of weight per rocket. The Raptor 3 has also increased its power, delivering 9% more overall thrust. Additionally, the V3 Booster has 12% more propellant than the V2, and the V3 upper stage has 6% more. To accommodate this, the V3 is noticeably longer than the V2, yet its dry mass is reported to be a huge 20% to 30% lower than the V2’s, or 100 tonnes less.

This should increase the payload of Starship, but is it enough to go from a 16 ton payload on a suborbital flight to 100 tons to orbit? I very much doubt it.

There is also the issue of reliability. The Starship V2 upper stage failures were the result of engine flash events (caused by premature propellant ignition, which methods like ablative cooling can induce) and lack of structural integrity. So, increasing the rocket engines’ power, removing the rocket engine heat shields, making the rockets carry far more propellant mass, and transforming the rocket structure to be larger, yet lighter, could very well dramatically increase the risk of catastrophic failure. There is no point in increasing orbital payload if the chances of actually getting to orbit are minimal.

Starship V3 needs to prove that these changes can not only increase payload capacity by over 600% to 100 tons, but that it can achieve this feat reliably. I cannot exaggerate enough how gargantuan a task that is from Starship V2’s currently horrifically low baseline.

Refuelling

Starship V3 has so little time to test and prove it is capable of orbital refuelling that it basically has to do it straight out of the gate. This creates four major challenges.

Firstly, Starship has to prove it can safely remain in LEO for extended periods. This region is incredibly crowded, full of satellites like Starlink and the ISS, as well as a great deal of space debris, so it will need to constantly and actively avoid potential and catastrophically expensive impacts. This might sound trivial, but the “tanker” variant will be one of the largest and by far the heaviest object in LEO, making it incredibly arduous to manoeuvre. SpaceX needs to prove there is functionally zero risk of Starship becoming an LEO bulldozer before they can receive permission to park there for months or years.

Secondly, docking. Starship V2 wasn’t exactly controllable when in space, with the upper stage losing control multiple times during Flight 8 and Flight 9. Yet, to accomplish orbital refuelling, two 70-metre-long Starships, with a total mass of over 3,000 tons, will have to successfully dock with each other in LEO. Orbital docking of such large vehicles has never been attempted before, because it is insanely difficult! It will require pinpoint accurate control, and even then, the forces involved will be colossal. Not to mention the docking systems need to be strong enough to secure these massive bodies together and not fail. Furthermore, Starship needs to be able to pull this off with almost 100% reliability, as a mishap docking with a near-full tanker Starship could cause a horrifically expensive mission-ending explosion.

So, once V3 reaches orbit, it needs to prove it has totally reliable and near-perfect levels of in-orbit control.

Thirdly, there is the issue of actually transferring the propellant (fuel).

For its propellant, Starship uses pressurised cryogenic liquid oxygen and methane. These kinds of fuels are highly challenging to handle. Their freezing cold temperatures apply thermal pressure on components, as they unevenly contract with the drop in temperature, making leaks and total failures far more likely. Likewise, their high-pressure and extremely volatile nature means even a small leak will turn into a devastating explosion. In fact, as I covered in a previous article, handling these fuels is so risky that a Starship exploded while being refuelled down here on Earth! I even roughly calculated that if orbital refuelling has the same explosion rate as the current terrestrial refuelling, a Starship mission to Mars or the Moon has an 82.6% chance of ending in a giant fiery explosion during orbital refuelling.

In other words, Starship V3 not only has to prove it can handle orbital cryogenic refuelling but also that it can handle it with basically zero risk of explosive failure.

This is especially true when you look at the fourth major issue. Most outlets claim that the “tanker” variant of Starship will need refuelling ten times before it can fully refuel a Moon- or Mars-bound Starship. But that isn’t true, because boil-off exists.

Again, I covered this topic in a previous article, but even in space, the warmth of the sun will heat up this cryogenic propellant and cause it to boil-off. The rate of this boil-off could vary from 0.5% to 5% per day in orbit, and this alone could dramatically increase the number of refuelling launches required to get a Starship out of LEO.

For example, in my previous article, I was very generous and assumed a 1% per day boil-off rate, a 100-ton propellant refuel per flight (assuming Starship reaches 100-ton LEO payload), and a refuel flight rate of once per week. With all of those estimates combined, how long do you reckon it would take for the orbital tanker to accumulate the 1,600 tons needed to fully refuel a Starship?

Well, after 110 refuelling missions, which will take more than two years at this refuelling rate, it hits an impassable equilibrium of 1,428 tons of fuel, or 89% capacity. At this point, the tanker is losing 100 tons per week to boil-off, meaning the weekly refuelling flights are simply stemming the losses.

Oh, and I estimate that a realistic cost for a reusable Starship launch is $70 million. That would put the price of a Starship lunar mission launch at $7.7 billion, or nearly four times the cost of an equivalent NASA SLS lunar launch.

Even a tiny amount of boil-off dramatically increases the cost and time required to send a Starship to the Moon. Because each refuelling mission will create a risk of mission-ending tanker explosion, boil-off also dramatically increases the risk of the mission flat-out failing.

So, to be able to conduct the Artemis III mission in late 2028, V3 needs to launch and begin refuelling its orbital tanker by the end of this year. That means that over the next nine months, V3 has to reach orbit, achieve its promised 100-ton payload to LEO, test and demonstrate near-perfect reliability in orbital refuelling, reduce orbital boil-off rates to 1% per day or lower, and increase its launch rate from its current once every two months to once a week. But don’t forget, NASA wants a demo beforehand, so really, SpaceX needs to send a tanker to orbit and refuel it months before this, as well as reaching a launch rate of twice per week to get a demo lunar mission tanker and an Artemis III tanker ready.

That is a monumental leap forward that Starship has to take in just a few months.

Upper Stage Reusability

For this colossal expansion of Starship launch frequency to even be financially possible, the V3 has to crack upper-stage reusability. As I have covered before, a Starship upper stage likely costs well over $100 million to build. So if SpaceX can’t figure out reusability, launch costs will more than double!

For some context, my $70 million-per-launch estimate assumes that a Starship upper and lower section has a lifespan of 33 launches with minimal maintenance. That might not sound that bad, considering a Falcon 9 booster has a lifespan of up to 34 flights, and while it still needs serious refurbishment between flights, they have lowered the reflight time to as little as nine days! Not only that, but SpaceX has conducted soft splashdowns with the upper stage. So, surely they are close to reuse?

Unfortunately, relaunching a booster and an upper stage are very different things. The upper stage of Starship has to reach orbit speeds, and this extreme velocity means that during landing, it has 72 times more kinetic energy to scrub off than the booster! Air friction dissipates a large portion of this energy into the giant heat shield of the upper stage.

And even worse, it doesn’t look like SpaceX has a heat shield for Starship that is capable of surviving this heat and being reusable. Thunderf00t posted a great video analysing a fragment of a Starship heat shield tile, and its old-school material choice and janky construction seem to be optimised to be replaced each flight rather than being suitable for multiple flights. Considering how large and expensive such a heat shield is, that is not a suitable alternative. Indeed, there is speculation that these heat shield tiles aren’t actually good enough and that the inside of Starship is getting excessively hot during landing, which could damage structural integrity and systems, causing it to require extensive refurbishment before reflight can be attempted.

V3 has to demonstrate that SpaceX can successfully land and rapidly reflight a Starship upper stage. That means a heat shield that adequately protects internal systems and structures from thermal damage while being robust enough to survive multiple launches without major maintenance. However, as Thunderf00t points out in his video, science hasn’t yet produced a material capable of this that doesn’t incur a colossal weight penalty (which Starship can’t afford to carry). So, really, V3 has to pull a rabbit out of the hat here.

In Time

Again, Artemis III is slated to launch in late 2028. Even optimistic assumptions about Starship’s payload and boil-off rates mean SpaceX has to put two tanker versions of Starship in LEO and launch two refuelling missions a week by the end of this year to meet that deadline. Consequently, for Artemis III to happen on time, Starship V3 needs to finally reach orbit, increase payload by over 600%, improve launch safety, figure out orbital docking between two Starships, resolve the orbital cryogenic fuel transfer problem, improve the safety of orbital cryogenic fuel transfer, make the entire rocket fully reusable, and scale up launch frequency from once every other month to twice a week (or, in other words, increase launch rate by 1,700%).

Can SpaceX do all of this in time?

Personally, I don’t think so. At most, I think the V3 will be slightly less of a failure than Starship V2. But that is my opinion, and I’m just some guy on the internet. The real question is, do you think SpaceX can do this?

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

You could fit every techbro CEO’s colossal, distended ego inside the gaping chasm between what AI promised to do and reality. Every now and then, we get a perfect snapshot of this discrepancy, and internet sleuths have just provided us with another one. Microslop has been shoving Copilot down users’ throats for a while now, heavily pushing it as the future of professional productivity. However, outlets like TechCrunch, TechRadar, PCMag, and others recently broke the news that Microsoft’s Terms of Service (ToS) state that Copilot is for entertainment purposes only, sparking a tsunami of online derision. Unfortunately, this is so much worse than people realise, because this isn’t just limited to Copilot, and the implications of this run significantly deeper than you might think.

The Terms of Service

To use Copilot, like most modern software, you have to agree to its Terms of Service (ToS). So, let’s start with what Copilot’s ToS actually says. In the “Code of Conduct” section, it states that “Copilot is for entertainment purposes only. It can make mistakes, and it may not work as intended. Don’t rely on Copilot for important advice. Use Copilot at your own risk.” In other words, Copilot isn’t accurate or reliable enough to be trusted with even remotely important tasks or as a source of information, meaning it should be treated more like a toy than a tool.

But Microsoft also doubled down on that “at your own risk” statement. In the “Important Disclosures And Warnings” section, it states that “You agree to indemnify us and hold us harmless (including our affiliates, employees and any other agents) from and against any claims, losses, and expenses (including attorneys’ fees) arising from or relating to your use of Copilot, including without limitation your use, sharing, or publication of any Prompt, Responses, or Creations, or your breach of these Terms or violation of applicable law.” In layman’s terms, if you use Copilot as a tool and not just for entertainment (you know, like how it is advertised), then you can’t hold Microsoft liable for any damage it may cause to you personally or the business you work for.

Subscribe now

So, let me get this straight — this tool, which has been slated as the next big thing in professional productivity, is so catastrophically unreliable that Microsoft had to not only explicitly state it is just for entertainment purposes but also force users to completely surrender any right to hold them accountable for damages caused by using this unreliable, inaccurate AI as a professional productivity tool. That doesn’t seem right, does it?

Some have pointed out that these ToS only cover Copilot as a personal chatbot and don’t apply to Copilot business tools. But that isn’t really a defence. For one, many businesses still use these “individual” versions of Copilot as a chatbot assistant. Secondly, people also use Copilot as a professional assistant chatbot in both their personal and professional lives, as that is essentially how it has been advertised. But finally, these Copilot business tools are based on the same AI models, so they will have similar issues.

And here is the thing: it isn’t just Microslop pulling this legalese bulls**t.

Anthropic, the less-evil AI company, is also doing this. Their consumer ToS (when viewed from a European IP) states that their services are “non-commercial use only” and that users “agree not to use our Services for any commercial or business purposes and we (and our Providers) have no liability to you for any loss of profit, loss of business, business interruption, or loss of business opportunity.” I wonder how many developers vibecoding with Claude Code know that.

To be fair to Anthropic, they are still better than Microsoft. They don’t completely retract any liability they have for damage their models may cause — but they do heavily restrict it. The ToS states that Anthropic’s “total liability to you for any loss or damage arising out of or in connection with these Terms, whether in contract (including under any indemnity), tort (including negligence) or otherwise will be limited to the greater of: (a) the amount you paid to us for access to or use of the Services in the six months prior to the event giving rise to the liability, and (b) £100.”

This is the exact same schtick Microsoft has pulled. The AI is so inaccurate that it can’t be trusted with meaningful tasks, so they claim it can’t be used for commercial purposes (even though that is primarily how it has been marketed) and then retract as much liability as possible for their AI’s actions.

OpenAI is not much better.

Their ToS does not contain an agreement not to use their services commercially. But it probably should, as Copilot is practically just a reskinned ChatGPT. Therefore, this lack of a disclaimer can more likely be attributed to Sam Altman’s far looser and riskier “break things and move fast” mentality than an improved product.

What it does include is a giant section in which it painstakingly tells users not to trust the output of their AI. For example, one section reads, “Output may not always be accurate. You should not rely on Output from our Services as a sole source of truth or factual information, or as a substitute for professional advice.” It goes on to say, “You must evaluate Output for accuracy and appropriateness for your use case, including using human review as appropriate, before using or sharing Output from the Services.” In other words, it shouldn’t be used to automate anything professional, and its output should always be reviewed by a human. That is so close to saying “not for commercial use” without actually saying it.

Also, they fully remove any liability they might incur for their AI’s actions. The ToS states, “NEITHER WE NOR ANY OF OUR AFFILIATES OR LICENSORS WILL BE LIABLE FOR ANY INDIRECT, INCIDENTAL, SPECIAL, CONSEQUENTIAL, OR EXEMPLARY DAMAGES, INCLUDING DAMAGES FOR LOSS OF PROFITS, GOODWILL, USE, OR DATA OR OTHER LOSSES, EVEN IF WE HAVE BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.” And you can tell they really meant this because they literally wrote it in block capitals.

These AI companies are trying to have their cake and eat it too. Supposedly, these AIs are so good that they can replace workers and take businesses to the next level, which is one of the main pumps inflating the AI bubble. Yet they still have to warn users not to trust their AIs, as they are dangerously inaccurate, as well as explicitly rejecting any liability for the tools they release into the world. It’s either one or the other — you can’t have it both ways.

The Liability Paradox

All three ToS documents say virtually the same thing. Firstly, the AI is unreliable and can’t be trusted. Secondly, it can and will cause damage if unchecked, and these companies won’t take liability for such damage. And thirdly, these AIs need to be micromanaged by humans checking every single output to work.

This creates a bit of a paradox. AI is meant to be a productivity tool, capable of augmenting or automating tasks and jobs. But, if it is such an inaccurate liability that a human has to micromanage it, how on Earth can it boost productivity?

I discussed this subject in a previous article, but a number of studies, such as this one from the University of Melbourne, demonstrate that AI is too unreliable to be a productivity tool because it requires more time to micromanage the AI in order to catch and correct its constant errors than it saves.

This Can’t Be Solved

So, what are the implications of all of this?

Well, let’s start with the fact that we knew this was the case, and it isn’t going to improve.

Take this study from Carnegie Mellon University, which found that even the best “agentic” AIs fail basic tasks 70% of the time. Or, what about this study that found that the best current AIs failed 97.5% of realistic real-world freelancing jobs given to them due to AI hallucinations and total failures? We know that these AIs are just statistical models; they aren’t thinking machines, meaning they will statistically get things wrong and hallucinate very frequently. We shouldn’t be surprised that AI companies’ small print warns against using their models commercially and washes their hands of any meaningful liability. Everyone knows, including AI companies, that these models are dangerously inaccurate.

But the promise that these AIs may eventually be good enough is what is driving the currently insane levels of investment in AI. In order to produce AI models capable of replacing human workers, hyperscalers are literally pouring hundreds of billions of dollars into building larger data centres and providing AI models with more compute power.

Only, even OpenAI knows that isn’t going to happen. They published a study last year that found that increasing the computing power behind AI, or providing it with more data, can’t reduce AI “hallucinations” (or error rate) below its current level. In fact, they found no viable way to reduce AI hallucinations, which strongly implies these models are doomed to remain as unreliable as they currently are. A more recent paper by Vishal Sikka and his son Varin Sikka has supported this idea, revealing that AIs are mathematically incapable of being reliable or carrying out computational and agentic tasks beyond a certain complexity.

So, what does that say about the hundreds of billions of dollars being poured into AI to “unlock” it?

The Big Implication

Quite simply, all of these factors point to generative AI being a dead-end economic bubble. The hype and investment are not at all reflected in reality. These ToS documents prove that these AI companies are, at least to some degree, aware of this discrepancy. They know their models aren’t good enough. They know they can and will cause damage. They know they aren’t going to meaningfully improve. Yet, rather than stopping the madness, they are actively attempting to protect themselves while still forcing the hype and inflating the bubble, all to line their already cavernously deep pockets. This is a tale of greed and deception, and we all know how it will end.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

I have called the AI boom a death cult before because it is in so many different ways. From its total lack of financial sustainability to its horrific environmental impact to the veritable psychopathic psychosis of the tech bros pushing this technology onto us, every decision made is an almost fetishistic attempt to dominate us and prove to Daddy Shareholder that they are still the golden child. But the elephant in the room is the other way AI is spiralling towards destruction. You see, AI is choking the data economy to death, which is the foundation that the generative AI industry depends on, in two distinct ways.

The “Biting The Hand” Problem

Let’s start with what I’m calling the “biting the hand” problem.

It’s not exactly a secret that AI needs to be constantly fed a metric f**k ton of training data in order to stay up-to-date, which has meant tech companies have been compelled to scrape the internet to source enough. After all, copyright isn’t a thing… right? It’s also no secret that, to justify the insane cost of AI, tech companies have been shoving it down our throats and trying to get us to access services or information through these bots.

Subscribe now

For a while, most people seemed to think these problems only materially affected other people, such as professional authors or writers. But, as it turns out, this is a problem that affects us all.

A recent report from Chartbeat and Axios investigated how Google’s AI search summaries were affecting online publishers of all sizes. To no one’s surprise, it turns out they are being utterly mullered.

Small publishers, such as sites with 1,000 to 10,000 daily views, have experienced a 60% decline in traffic from Google. Medium publishers, such as sites with 10,000 to 100,000 daily visits, saw a 47% drop, and large publishers, with more than 100,000 daily visits, saw a 22% drop! Another study found that online publishers saw an 80% decline in traffic from AI summaries.

Pew has the answer to what is causing this decline. They found that users are far less likely to click on links when Google’s AI summary appears in the results. In fact, users are twice as likely to click on links when this feature isn’t active!

Why is that bad? Well, even after this monumental downturn, Google Search remains the dominant source of traffic for these publishers.

Just to remind you, the term “publishers” doesn’t just include news websites. It also refers to recipe websites, how-to websites, blogs, and independent voices — basically, anything you read online.

Now, here’s the thing: these publishers need traffic to generate the income that sustains them. Less traffic means less money. Less money means these sites produce less content. That means that these AIs will have less valuable data to be trained on.

Can you see the problem here? AI is biting the hand that feeds it.

Apparently, using AI to replace human connections crushes the human output AI depends upon. Who would have thought?

Because the smaller publications, which tend to be more diverse and independent, are being affected the most, this will impact our media landscape and these AIs directly. It will flatten the curve and make both the written media we consume and the AIs trained on it more narrow and generic by squashing the more niche, diverse, and unique voices. I can’t stress enough that this is bad for both AI and these publications.

However, in an effort to cut costs, maximise shareholder value, or brute force algorithms, a sizeable portion of this written data is AI-generated anyway. In fact, this squashing of small and medium-sized publications is likely going to make AI-generated online content more prevalent. This brings up the second major problem with the AI data economy.

The “Eating Its Own Tail” Problem

The only thing generative AI does is find and replicate statistical trends in giant datasets. But even the best AIs aren’t perfect. After all, a trend can exist statistically but not technically be real. We, humans, have this problem too. Have you ever seen a face in something that doesn’t have one?

This means that weird things begin to happen when you train an AI on AI-generated dataYou see, AI-generated content contains tiny, almost indistinguishable trends that human-generated content doesn’t have. That is why we feel we can spot when AI has written something. If you then start feeding this data back into the AI model, it will place more and more weight on these non-human trends. At first, this looks like the AI becoming less capable, but if you continue feeding the AI its own output, it will eventually place more weight on these generated trends rather than the human ones, causing the model to collapse and produce gibberish. We call this “model collapse,” and it is a surprisingly well-studied phenomenon.

Why does this matter? Well, let’s quickly run through some stats.

According to Axios, by the middle of 2025, over half of the content being posted online was AI-generated.

Online scraping accounted for at least 82% of ChatGPT-3’s training data. We do not know what proportion of the current AI models’ training data was scraped from the web. However, AI companies have claimed they are “running out of data” and have had to resort to scraping low-quality data from the web to fill the gap. So we can safely assume this proportion is at least as high, if not higher.

Ready for another doozy? AI detection tools are terrible! They can detect AI content 57% to 95% of the time. But their false positive rate, where these tools label human-written content as AI, is considerable, with one study even finding a 50% false positive rate for a leading tool. Even worse, the false-positive rate of these tools is biased against diversity, labelling writers who are neurodiverse and who speak English as a second language as AI more frequently.

Put simply, AI companies are using considerable amounts of web-scraped data to train their AI. But more than half of the content posted online is AI-generated, and AI detection tools fail to consistently filter out AI content and incorrectly filter out human-made content. This means that today’s AIs are being trained on their own output, or even the output of their predecessor models (which also causes model collapse). This isn’t a hypothetical problem; researchers have found a very real, tangible risk that current AI models are headed toward model collapse.

There is a proposed solution to this problem: synthetic data. This is AI-generated data designed specifically for training the AI. However, this training method falls short, as it doesn’t improve the models much, exaggerates their flaws, harms real-world performance, and risks model collapse. As such, synthetic data is more of a gimmick than a replacement for genuine, high-quality human-derived training data.

This really is a lesson in not shitting where you eat.

AI companies have polluted the web with a tsunami of AI-generated content. Naturally, the very data these AI models depend on is being contaminated beyond recognition, which contaminates the AI themselves and sets off a vicious cycle. Not only that, but the sheer volume of AI-generated slop online is drowning out interesting, diverse and valuable human voices, reducing the internet to a bland monoculture. Of course this is horrific news for us humans, but it ain’t good for these AIs either, as it means there is most likely less “high-quality” human data being posted online than before.

It’s almost like it was a bad idea to unleash an unregulated plagiarism machine that is, at best, a hollow and deeply flawed attempt to mimic humanity onto the world…

Summary

In layman’s terms, the AI information economy is detrimental to everyone, including AI companies. It is crushing and robbing blind the very people producing the data it depends on and is flooding the internet with so much slop it is at risk of destabilising itself. This is a totally unsustainable situation. Can it be solved? Yes, regulation, copyright laws and copyright reform could all help. But these solutions involve taking power away from Big Tech, which is becoming harder by the day. It seems we are all trapped in yet another downward spiral. The question is, will we take the steps to escape before it is too late?

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

We have all heard of AI brain rot, AI psychosis, and AI slop. If you spend any time online, it’s quite obvious that the combination of social media and AI isn’t exactly healthy for your neurons. What isn’t talked about as often is the mental impact of using AI at work, despite it being potentially more damaging. Thankfully, this issue is now beginning to receive headline coverage. But most publications fall short of explaining why using AI at work can be so harmful and totally neglect to mention that we were warned about all of these issues from the start. Welcome to the world of AI deskilling.

Business Insider recently published one of these articles. The piece profiles Josh Anderson, a highly experienced software consultant, who shared his experience developing a new app, Road Trip Ninja. He conducted a little experiment and tried to get AI to write the entire codebase. Initially, things went great, but as the code ballooned past 100,000 lines, and interactions with the chatbot grew from minutes to hours, Anderson became increasingly frustrated when progress slowed to a halt.

Subscribe now

Of course, this was just an experiment; Anderson could have stepped in at any time and coded the app himself, even if sorting out such a huge block of AI-generated code with very few in-code comments is insanely difficult. But Anderson’s experience highlighted a glaring problem. You see, even Anthropic has found that using generative AI coding tools dramatically reduces a coder’s skills in debugging and code comprehension. So, with the direction the software industry is heading, could a coder actually step in and finish what the AI couldn’t? This article explains that Anderson’s experience “raised questions about the real impact of AI on skill retention and development” and that it “highlights a broader concern among workplace researchers: the risk of deskilling in an environment increasingly reliant on AI.”

Given that this is an industry-wide issue, the article didn’t just look at Josh’s experiment. It also highlighted that developers admitted to finding tasks considerably more challenging during Claude’s recent outage, which rendered their AI assistant useless and indicated a “dangerous dependency.”

This issue has a variety of names and similar explanations that the article quickly glances over.

For example, Josh Nosta calls it the “AI rebound effect.” He describes it as when an AI-driven increase in productivity hides a decrease in skill levels. As he put it:

“When automation handles the details, situational awareness dulls. And in that context, we scan less, anticipate less, and make fewer micro-adjustments. Simply put, the mental models we rely on to navigate complex situations shrink because the system is doing what we once did ourselves. Over time, this isn’t just about pausing a skill; it may be more akin to erosion. And when the technology steps away, the skill doesn’t simply return to baseline. It can come back lower.”

In other words, skill and expertise are muscles that need to be trained to be maintained or they will be lost. So, automating these decisions with AI can lead to us losing critical skills.

Dr. Rebecca Hinds calls this “cognitive debt.” As I covered in a previous article, Dr. Hinds is equally worried about the atrophying of critical skills. She discovered that if AI is used as a shortcut to automate tasks, increase work scope, or reduce workforce size, workers will lose critical expertise and skills because they aren’t being maintained and reach a dangerous level of false confidence, increasing the likelihood that mistakes will be missed. Dr. Hinds instead suggests that AI should be used in tandem with experts, providing them with options while ensuring the expert remains the one making the decisions. Sadly, that is not how AI is being used, and whether using AI like this genuinely increases productivity is still questionable to many.

Colloquially, this issue is known as AI deskilling. This is commonly understood to be when AI is used to automate or augment workers and it shoulders most of their cognitive load. But it is that load that builds and maintains these workers’ critical skills. So, deploying AI in this manner inherently erodes critical skills in a workforce.

This isn’t a hypothetical. We have known about this for quite a while.

For example, there is the 2023 study from JYX, which analysed how automation in an accountancy firm directly led to skill erosion and a marked reduction in critical thinking skills (complacency) that was negatively impacting the business. Or the early 2025 study by Carnegie Mellon, backed by Microsoft, which surveyed 319 “knowledge workers” and found that generative AI automation and augmentation caused a serious loss of critical skills and critical thinking. Or what about this more recent study, which found that generative AI augmentation and automation in medicine are eroding physicians’ critical skills, meaning performance would drop below their previous non-AI baseline if AI were removed?

If you think all of this means AI can still be used to automate or augment low-skill jobs, read my previous article to find out why this simply isn’t true.

Okay, but why does it matter that these workers are losing their skills and expertise? Take the physician example. This study found that generative AI augmentation actually improves the physicians’ efficiency. So, does it matter that these skills are gone if they are no longer needed?

Well, yes, it does matter, because this expertise and these skills are still very much needed and haven’t been truly replaced.

Firstly, you can’t rely on AI. Take the Claude example from before. These AIs sometimes experience outages. A marked decrease in performance or capability during those times could be unacceptable. Imagine if a physician struggled to diagnose patients because ChatGPT was offline! But these AIs are also regularly updated or have their functionality tweaked, which can interfere with how they integrate into a work environment, causing workers to need to use their critical skills to address the situation.

Then, there is the issue of training these AIs. To operate in these fields, they need to be constantly trained on an enormous amount of detailed data to ensure the models are as accurate and up-to-date as possible. These skilled experts provide that data, but if they are being deskilled, where is that data going to come from? There is a possibility that deskilling a workforce could make these AIs noticeably worse.

Let’s also not forget that these AI companies are not profitable and could go bankrupt and disappear in the not-too-distant future. We really shouldn’t make ourselves dependent on them.

But we also seem to forget that AI is not a complete solution and that AI augmentation doesn’t mean these skills are not required. We saw this clearly with Josh Anderson. Anderson had to put in exponentially more effort to get the AI to complete the task because it was incapable of doing so on its own. It really required someone to come in and independently finish the job, which calls for every bit of coding skill needed to comprehend the code, follow its logic, debug issues, connect the separate parts together correctly, and make the whole thing efficient. In other words, the deskilling might go totally unnoticed until the day these skills are required to finish a critical task, which then can’t be completed.

This was obvious during Amazon’s recent outages. I covered this topic in a previous article, but Amazon has laid off a significant number of engineers and has effectively tried to replace them by augmenting the engineers it has left with AI. However, it turns out many of these laid-off engineers were highly skilled in preventing and resolving outages, and this expertise was lacking in the remaining teams. This has created a wave of gigantic, frequent, and extremely costly outages. Who would have guessed?

Oh, and all of these attempts are likely in vain. Remember that Anthropic study from before? It found that the productivity gains from using AI coding tools were “failing to reach statistical significance.” So, at least in some industries, there is no measurable payoff to offset the huge downside of deskilling. It is a lose-lose situation. Generative AI coding tools are more expensive, but they fail to deliver a boost in productivity while also causing the coders to rapidly lose critical skills that are still required… Sounds like a bum deal to me.

But is this really an industry-wide issue?

Well, for developers, it might be. Everything we have discussed could explain the recent findings of METR. Their previous 2025 survey revealed that AI coding tools slowed down expert coders by 20%, as the amount of time spent correcting the AI was more than the time saved using AI. However, because too few coders were willing to work without generative AI assistants, the 2026 survey was deemed “unreliable.” Why? As Anthropic found, it isn’t like these coders are making considerable productivity gains by using AI. Well, if much of the industry has experienced AI deskilling, this would explain why. Many of these coders could now lack the skills to code without AI help and so are unwilling to do so.

There is also a secondary issue here. Roughly 84% of coders use generative AI coding assistants. That means surveys like METR’s might struggle to find a control group of coders who aren’t AI-deskilled and capable of coding well independently. In other words, this industry-wide deskilling issue could skew their future studies.

What does all of this mean? Well, without serious restrictions in place, AI is a skill and expertise bomb. This would be a problem if it were being deployed this way in a few small, isolated cases. But AI is being slapdash-deployed across whole sectors and industries, and, as such, it threatens to erode our collective skills and abilities. This makes it a major problem that will affect us all. How we solve this problem is a conversation for another day. AI regulations, expanding workers’ rights, and corporate restructuring are just a few of the possible options. But for now, I am just grateful that this issue is hitting the headlines — even though it deserves far, far more coverage.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

Sora, OpenAI’s text-to-video AI model, was the pinnacle of AI slop. As soon as Altman launched it in late 2024, the internet was flooded with a tsunami of objectively bad, utterly nonsensical, and often legally dubious 20-second videos that fell firmly in the uncanny valley of not-real-enough. Yet, AI glazers and tech bro fanatics still ran around like headless chickens in response, claiming this technology was enough to make Hollywood obsolete. The mismatch between reality and speculation had never been wider. Now, a little more than a year later, OpenAI has shut down the Sora model and its TikTok clone app. Not only that, but Disney has cancelled its $1 billion investment in OpenAI to licence its IP for Sora to use. So I guess Hollywood is safe for another day — what a shocker! But many have pointed to Sora’s demise as a sign that the AI bubble is bursting. So, is it? Well, yes and no. Let me explain.

We first need to ask why OpenAI killed Sora, as that is the key to this entire ordeal.

Why?

In short, money. Sora was nothing more than a liability and a cash bonfire.

Subscribe now

Let’s start with the fact that AI video isn’t very useful for commercial (i.e., paid) purposes. Its lack of consistency between clips, total lack of emotional nuance, distracting levels of uncanny valley, awful artifacting, and noticeably inconsistent depth and placement make AI-generated video functionally useless for commercial purposes. So, Sora was never going to land any big-paying customers and would have to rely more on “hobbyist” use.

This is a problem because AI video models are insanely expensive to operate. One study found that a ten-second AI-generated video uses 30 times more energy than an AI-generated image, and that AI-generated image uses 2,000 times more energy than typical text generation. Indeed, one analyst has estimated that Sora cost OpenAI $15 million a day to operate! Yet, because Sora was only ever really used by people “playing around,” it brought in virtually no revenue.

So, this product cost OpenAI billions of dollars a year to operate, yet it produced practically zero revenue because it was only really useful for making slop videos.

There is only one possible justification for keeping such a profound waste of money on the books: social impact. If Sora had a huge social impact and functioned as incredible PR for OpenAI, they could justify keeping the service going.

But this is where the Sora 2 app comes in, because, despite the hype, it was a total failure.

Read more

There are so many signs that the AI industry exists in the mother of all bubbles that it can be hard to see the forest for the trees. For example, the total lack of productivity growth, zero GDP growth from AI, OpenAI’s own research on the limitations of today’s models, and the countless studies that show just how useless these machines are. But possibly the most interesting is the recent revelation that AI insiders, who arguably profit the most from this bubble, are preparing for the entire thing to collapse in just a few years. This isn’t as significant a red flag as it sounds. Businesses make such contingencies. But it is a deeply insightful piece of context that reframes the entire AI hype train.

To explain why this revelation is so important, we need to recap a few things.

As I have covered before, the AI industry is currently neck-deep in circular financing. AI companies like OpenAI, AI hardware operators like Amazon or Coreweave, and AI hardware suppliers like Nvidia are funding each other to the tune of hundreds of billions of dollars in a large-scale corporate circle jerk to increase each other’s value and prevent loss-making AI labs from collapsing. For example, Nvidia and Amazon recently gave OpenAI tens of billions of dollars, but in return, OpenAI will use almost all this money to buy their AI chips and use their AI data centres. Unfortunately, OpenAI’s annual losses are only growing, as its models become more and more expensive to train and operate. So it needs a constant flow of these gargantuan cash injections to stave off bankruptcy. In other words, established data centre giants like Nvidia and Amazon are funding colossal, unprofitable AI companies to drive up demand for their hardware, operations, and sales and, in turn, increase their share value.

When viewed through this lens, the AI bubble could be seen as Big Tech using AI to grift the economy and increase its value.

Subscribe now

Now, admittedly, AI companies like OpenAI have other sources of financing, such as debt. But they can only rack up so much debt before it completely decimates them. As such, this drive to increase demand for AI data centre sales and operations through circular financing is mainly responsible for the AI bubble.

Why does that matter? Because it has caused the RAMpocalypse.

AI infrastructure requires truly enormous amounts of high-speed computer memory. This has caused demand for high-speed memory such as DRAM, VRAM and HBM to skyrocket over the past year and a bit. Sadly, production of these chips is slow to expand, causing demand to significantly outstrip supply and prices to explode. Tech enthusiasts have labelled this the “RAMpocalypse”, as the consumer computer component most affected by these insane price hikes is RAM. Some variants of consumer RAM have increased in price by 500% over just three months!

Those inside the AI circular financing bubble are not the manufacturers of these chips. In fact, there are only three main manufacturers involved: Samsung, SK Hynix, and Micron, and they sell the chips to companies like Nvidia. All three are set to make record profits from these price spikes and arguably might be the ones profiting the most from the AI boom, with them set to pocket around half a trillion dollars. Selling shovels in a gold rush and all.

Now, all three manufacturers are trying to expand production to meet this higher demand. SEMI forecasts that overall 300mm wafer output (used to make these types of memory) is set to increase at a 7% annual growth rate from late 2024 through 2028. But, when you consider that new AI data centres are predicted to use 70% of the memory chips produced in 2026 and that we are expected to deploy exponentially more AI data centre capacity each year until long beyond 2030, that supply increase likely can’t keep up with demand.

Simply put, the AI bubble is directly causing the RAMpocalypse. If the AI bubble continues to grow, the RAMpocalypse will follow in its footsteps. But the opposite is also true. If the AI bubble pops, then the RAMpocalypse will end because memory demand will crash.

And guess who knows that? These memory manufacturers.

As reported by Chosun, Samsung, the largest of the three memory manufacturers, expects the current memory shortage to end by 2028. Understandably, Samsung is cautious about overinvesting in production capacity expansion so as not to be stung by this predicted downturn. In other words, they don’t want to get whiplash from trying to capitalise on the current skyrocketing demand, only for that demand to collapse, leaving them with cavernous losses and a supply-side glut.

This is a surprising revelation to say the least, as back in January, Samsung was reportedly looking to expand its memory production capacity by 50% by the end of this year.

If you add one and one together here, it is very obvious that Samsung has realised there is a very real chance that the AI bubble will burst in the next two to three years, has developed cold feet, and is seriously rethinking its planned expansion. This doesn’t necessarily mean they believe the bubble will burst but that there is a strong enough likelihood of this happening for them to create contingency plans that could limit their near-term growth.

So, what happened in the past few months to spook Samsung so deeply?

We don’t know, but it does align with a broader realisation that the AI bubble is rapidly spiralling towards destruction.

For example, by late January, it was becoming increasingly obvious that OpenAI was growing desperate. Its Sora and Atlas projects had functionally failed, its deeply worrying financial issues had become more publicly known, and Altman announced OpenAI would roll out ads, which he previously called a last resort. This led many analysts and pundits, including myself, to predict that OpenAI would go bankrupt in the next few years. One of the most prominent was The New York Times’ Sebastian Mallaby, who estimated that OpenAI will declare bankruptcy between now and sometime in 2027.

While Anthropic is beginning to overtake OpenAI in terms of users, OpenAI is by far the largest driver of the financial side of the AI bubble. So, if OpenAI goes bust, it very well could pop the entire bubble.

And I think it was revelations like these that made Samsung reassess what they were getting themselves into.

Again, Samsung’s newfound caution isn’t the red flag it might seem at first glance. This isn’t some ground-breaking scoop where I can predict the exact date the tech bros’ empire will crumble. But it is deeply telling that the company that is arguably profiting the most from the AI boom is beginning to view it as a bubble that could blow up in its face soon. Tech bros are on a colossal propaganda campaign to give us all AI FOMO (Fear Of Missing Out). Yet, those who would profit the most from this bubble have a fear of being burnt by hollow promises and an imploding industry. This simple piece of context potentially reframes the whole narrative.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

Nvidia CEO and professional AI glazer Jensen Huang recently claimed that we have already achieved AGI (Artificial General Intelligence). Firstly, that raises serious concerns about his definition of intelligence. Current AI systems are more akin to a deeply hallucinating, plagiaristic sycophant than to any form of coherent intelligence. The toothless, tin-hat-wearing cider-addled man propping up my local pub from 11:00 AM every morning has infinitely more intelligence than these “flatten-the-curve” statistical slop machines. That guy is also infinitely more fun to talk to. But secondly, that simply ain’t happening, Chief! And Jensen would know that if he took a break from counting the billions of dollars he has earned in circular financing and actually looked at generative AI’s capabilities in the real world. You know, where intelligence isn’t some pseudointellectual, speculative bullshit concept but instead critical to real-world results. Take Amazon, for example. For the third time, they have learned the painful lesson that generative AI is not intelligent, can’t replace human intelligence, and isn’t a productivity tool. Well, I say “learned” — what is that fake Einstein quote about the definition of insanity? Something about doing the same thing over and over again, expecting different results?

The Recent “Lesson”

Earlier this month, the Financial Times reported that Bezos’s favourite little monopoly had effectively called a giant emergency meeting of its remaining engineers to try and fix the rapidly increasing number of outages taking Amazon.com down. These aren’t little blips either. A week before this meeting was called, Amazon’s main shopping website was down for six hours! This one outage could have cost Amazon over $490 million in sales, given that $717 billion was spent on Amazon.com in 2025. Let’s just say that the bald man with more in common with Smaug than the rest of humanity wasn’t too happy about that. This meeting was an all-hands-on-deck moment. The engineers were expected to find the source of the problem and fix it.

And guess what the problem was?

Amazon’s own AI…

Subscribe now

According to the official line, generative AI was a “contributing factor” in the botched “software code development” that caused these outages. But that is a bit like saying the untimely death of Archduke Franz Ferdinand was a contributing factor to World War I. This reeks of a PR spin designed to hide the embarrassment of the own goal that is Amazon’s AI “transition”, particularly when you consider the actual problems causing these outages, the engineers’ solutions to prevent them, and the wider context of Amazon’s recent business decisions. It all points to AI being the culprit.

Take the 13-hour AWS outage incident from December of last year. Last month, the Financial Times reported that Amazon’s own “agentic” Kiro AI coding tool was to blame. Engineers had allowed Kiro to make changes to Amazon’s AWS code and make “autonomous decisions”. As it turns out, Kiro ain’t that clever; it pulled a Musk move and deleted the entire working code environment before recreating it from the ground up with a ton of fatal bugs. In fact, the FTfound that Kiro caused outages like this not once, but twice!

Indeed, it seems to be both wild “agentic” AI and AI slop coding that are the culprits behind Amazon’s outages, and the smoking gun is the emergency solution these engineers came up with. Are you ready? Their solution is to require junior and mid-level engineers to ask senior engineers to sign off on any AI-assisted changes. This is almost fully admitting that AI caused all these outages.

But why are these engineers using AI like this? After all, 96% of professional coders explicitly don’t trust AI-generated code. These guys know giving it the keys to the kingdom was a bad idea.

Well, they were basically forced to.

Amazon has laid off thousands of engineers and plans to soon lay off around 30,000 workers, all while their major services, like AWS, expand dramatically. These services simply can’t be run on a skeleton crew, which makes this an obvious attempt to replace workers with AI automation. Indeed, last year, while these layoffs were happening, leaked documents showed Amazon’s plans to replace 75% of its workforce with automation and AI.

In short, these engineers are likely so stretched that they are forced to turn to AI to speed up their output. On top of that, Amazon recently mandated that 80% of its engineers use Kiro at least once a week. This isn’t necessarily a problem, but because they are so stretched, they don’t have the time to check the AI’s outputs, which practically guarantees these fatal mistakes will happen over and over again.

In other words, AI, despite its name, isn’t actually intelligent and is no replacement for genuine human intelligence in the real world. (I hope you are taking notes, Huang.)

But once again, Amazon has also proved AI isn’t a productivity tool either.

It completely kills productivity for these engineers to ask junior and mid-level engineers to obtain senior engineers’ approval for AI-assisted changes. Amazon engineers are expected to use AI coding tools like Kiro. So this means almost every line of code now has to be reviewed and approved by a senior engineer. Being a jumped-up debugger is not part of a senior engineer’s job description! This is a huge bottleneck for junior and mid-level coders, who are already far too understaffed, and it burdens senior engineers with heavier workloads and scope bloat, which detracts from their main responsibility of ensuring the entire project actually functions on a wider scale. In other words, this AI was implemented to make these departments more productive, but that decision led to a steep and damaging decline in quality. So, Amazon’s solution is to make these teams far less productive from top to bottom through enforcedmicromanagement.

Once Bitten, Twice… Bitten?

In a previous article, I covered an eerily similar situation that took place at Amazon back in October 2025. AWS had totally crapped the bed and briefly took out half the internet. Nearly all of AWS was down for 16 hours straight due to a simple DNS resolution issue, which impacted thousands of businesses, including Medium and Substack. I can vividly remember being unable to log in to either of my accounts that day.

Why did it take so long to fix such a simple yet devastating issue?

A few months prior, Amazon had laid off a significant number of engineers at AWS whose jobs were specifically to resolve these kinds of problems. Officially, these layoffs had nothing to do with Amazon trying to replace workers with AI. But, again, this is not a task that can be completed by such a small skeleton crew, and this was when Amazon was beginning to enforce AI usage on its AWS engineers and deploying autonomous “agentic” coders. I cannot prove it, but it’s kind of obvious they attempted to replace these engineers with AI, and the AI was unable to fulfil the role due to its lack of intelligence, which caused this catastrophic outage.

You’d think such a public and humiliating failure would teach them a lesson, but here we are, just a few months later, and they have made the same mistake again!

Truth be told, they should have learned this lesson all the way back in early 2024.

I wrote about this hilarious failure in one of my previous articles. Do you remember Amazon’s “Just Walk Out” grocery stores? The idea was that facial-recognition cameras, shelf sensors, and AI would track which items a customer had taken and charge their Amazon account when they left, eliminating the need for a cashier or self-checkout. This innovation was hailed as one of the first cases of AI directly replacing human workers and a way to lower the cost of operating a store. But, in reality, it really wasn’t. A report found that over a thousand remote workers had to be hired to monitor the video feeds and verify 70% of the customers’ purchases, given that the AI was consistently making mistakes. This amount of labour isn’t cheap, even if it is outsourced overseas, and Amazon’s “Just Walk Out” AI became significantly more expensive than simply hiring regular cashier staff. As such, Amazon failed to sell the system to third parties, which resulted in the closure of almost all of these stores and fancy non-AI self-scan systems being used as a replacement. Again, the AI isn’t intelligent and can’t reliably perform simple tasks. This is because it is a statistical machine, meaning it will statistically get things wrong. So, the amount of human oversight needed to correct its simple but potentially devastating mistakes is almost always more work than is saved by implementing the AI.

What is the lesson to learn from this? These systems aren’t intelligent, they can’t even replace basic human intelligence, and they aren’t a productivity tool.

The real question is, after a third attempt, do you think Bezos and his band of ravenous executives have the awareness, empathy, or understanding to learn this lesson? That isn’t a leading question; I genuinely want you to answer it for yourself.

They Should Have Known…

You could argue that this is the clash between theory and practice. AI works in theory, in labs, and in controlled conditions, and the only thing Amazon is doing is ironing out the kinks of transitioning from theory to reality. I will happily point out that Amazon could easily test AI in the real world in a controlled and restricted way, rather than unleashing it basically untested and unrestrained on the bedrock of their business, all because some sweater-vest-wearing business consultant thinks it’s the easiest path to buying a third yacht. But here is the thing: generative AI doesn’t work in theory, and we have known that for a while.

Take the Carnegie Mellon University study, which found that even the best “agentic” AIs completely fail basic tasks 70% of the time, thanks to hallucinations and obviously incorrect responses. Or what about the recent study that found that the best current AIs failed 97.5% of realistic real-world freelancing jobs given to them due to AI hallucinations and total failures? What about the University of Waterloo’s research, which found that even the best generative AI coders only have a 75% accuracy rate when tasked with very basic coding tasks? In other words, even basic AI-generated code doesn’t work a quarter of the time! Or, what about research from Veracode, which found that 45% of AI-generated code contained security flaws? Or the study from Coderabbit, which found that AI-generated code has 70% more bugs than human-written code? All of these factors combined explain why a recent Harvard Business Review report discovered that AI is not boosting productivity, but instead intensifying work. Ultimately, AI is more like a burnout machine than a productivity tool. The time saved by using an AI is greatly overshadowed by the time spent micromanaging the little slop-producing plagiarism monster.

Possibly my favourite example was some research from the University of Melbourne. They found that AI only increases productivity in “low-skill” tasks, such as taking meeting notes or providing customer service. Here, they discovered that AI can help smooth the outputs of workers who may have poor language skills or are learning new tasks. For higher-skilled jobs where accuracy is essential, AIs make errors so frequently that the extensive human oversight required to catch them makes the entire effort less productive than not using AI at all. What’s the problem here? Well, the workers who stand to benefit the most from AI — such as “low-skill workers” — don’t possess the skills or awareness to oversee AI and identify and correct its frequent mistakes. So, even though it “improves productivity”, potentially critical errors go unnoticed, which creates an obligation to micromanage its output, meaning it doesn’t improve overall productivity.

And, yes! It does get worse.

You might argue against this point by claiming we are providing these AIs with more data and compute power, which is causing them to improve, meaning they could have already overcome all of these limitations.

Well, not so fast.

As I have covered before, OpenAI’s latest research paper found that “hallucinations” (where the AI gets things wrong) are a fundamental part of generative AI technology and aren’t going away any time soon. They mathematically proved that adding more training data, ensuring perfect training data, and providing the models with more compute power won’t lower their current hallucination rate. In fact, the paper concluded that there are no viable options for improving overall accuracy.

The body of research is very clear — generative AI is not intelligent; it is not reliable; it can’t replace humans; it can’t be widely used as a productivity tool; and this is how it will remain for a long time.

Amazon should have known this from the get-go, and that is exactly why those “in the know” are pointing and laughing at Huang’s AGI remarks.

Summary

So, well done, Jeff; you have fallen on your own multi-billion-dollar artificial sword. The central product of your empire is flickering out like a broken lightbulb, and you have fired all the talent that could fix it because you wanted to cosplay as Tony Stark and J.A.R.V.I.S. (not to mention that your AI has more in common with the senile Holly from Red Dwarf)… Still, I wonder if any of those employees will be willing to come back to clean up your mess after being so crudely tossed out into the cold. I wonder if, instead of chasing speculative value, techbros will learn their lesson and place more value on real, brilliant human intelligence. All I know is I can hope.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

Whatever Musk has been smoking, I don’t want any, because it seems his last few functioning neurons have fallen out of his nose. Originally, his orbital data idea was laughable. But now that details have emerged about how Musk plans to pull off this grand scheme, it seems even more idiotic than once believed. These are either the drivelings of a power-crazed moron or something more insidious is going on.

Let’s start at the beginning. Despite what Musk claims, orbital AI data centres are not cheaper than terrestrial ones. As I have covered before, it costs roughly nine times more to launch AI data centres into space than to operate them here on Earth, so even during an energy crisis, orbital data centres are significantly more expensive. On top of that, it would cost tens of trillions of dollars to build and deploy the 100 GW of solar arrays to space that Musk has promised, and they would need to be completely replaced every five years or so when the satellites they are attached to deorbit. Oh, and building these satellites on the Moon, as Musk has suggested, doesn’t solve any of these problems and, in fact, will make them worse. Yet, Musk still wants to deploy an orbital AI data centre constellation of one million satellites!

The Satellite

As usual, Musk is pig-headedly carrying on, and in a recent presentation for Terrafab (with more on that in a bit), he showed off the first rough rendering of one of these AI data centre satellites and some details about SpaceX’s plans to get them to orbit.

The “AI Sat Mini”, as shown above, has a solar array size of roughly six metres by 150 metres, for a total area of 900 square metres (using Starship V3 as a scale reference) — which, in orbit, can theoretically provide 216 kW of powerwhen not in Earth’s shadow. This makes total sense, as Musk stated that this satellite can provide 100 kW of AI computing power, and SpaceX recently confirmed that its planned AI satellites would operate at altitudes of 500–2,000 km. At these orbits, the satellite would experience roughly 35 minutes of darkness per orbit, meaning it would require a significantly overpowered solar panel and a sizable battery to remain operational during that time. As such, the power demands, the computational power delivered, and the orbital location all make sense so far. That’s a great start; well done, SpaceX!

Subscribe now

This is going to sound crazy, but the AI Sat Mini is nearly identical to a hypothetical estimate I made months ago for Musk’s orbital data centre satellites, based on the Nvidia GB200 NVL72 rack (read more here). I estimated that an LEO AI satellite based on this 1,360 kg rack would require 880 kg of solar arrays, 471 kg of batteries, 172 kg of radiators for cooling, and 262 kg of radiation shielding to protect the chips inside, for a total mass of 3,145 kg. This Nvidia rack delivers 120 kW of AI compute power, and luckily, all of these components scale with power demand, so we can estimate that Musk’s AI Sat Mini has a mass of roughly 2,620 kg, equivalent to a large car.

That is lighter than you might expect. Let’s optimistically assume these satellites can fold up incredibly small, and Starship’s ability to launch a large number of them is only limited by their mass (rather than volume). Let’s also optimistically assume Starship can reach its promised 100 tons to LEO payload. In that case, Starship can launch 38 of these AI Sat Minis per launch. And, according to my calculations, a fully reusable Starship, if it is ever possible, would cost $70 million (read more here).

So, how much would this satellite cost?

Well, that Nvidia rack, which I based my estimate on, costs $5.9 million, and the cost of the space-rated solar panels, radiators, shielding and construction required to make it a satellite will run into the millions of dollars. So, let’s be insanely generous and say that each AI Sat Mini will cost $8 million a pop.

With these satellite estimates and very optimistic launch assumptions, we can figure out how Musk plans to deploy, maintain, and use this constellation. And guess what? It makes no sense at all.

Launching & Maintaining the Satellites

Okay, so how long would it take to deploy a million of these satellites? And how much would it cost?

At a rate of 38 AI Sat Minis per launch, it would require 26,315 Starship launches to get a million of these satellites into orbit. If SpaceX launched a Starship full of these satellites every single day, it would take over 72 years to reach a million in orbit, as Musk has promised. That many satellites would cost $8 trillion, and that many launches would cost $1.8 trillion for a total cost of $9.8 trillion!

Except these costs aren’t even accurate, because the satellites won’t last that long. To reduce orbital debris, SpaceX places its satellites in an orbit that will decay and deorbit roughly five years after deployment. I can’t find anything to confirm or deny whether the same will be true for these AI satellites.

But, even if Musk plans to keep these AI satellites in orbit perpetually, the chips inside them won’t last that long. Meta found that the AI chips it was using were failing at a rate of 9% annually. These satellites will be operating in far more extreme conditions, so the failure rate could be significantly higher. But let’s be generous and assume the failure rate will be the same.

That means that after 72 years of daily launches and deploying a million AI satellites to orbit, there will only be 154,000 functional satellites in orbit. At this point, equilibrium is met, as the number of AI satellites SpaceX can deploy from its daily launches per year will be the same as the number of satellites lost to failure per year.

So, what does SpaceX have to do to get a million operational AI satellites in orbit in a realistic amount of time?

Well, let’s assume they want to complete the constellation in 15 years. To achieve that, they would need to launch 120,000 satellites per year. Over the 15 years, they would launch 1.8 million satellites, but 800,000 of them would fail (as part of our 9% failure rate), leaving a total operational fleet of one million satellites. This equates to 3,158 Starship launches per year, or nearly nine launches per day. For some context, the current launch rate for Starship is just five per year.

How much would this cost?

So, that is 1.8 million satellites costing $8 million each as part of 47,370 rocket launches costing $70 million each, for a total cost of $17.759 trillion.

Youch, that is a lot. But it gets worse.

In order to keep a million satellites in the constellation, it needs to be maintained. So, each year, SpaceX would have to launch 90,000 AI Sat Minis to replace the roughly 9% of the constellation that failed. That equates to 2,368 Starship launches per year, or 6.4 per day. Using our previous costs ($70 million per launch and $8 million per satellite), that equates to $886 billion per year in maintenance!

Compared to Earth

Back down on Earth (and in reality), it costs $35 billion to build 1 GW of AI compute power and $2.5 billion in annual operational and maintenance costs. How does Musk’s plan compare?

Well, a million of these 100 kW satellites would provide 100 GW of computational power. Let’s keep it simple and say that it will cost Elon roughly $9.8 billion to build and launch a million of these satellites and $886 billion a year to maintain the constellation’s fleet size. That means Musk’s orbital AI data centre would cost $98 billion to construct per GW and have an annual operational and maintenance cost per GW of $8.86 billion per year.

In other words, Musk’s orbital AI data centre is roughly 2.5 times more expensive to build and 3.5 times more expensive to maintain and operate than the current ones we have on Earth!

That is not an expense the AI industry, let alone xAI, could stomach. A recent report found that, even with very optimistic revenue projections, the AI industry will be $800 billion a year short of breaking even by 2030. In other words, instead of drastically increasing expenses, the industry needs to focus on reducing them! Even in the face of an energy crisis or industry regulation, this colossal expense makes this entire idea a non-starter from the get-go.

Oh, and increasing the size of these AI satellites, as Musk has suggested, won’t make any of these problems better. This technology scales very linearly, so it doesn’t matter whether it is one giant AI satellite or loads of these AI Sat Minis. The most a Starship can launch is roughly 3.8 MW of AI compute power, whether it is a single giant satellite or 38 100 kW-capable AI Sat Minis.

The Laughable Scale

Starship’s development began nearly a decade ago when Musk abandoned the idea of a fully reusable Falcon 9. Yet, even after all this time and billions of dollars spent on the project, it is still utterly useless. Its theoretical payload is a fraction of what was promised; it has yet to even make it to orbit; it has yet to become fully reusable; and it only launches at a rate of five per year (with a sizeable failure rate). To give you a sense of just how slow Starship development is progressing, it was supposed to have launched for Mars by now (read more here).

In order to build this million-strong constellation in 15 years, Starship needs to successfully reach orbit, nearly triple its claimed 35-ton payload capacity to 100 tons, become fully reusable at that payload capacity, and increase its launch rate by a factor of 631 from five per year to 3,157 per year.

The idea that Starship will be able to do anything like this in even the mid-future is side-splittingly laughable. They have so far to go before any of this is even remotely possible, and their rate of development is practically glacial. This is beyond fantasy thinking.

Terrafab

All of these factors make Musk’s recent Terrafab announcement even funnier.

Terrafab is a $20 billion joint venture between Musk’s recently merged xAI and SpaceX and Tesla. The idea is to build a fully consolidated semiconductor fabrication facility in Austin, Texas, that contains every stage of chip production within a single facility. As the name suggests, the goal is to produce a terawatt of computing power hardware per year. But this output will be split between two chips: the AI5 chip, which will power Tesla’s FSD and Optimus, will receive 20% of production, and the D3 chip, which will be used for SpaceX’s AI data centre satellites, will receive 80%.

For some sense of scale, Musk himself estimates that the current global AI compute production is 20 GW, meaning this one facility will produce 50 times more AI compute-power hardware than the globe currently does!

There is so much wrong here.

Let’s start with the total lack of coherence.

Musk wants to create a factory that can produce 800 GW of orbital-grade AI data centre chips. Yet, Musk recently said that SpaceX will deploy 100 GW of solar power in orbit per year to power its orbital AI data centres (read more here). The satellite Musk has previewed has 100 kW of computing power, and he wants to deploy a million of them. But that will take decades and cost trillions of dollars, and after all that, a million of these would amount to only 100 GW of computing power.

This doesn’t seem consistent at all. Why is Musk planning to overproduce these chips by a factor of eight compared to how much solar he intends to deploy? Why is the amount of solar he wants to deploy so colossal compared to what is even highly optimistically possible with Starship? The pieces of this puzzle aren’t fitting together.

And then there is the cost-to-production ratio of Terrafab. TSMC Arizona has cost north of $40 billion, yet it currently produces only 10,000 12-inch (300mm) wafers per month, with plans to ramp up to 30,000 in the near future. For some sense of scale here, Nvidia was projected to consume 535,000, or 77% of the wafers used in AI chips in 2025, meaning that the total wafers used for AI in 2025 were somewhere around 690,000.

So, this one facility, built by the industry experts at twice the cost of Terrafab, can only deliver the equivalent of 4.3% of the chips consumed by the AI industry last year. Yet somehow, Terrafab is expected to produce 50 times more chips than the entire AI industry consumed last year?

For this idea to work, Terrafab has to produce 1,163 times more annual compute power than TSMC Arizona, while simultaneously costing half as much to build.

It’s not even like Musk has a good track record with developing in-house silicon. Remember Tesla’s “revolutionary” Dojo supercomputer, which it sunk more than a billion dollars into? It was expected to use chips designed in-house by Tesla that would give the company a gargantuan advantage over the entire AI industry. Well, Tesla cancelled it because the chips sucked, as they were an “evolutionary dead end”.

Terrafab simply won’t accomplish what Musk claims it will; I don’t know how else to phrase it. It’s like me claiming my old VW Golf could reach 400 mph after you watched me crash it into a tree a few moments earlier. Only someone with a truly pickled brain would think it was even remotely possible.

Why?

Actually, since SpaceX has an upcoming IPO, it is more akin to me claiming that if you gave me a billion dollars, my old VW Golf would reach 400 mph. Let’s not forget this is possibly the largest IPO in history.

I think this is the reason why Musk is throwing out these insane, nonsensical ideas and numbers. They aren’t meant to make sense or even be possible — they are meant to make headlines and woo investors.

It doesn’t matter that the current science heavily suggests that AIs simply won’t get better by putting more data and compute power behind them (read more here). It doesn’t matter how utterly preposterous it is to use Starship to launch this many satellites. It doesn’t matter that the economics of building and maintaining AI data centre satellites are orders of magnitude worse than terrestrial ones. It doesn’t matter that Musk hasn’t figured out where he will get enough data to justify this huge AI data centre expansion. It doesn’t matter that Musk’s chip-making track record is horrific, yet he plans to outperform the industry leaders by more than a factor of a thousand. It doesn’t matter that none of the figures he claims as part of this harebrained scheme add up.

The whole thing is messy, incoherent, moronic, and flies in the face of the laws of reality. But it isn’t supposed to make sense. It is, in my opinion, intended to be a barrage of chest-beating to bamboozle FOMO-drenched investors out of every cent they have. After all, it isn’t like Musk hasn’t stooped that low before.

Musk has proven time and time again that we should take everything he says with a heavy pinch of salt. With regards to this upcoming IPO, it might require an ocean’s worth of the stuff.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

The notion that the current AI industry is in a negative-sum game isn’t new. So much money has been dumped into this technology that we have practically bet our economy on its success. So, if it doesn’t succeed, we become the newest residents of recession town. But if this bet does pay off, then millions will lose their jobs, and our economy will collapse, sending us to the same bleak place. It’s shooting ourselves in the foot whilst screaming “innovation!” However, we are starting to see the details of how these scenarios might play out. So, let me explain.

If AI Fails

If you have read any of my work before, you know I am slightly sceptical of AI, so let’s start with the AI failure scenario.

We are now starting to see how AI could fail.

Subscribe now

Quite simply, it isn’t the productivity tool it was sold as. A Harvard Business Review study found that AI didn’t reduce work, but consistently intensified it. The employees “worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day, often without being asked to do so.” In other words, AI isn’t a productivity tool, but a burnout tool. This is supported by the recent report from Oxford Economics (read more here), which found that companies “don’t appear to be replacing workers with AI on a significant scale” and instead suggest that they are actually using the AI layoff narrative to cover up their own shortcomings. Meanwhile, Forrester principal analyst JP Gownder has said that they simply don’t see any boost to productivity from AI in any dataset. Senior US economist Ronnie Walker found something similar, stating that “We still do not find a meaningful relationship between productivity and AI adoption at the economy-wide level.” The only areas he says had productivity gains were “customer support” and “software development tasks.” Note that it isn’t software development, but specifically granular tasks within the job. That doesn’t mean it equates to software development jobs increasing productivity, without decreasing quality, as there could be (and almost certainly are) bottlenecks that AI doesn’t address.

So, even though it might feel like AI makes you more productive on an individual level, that isn’t scaling to an economic-wide impact. Which is a shame because it needs to do that.

That isn’t just my opinion, but that of Nobel laureate Geoffrey Hinton, often called one of the “Godfathers of AI” for his enormous contributions to the artificial neural network technology that powers AI. In an interview with Bloomberg, he was asked whether the eye-watering investments in AI will ever pay off. Hinton replied, “I believe that it can’t,” and elaborated, “I believe that to make money, you’re going to have to replace human labour.” Other reports back this up. As I covered in a previous article, a recent report found that the AI industry will need to generate $2 trillion in annual revenue just to pay for the data centres they plan to build by 2030. This report used very optimistic revenue projections to estimate that the AI industry will be $800 billion short of breaking even by 2030!

To give you an idea of that scale, Google, one of the most monopolistic companies on the planet, made $400 billion in revenue in 2025. So, just to pay for the datacentres AI runs on, the industry will need to pull in more than four times that, and AI can only generate that kind of revenue by replacing huge swaths of the job market.

So, what happens if AI doesn’t replace labour?

Well, the entire industry will be cannibalised by its own debt.

Read more

With vehicle sales crashing faster than Musk’s favourite airship, ‘self-driving cars’ which drive about as well as my myopic Gran after ten Jäger bombs, and humanoid robots that are comfortably outclassed by Disney’s animatronics, Tesla is on one hell of a downward spiral. But, there is still one lifeline stopping Tesla from slipping into oblivion, one it has clung to for more than a decade, emissions credits. It’s hard to explain just how vital this lifeline was to Tesla’s success. They raked in over $11 billion from selling these to other vehicle manufacturers. Tesla simply wouldn’t be the giant it is today, or even here at all, without this colossal cash injection. But with sales and profits sliding, they are relying more and more on these credits to stay profitable. But just when they need it the most, this lifeline is being severed.

Let’s start with how these emissions credits work.

Subscribe now

The EU has a ‘cap and trade’ model. Vehicle manufacturers have to meet a fleet-wide average emission target; if they exceed this target, they face insanely high fines for every gram of carbon emissions per km per vehicle over. But manufacturers can pool their fleets together in order to meet this target, enabling EV makers to offset their carbon savings against automakers who are exceeding the fleet targets. EV makers can charge an awful lot for this service, as the potential fines are exceedingly high. As Tesla has historically dominated the EU EV market, they have also dominated this market too.

The US used to have something similar, but Trump shut it down back in September 2025. And, as China is awash in EVs, Europe is now the largest market for these emissions credits.

Okay, so why does this matter so much to Tesla?

Well, in 2025, thanks to nosediving sales, Tesla’s net profit dropped 46% compared to 2024 to just $3.79 billion. In 2025, Tesla also sold $1.99 billion in emissions credits globally. Since these credits are basically 100% profit, that means 52% of Tesla’s net profit last year came from emissions credits. That’s right, Tesla currently earns more from regulatory credits than from selling their cars. With their high-revenue-generating software services, like FSD, being a worse sales flop than the Cybertruck, its profit margin continuing to fall, and its vehicle sales falling off a cliff, these regulatory credits were set to become an even bigger share of Tesla’s profit, and critical in stopping them from hitting unprofitability in the near future.

But the emissions credit market is drying up rapidly.

Tesla made $2.76 billion from emissions credits in 2024, so their profit from credits shrank by 28% in 2025 alone. But they are set to fall even further.

It is estimated that roughly half of Tesla’s 2025 emissions credit revenue came from the US. But Trump closed that market back in September, so straight away, that is a billion dollars of potential profit wiped away.

But things aren’t looking much better in Europe. In 2025, Tesla’s European emissions pool included Toyota, Stellantis, Subaru, Suzuki, Ford, Mazda and Honda. But Toyota, Stellantis and Subaru have now left the pool. This is a major blow as Toyota and Stellantis were reportedly some of the largest purchasers of Tesla’s emissions credits in Europe.

Why leave? Well, the EU is relaxing emissions targets and assessing manufacturer average fleet emissions from 2025 to 2027, rather than on a yearly basis. So, if these manufacturers know they have a ton of hybrid and electric cars coming, they simply don’t require the credits. But in Toyota and Stellantis’ case, both have heavily transitioned to hybrid and fully electric cars, meaning they likely hit the emissions targets themselves already, and so have no need for Tesla’s credits.

We know Tesla generated roughly $1 billion in emission credits in Europe in 2025. What we don’t know is how much each manufacturer paid into that. But, if we assume they all spent the same, then we can estimate that losing Toyota, Stellantis and Subaru will shrink this revenue by some 42%. Keep in mind, this is likely an underestimate, as we know Toyota and Stellantis were likely spending more than the others.

With all of this, we can make a rough guess that Tesla will only make $600 million from emissions credits this year, or a whopping 70% drop from 2025!

That means that even if Tesla’s core business doesn’t shrink in 2026, this alone could take its profit down 36% to just $2.4 billion!

But, Tesla’s core business is shrinking. Tesla’s global sales dropped by 8% in 2025 compared to 2024, and recent sales figures suggest that trend is only intensifying. Especially as Tesla’s limited and ageing line up is being swamped by more affordable, more capable, and more varied EVs entering the market.

Let’s also not forget that Tesla’s Optimus and Robotaxi projects are going nowhere, with even Tesla insiders calling them dead ends (read more here, and here). As such, as Musk continues to push them, they could drag Tesla’s balance sheet way down.

As such, my back-of-the-envelope prediction is that Tesla’s 2026 profits could easily be less than half that of 2025, if not considerably less.

In other words, Tesla is about to take a huge step towards becoming unprofitable.

That isn’t necessarily a problem; companies can run at a loss as long as they are growing or developing new revenue streams. But arguably, that isn’t what is happening at Tesla. Again, even the Tesla executives behind Tesla’s Robotaxis and Optimus robot don’t think they will lead to revenue (read more here and here). On top of that, much of what is propelling Tesla’s share price so high is the hype that profit is going to explode, so when that doesn’t happen, it could severely damage the share value.

Tesla losing its emission credit revenue isn’t going to kill it. But it is yet another major step on its downward spiral. Companies as large as Tesla rarely die in one fell swoop; they almost always die by a thousand cuts. Tesla doesn’t have to die; this isn’t a terminal tail-spin, but the blows are starting to add up, and if Musk doesn’t do something to get solid business fundamentals under Tesla soon, its fate might be sealed.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

Tesla had the biggest head start in the EV market. The Model S launched in 2012, and it took six years for legacy automakers to offer anything that could even try to compete with it. Even then, Tesla’s charging network was light-years ahead of what anyone else was offering. Likewise, when the Model 3 launched in 2017, it had no genuine competition for over three years. At the turn of the decade, it looked like Tesla would extend this lead with new models and the ‘revolutionary’ 4680 battery. But that simply hasn’t happened, and now Tesla’s sales are crashing as every single one of their competitors now offers comparable or slightly better EV options. But the leapfrog moment has just happened. BYD have just annouced their 1,500 kW-capable Blade Battery 2.0 and the stupidly affordable EV it is going into. Now it is Tesla who is playing catch-up. So, what about this puts Tesla on the back foot? And why did Musk fumble the bag?

Let’s start with this new EV from BYD, and what makes it so damn special. In China, it is called the Seal 07 and will be equipped with BYD’s second-generation LFP Blade Battery, specifically the “short blade” variant. This new battery retains the same stupidly high safety standards and long lifespan of the previous version, but it charges WAY faster! The 69 kWh pack in the Seal 07 has a peak charging rate of 1,500 kW, enabling it to charge from 10% to 70% in just 5 minutes, and from 10% to 97% in just 10 minutes.

Subscribe now

In other words, this car can do a normal rapid charge in the same time it takes to fill the tank on a combustion vehicle. This is the kind of tech we thought would be more than a decade away just a few years ago.

But BYD hasn’t put this in some silly, expensive flagship halo car. No, the Seal 07 is a mass-market Model 3 competitor. I have compared the two in the table below:

*Estimated from the price difference between the current BYD seal in China and the UK

**Estimate based on the ratio of the Model 3 CLTC to WLTP

So, the BYD Seal 07 is RMB 90,500 (£9,700 / $13,000) cheaper than the equivalent Model 3 in China, whilst charging more than five times faster! Yes, the Model 3 has a slightly longer range and accelerates slightly faster, but is that enough to justify being 54% more expensive?! It’s not like this version of the Model 3 is better spec’d either, as the Seal 07 will have BYD’s “God’s Eye” system, which has been touted as more advanced and safer than Tesla’s FSD.

There is some important context here. The Chinese EV market is in a price war, and BYD has had to slash prices to stay competitive; even then, its sales have fallen sharply in recent months. So, you might think the price of the Seal 07 isn’t directly comparable to the Tesla.

But, don’t be so sure. For one, Tesla has also had to drop its prices as its sales have collapsed across the globe. But the main driver of BYD’s falling sales and discounts was that other Chinese EV makers, like Xiaomi, offered better-equipped, more capable EVs at lower prices than BYD. However, with the launch of the Blade Battery 2.0, that seems set to rapidly change, as it puts BYD back at the forefront of the industry. Plus, BYD could sell this car in Europe for £10,000 more (or equivalent) and still be competitive with Tesla’s offering. So, the prices are more comparable than you might think, as they strongly suggest BYD has a gargantuan cost-of-manufacturing advantage that could easily become a profit-margin advantage (if it isn’t already).

There is a really important part of the Seal 07 that I haven’t talked about yet. You see, it is all well and good having an EV that can charge at 1,500 kW, but unless there are thousands of chargers capable of delivering that power. Luckily, this is exactly what BYD are doing. They developed their new 1,500 kW “Flash Chargers” to work with their second-generation Blade Battery and unlock these charge times, and they have been proven to deliver BYD’s claimed charge rates in the real world. On top of that, they have already deployed 4,200 Flash Charger stations in China and plan to build 20,000 more by the end of this year. But this rollout isn’t limited to China, as BYD are planning to build thousands of them in Europe, the Middle East, and across Asia.

In other words, if you are in a region where you can buy the new Seal 07, there will already be a ton of 1,500 kW-capable chargers available, and if not, there will be very soon.

So, why? Why has BYD been able to take such a catastrophically huge leap forward that it makes Tesla’s technology and pricing look outdated?

In a word, hubris. Basically, Musk fumbled the bag by taking Tesla down a dead-end.

Like BYD, Tesla recognised the cost and technological advantages of developing and building their batteries in-house. But they went down very different routes. While BYD optimised proven, cost-effective technology, Musk bet on unproven, risky technology.

BYD chose to optimise the proven, cost-effective LFP cell chemistry and prismatic cell formats to create the Blade Battery. Musk opted to bet it all on unproven dry-coating technology.

Now, making big bets isn’t the end of the world if you understand what you are dealing with, but Musk simply doesn’t understand batteries. That isn’t my opinion, but the CEO of the largest battery maker in the world, CATL, who back in 2024 said to Musk’s face that Tesla’s 4680 battery “is going to fail and never be successful” because “He doesn’t know how to make a battery. It’s about electrochemistry.”

But Musk thought he knew better than the experts and wasted precious years and billions of dollars in R&D to make a battery that is already obselete, currently only powers the Cybertruck, and, in fact, it looks like the 4680 as we know it might soon end production.

Meanwhile, BYD invested in genuine research with leaders in the field for six years and now has a battery with charging speeds, safety standards, and, almost certainly, cost advantages that no one else has. They took ego, hubris and dogma out of the equation, did the hard graft, and have ended up leapfrogging the entire industry in one fell swoop.

Tesla could have done the same. But Musk’s ego got in the way. They wasted their lead, and now they are in catch-up.

Especially as BYD is not some standalone distant player. They have deep strategic partnerships with many big players. Toyota has created a 50/50 joint venture with BYD to use their technology in their EVs, so the Seal 07 could easily become the next Camry. Ford is in the final stages of securing a battery supply deal with BYD. Stellantis (which owns the likes of Jeep, Ram, Citroen and Alfa Romeo) is also seemingly moving to use BYD’s battery technology. Basically, all the big players in the automotive world have some close connection to BYD to use their technology.

In other words, it isn’t just BYD that is leapfrogging Tesla, as almost every major player in the EV market will soon use BYD’s technology in some way.

Tesla could do this, too. After all, they have a longstanding relationship with BYD to use their first-generation Blade Battery in their European Model Y (kind of proving that BYD’s battery technology has always been better than Tesla’s 4680). But unlike all the other automakers, Tesla’s value in based on the idea that their in-house tech would be the leader. So, adopting BYD’s tech would likely damage the value of the company.

BYD is a beautiful reminder of what Tesla could have been if it had kept grounded. If it ensured its direction was set by reason, reality, and science, not by a greedy billionaire oligarch snake-oil salesman. They show just how much Musk has slept on his laurels. As such, the next few years in the EV market are going to be interesting to say the least, as we could be about to watch Goliath get slain.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

The memification of Musk’s constant failure to deliver even a fraction of his self-driving promises has spiralled from funny to downright painful. I am convinced that this is what broke the public’s golden-boy perception of Musk more than anything else, and I’m including that ‘wave to the crowd’ moment. Musk’s questionable politics were tolerated for years until everyone realised he is full of s**t. But, somehow, these ‘untruths’ are getting worse, and I think I might know why.

For a while now, Musk has been saying that Tesla robotaxis are imminently coming to California, as soon as regulators give them permission. Back in October 2025, Musk said that robotaxis would be available in California “in a few months” and that they were “waiting on regulatory approval in California.” However, Reuters recently found that Tesla logged zero miles of autonomous test-driving on California roads in 2025. In fact, this marks the sixth year in a row Tesla has failed to clock any autonomous test miles in the state since getting a permit to do so.

Subscribe now

Why does this matter? Because it indicates that regulators are waiting on Tesla, not the other way around, as Musk insinuates.

Tesla currently holds an ‘entry-level’ California DMV permit, allowing it to test autonomous vehicles on California roads with a safety driver in the driver’s seat. But, to even apply for a permit to test fully autonomous robotaxis, Tesla needs to log 50,000 miles of autonomous driving on public roads in California with a safety driver. Once it has obtained this next permit, it needs to undergo even more testing before being allowed to offer autonomous ride-hailing to the public.

However, that is likely an unrealistic minimum, as Reuters pointed out, Waymo logged 13 million miles of testing hours before it launched its driverless taxi service in California. Considering Waymo has historically been far safer than Tesla’s Robotaxi (read more here), it is fairly safe to assume Tesla will likely need to log even more testing with the California DMV to prove its safe enough to get a similar permit.

So, why hasn’t Tesla started logging these miles?

After all, it could take Tesla an awfully long time to do so. Their current Californian test permit doesn’t allow them to offer supervised robotaxis service to the public, as they are currently doing in Austin, Texas, and limits them to offering rides only to employees or select individuals. That puts a serious bottleneck on how quickly Tesla can log these miles, and potentially makes logging them much more expensive.

However, even in Texas, Tesla isn’t logging all that many miles. According to Electrek, from June 2025 to mid-January 2026, Tesla’s Austin robotaxis have logged 800,000 miles, or equivalent to 40,000 logged hours (assuming an average speed of 20 mph). At that rate (6,666 logged hours per month), it would take Tesla 162 and a half years to log as many supervised autonomous test hours as Waymo did in California before launching their driverless service.

But, I can think of several reasons why Tesla hasn’t started logging miles in California yet, and they all revolve around Tesla’s Robotaxi’s crash rate.

You see, over those 800,000 miles, Tesla’s Austin robotaxis logged 14 crashes with the NHTSA. In other words, even with a supervisor, Tesla robotaxis crash on average once every 57,000 miles. To put that into perspective, Tesla itself estimates that the average American has a minor collision every 229,000 miles. That means Tesla’s robotaxis crash at a rate four times that of a human driver, even when there is someone in the car to correct its mistakes, so the actual crash rate if these robotaxis were allowed to operate autonomously would likely be far higher.

That could make logging these 50,000 miles not just a futile Sisyphean task, but also deeply damaging to Tesla’s narrative.

Let me explain.

Tesla’s robotaxi crash rate is horrific. For some context, Waymo got its full commercial licence to offer driverless robotaxi services in California in early 2024. In 2024, Waymo cumulatively logged 114 million miles driven and reported 462 accidents to the NHTSA, for an average of 1 accident every 247,000 miles. In other words, just after Waymo received its driverless permit, it operated at a crash rate more than four times lower than Tesla currently posts with a safety driver! But, it is worse than that. In 2024, Waymo had a disengagement rate (how often a safety driver takes control, or the system aborts, to avoid an accident) of once every 9,793 miles. For comparison, on the FSD tracker(which collects user-reported Tesla FSD data, making it likely a cherry-picked and too optimistic source of data), the latest and greatest system from Tesla had a disengagement rate of once every 1,029 miles, or nearly ten times worse than Waymo had more than a year ago!

If you want to know why Tesla’s self-driving tech is so far behind Waymo, read my previous article here.

All of this suggests the bar to pass safety requirements to offer driverless ride-hailing services in California is far, far higher than Tesla is currently at. As such, logging test miles might be pointless for Tesla, as it will only prove to regulators that they aren’t safe enough to advance to the next-level permit.

But, Tesla doesn’t just have to log 50,000 test miles in California; it also has to report a crash, with detailed context, within 10 days of an incident, and detailed disengagement data annually over these test miles. Moreover, the California DMV posts this data and the reports to public databases!

So, not only would logging test miles in California likely not get them closer to obtaining a commercial autonomous taxi licence, but it would also risk exposing the gargantuan shortcomings of Tesla’s autonomous technology. That could be a major threat to Tesla’s stock price, as the notion that it is a leader in autonomous technology is the only thing keeping its stock price so ludicrously high.

Compare this to Texas, where Tesla can easily obtain a permit to run a small autonomous ride-hailing service with safety drivers with no requirement to report incidents to the Texas DMV. So, they can make it ‘look like’ they are on par with Waymo, all while obfuscating their shortcomings. In fact, the only body Tesla has to report incidents in Texas to is the NHTSA, and there they can redact details about it, hiding how bad or how responsible Tesla’s technology is.

Oh, and those rumours of driverless Tesla robotaxis in Austin? They seem to be non-existent. So, that doesn’t change anything.

I suspect that Tesla isn’t logging testing miles in California because they know they aren’t safe enough to get the next-level permit, and to keep the true incident rate of their robotaxi under wraps to protect the myth that is holding Tesla stock prices so high. I just wonder how long Tesla can keep this charade up, and how long investors are willing to ignore reality.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

With it all kicking off among Anthropic, the Pentagon, and OpenAI, I think it is about time I explained what the real value of AI is, as this is the perfect lens for understanding what the hell is actually going on with AI right now. You see, AI is a Trojan horse. It is an exploitative oligarchy parading as a productivity tool. So stick with me while we fall down this terrible rabbit hole.

Let’s start with the fact that AI doesn’t broadly improve productivity, isn’t good enough for automation (read more here), and has been repeatedly shown to damage workers’ skills (read more here). Indeed, despite large-scale AI adoption, there hasn’t been a surge in productivity.

But that doesn’t make any sense. We have been told that the reason AI is so valuable and why corporations are pouring billions of dollars into AI is that it will be the next industrial revolution, unleash new levels of automation, boost productivity and deliver an economic miracle. However, we can see that this simply isn’t true. So, what is the real reason AI is perceived as so valuable? What actual beneficial utility do they actually deliver to their owners, investors and users?

Subscribe now

Well, if you look at what AI is actually good at, it becomes quite obvious. It is enabling and empowering a new technocratic oligarchy. That’s right, your favourite chatbot is a perfect tool of authoritarianism masquerading as a slightly more polished version of Microsoft’s Clippy.

How? Well, let’s start with this spat Anthropic has had with the Pentagon because the story is not as simple as it has been portrayed.

The AI Panopticon

The Panopticon was a conceptually and ethically flawed idea proposed by Jeremy Bentham in the 18th century to make prisons way more efficient through perceived surveillance and self-policing. A panopticon has two parts: a central observation tower and a ring of cells facing it. The observation tower and cells are set up, so guards can see every bit of the cells and the inmates, but inmates can’t see the guards. This way, the guards won’t know if they are being watched, creating a feeling of constant surveillance. Bentham hypothesised that this would make inmates self-police their own behaviour and, in turn, allow a single guard to keep the prison in check. In short, the watcher sees everything, the watched see nothing, and so the watched behave themselves.

I will give you a moment to let your subconscious figure out why this is a truly awful idea. But the way modern generative AI currently functions at the corporate and governmental levels effectively serves as a digital version of the panopticon, as it is perfect for unseen surveillance.

LLMs like ChatGPT and Claude deeply erode digital privacy. They can accurately identify locations from a single simple photo, effectively geotagging any video or photo ever taken, making them the perfect tool for stalking. Both ChatGPT and Claude have remarkably accurate facial recognition abilities, and while there are safeguards in place to stop this ability from being used for surveillance, these can be painfully easy to overcome. Considering we live in a world where our phones, cars, doorbells and laptops have cameras strapped to them, it is easy to see how these AIs can be used to surveil everybody.

Indeed, Palantir, which functions as a private government surveillance service masquerading as a data analytics company, has its own AI that does just this, and can recognise and track faces and locations. We also know that they use external LLM AIs to integrate, analyse and visualise massive, disparate datasets for defence and intelligence agencies, enabling them to make rapid decisions. In other words, these AIs are used to connect every part of your digital footprint and hand it to the authorities. We also know that Palantir uses both Claude and ChatGPT to do this. In our digital world, that is effectively 24/7 surveillance, and it is already being used against immigrants, as ICE uses all of these services.

But corporations can do the same thing to their employees with LLM chatbots like Claude and ChatGPT.

These bots can be trained on the vast amount of internal data and communications to find anything from ‘poor performers’ to identifying and squashing unionisation efforts. Indeed, many Bossware tools now use integrated AI LLMs to ‘interpret’ the tone and context of worker communications and behaviour for risk management, productivity assessment, and even behaviour prediction. Enterprise-level monitoring goes one step further. If an organisation pays for the enterprise version of an LLM (ChatGPT and Claude have this option), then the administrator can effectively see every interaction you have with the AI. So, by having employees work with these AIs, it enables them to be surveilled in every action they take at work. Burger King has somehow taken this one step further and integrated ChatGPT into their workers’ headsets to ‘help them with meal prep’ but also to monitor their behaviour. You could call this micromanagement, but it is just an AI panopticon.

This is why I see the Pentagon/Anthropic/OpenAI debacle as total bullshit. Don’t get me wrong, I’m glad Anthropic didn’t want their AIs to be used to make “Mass domestic surveillance” and “Fully autonomous weapons.” I’m also damn glad that the backlash to OpenAI jumping in to replace Anthropic has caused such a colossal backlash that ChatGPT uninstallations have skyrocketed by nearly 300%! But Anthropic is no better than OpenAI in the grand scheme of things. Let’s not forget that Claude was used in the recent strikes in Iran, and they are both part of our new-age AI panopticon, which enables corporations and governments to mass-surveil us at unprecedented levels. The sceptic in me thinks this was all a marketing ploy by Anthropic, and that I should be telling everyone that switching to Claude is not the ethical alternative they think it is. And as if to prove my point, while writing this, Anthropic is back in talks with the Pentagon…

So, what is so bad about all of this? If you have nothing to hide, why care? Well, because panopticons don’t fucking work!

Bentham assumed the threat of being constantly watched would create order, but he was wrong, as all it actually does is create undue anxiety, isolation, distrust and feelings of vulnerability. This makes sense when you think about it for even a second. But here is the thing: being anxious puts you in fight or flight, which can make you more aggressive and less orderly. So, not only is this ethically abhorrent, but it doesn’t lead to the behavioural changes it intended to create (see here and here)!

This anxiety is not unfounded. The panopticon famously has a “who watches the watchers?” problem. All discipline is concentrated in a select few, and their interpretation, sadly, observation and objective realities are two very different things, and the watched do not know how their actions will be perceived, no matter how well-intentioned they are. Taylor Lorenz has a great video on this, go watch it here.

As a side note, a panopticon only functions if it has a credible threat to those under it. So, there needs to be constant visible enforcement, otherwise those under it lose the fear of the watcher, and the pressure to conform disappears. So, it doesn’t necessarily reduce the amount of enforcement needed to control a population, and in fact might motivate additional, chaotic, indiscriminate ‘punishment’ to grow this necessary fear and anxiety. As such, I suspect ICE’s very public egregious acts and the current colossal number of corporate mass layoffs are, in part, to do just this: instil fear in the AI watcher to make the AI panopticon work.

But really, this is all a panopticon is good for, making those under it anxious and therefore easy to manipulate, while concentrating power in the watcher. All that has changed with this modern AI panopticon is that it places the oligarchs who own and operate these AI systems in the position of power, making them judge, jury, and executioner. Literally in some cases.

In a society where the capitalist elites are turning more to monopolistic extraction of power and wealth from the 99%, you can see why AI and its panopticon are seen as so valuable, and why AI’s inability to boost productivity, or even generate a profit, isn’t a problem. It is about manipulation and domination by an elite.

AI Elitism

But this is just the beginning of the rabbit hole because generative AI, particularly the modern LLM variety, inherently enables an entirely new form of pervasive elitism.

Elitism doesn’t mean what many people think it means. It is the belief that a select few ‘elites’ should have greater power than the rest. As such, elitism is more akin to diet-fascism, as it almost always has to turn to baselessly dehumanising others to create a false meritocracy *points at the Trump administration.*

AI enables this in a far more foundational way than we have seen before.

Remember when Grok suddenly started glazing Musk, and saying he was the best at absolutely everything? Even if it wasn’t Musk who made Grok do this, this shows just how much control those who own or operate AI have over their product. In other words, if they want their AI to make certain decisions, they can make it do just that.

But these LLMs are being used by governments, businesses, organisations and individuals to ‘automate’ tasks, jobs and decisions. While we know this likely doesn’t improve their productivity, it does place a significant amount of key decision-making power in the hands of those who control the AI. In other words, the entire structure of AI LLMs is inherently deeply elitist. In other words, by forcing data and decision-making through AI, it concentrates power in the elitist class.

This is why the revelations that AI LLMs cause burnout and significant cognitive decline are not a problem. It undermines and overwhelms the collective workforce’s decision-making and inherent power within the system, and hands it to the tech oligarchs and corporate wielders of AI.

Why is this bad? Well, elitism is a fundamentally broken ideology. Hubris syndrome means that the more power you wield, the worse decisions you make. The empathy gap and associated affluenza mean that the more wealthy and powerful you are, the less empathetic you become. So elitism creates a system led by those who don’t care about the majority and make awful decisions, *points to the Trump administration again.* In other words, elitism will always create rampant and wildly detrimental inequality.

Again, it is easy to see why tech billionaires and many C-suite executives see this as valuable, as it concentrates power in their hands. But in reality, it does much more than that.

Flattening Human Value

Remember how I said elitism is diet-fascism? Well, AI LLMs are inherently wildly dehumanising, which is a critical step in solidifying elitism, oligarchy and fascism.

LLMs are dehumanising in so many ways that it is challenging to count. It dehumanises those whose work it is trained on by failing to compensate them. It dehumanises workers through ‘micromanagement’ and crushing their creative value. Like a regular panopticon, our AI panopticon reduces our internal moral agency to fear-based conformity and paranoia. This not only reduces and devalues our perception of our own humanity, but legitimates violence against others, as our own understanding of morals is entirely fear-based, so dehumanising everyone else. AI isolates us by making us interact with it, rather than other humans, eroding our sense of society and reducing our empathy. Through AI surveillance and ‘automation’, it flattens the human experience and value to just numbers in the eyes of intelligence, governments, and corporations, inherently dehumanising those under their power. This flattening and distance dramatically reduce government and corporate social responsibility by deindividualisation, enabling wildly exploitative and violating actions against those not in the elitist class.

I could go on and on. But in short, should we really be surprised that a technology that uses maths to pull off a statistical parlour trick to badly emulate humans is dehumanising?!

Again, it is easy to see why certain powerful governments and corporations see immense value in AI’s rampant dehumanisation. It casts off any social accountability and makes those under its power less empathetic toward those around them, or even toward themselves. It basically enables the violation and exploitation of the masses to empower and enrich the elite that oligarchy and fascism need.

If I need to explain why that is bad, you need your head examined.

The AI Oligarchy

There are countless other ways AI empowers an oligarchy, such as being the perfect tool for techno-feudalism. But I feel I have made my point, and this article is already way too long.

We can now see why AI LLMs are the perfect tool for the modern oligarchy that needs power to retain and grow its assets. They can force it into everything under the guise of productivity, only for it to massively empower them.

That is why those backing AI don’t care that it isn’t increasing productivity because that isn’t the point. Likewise, they don’t care that the entire industry is increasingly far from profitability, and keeping their cash bonfire going is a small price to pay for solidifying and empowering their oligarchic domination, and turning the rest of us into modern-day serfs. That is why they have been happy to inflate the AI bubble to an economy-ruining size.

Summary

The current generative AI industry has nothing to do with productivity. It is about capital turning to authoritarianism to empower and enrich itself while protecting itself from democratic accountability. It is a perfect example of unregulated capitalism devolving into fascistic behaviours because ‘maximising capital’ almost always equates to crushing the human experience. This is why what is going on in AI is so closely linked to the imperialist crap going on right now. The current state of the AI industry is a symptom of our political landscape. So, fighting AI’s crushing march is also helping to fight against genocide, against needless world wars, for democracy, for empathy, for the human experience.

So, if you are leaving OpenAI because it was so desperate to lick Hegseth’s boots, maybe consider leaving generative AI for good.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

OpenAI recently announced it is raising $110 billion, including $30bn from SoftBank, $30bn from Nvidia, and $50bn from Amazon. That is more than double the $40 billion they raised last year in the largest private tech deal on record. It also valued the company at an astonishing $840 billion! But, OpenAI has also seemingly reduced its spending targets, with investors being told that it is targeting around $600 billion in total compute spend by 2030. That is significantly less than half its previous $1.4 trillion commitment. So, with more money coming in and less going out, is this the turning point for OpenAI? Can this turn OpenAI from a cash black hole into something even remotely sustainable? In short, no, not at all. $110 billion isn’t even going to touch the sides. Let me explain why.

Subscribe now

Let’s start with the fact that, by OpenAI’s own estimates, this isn’t really enough. Back in September 2025, OpenAI dramatically increased its projected cash burn, stating that it expects to rack up losses of $115 billion between now and 2029. That is $80 billion, or 44% higher than their previous projections! It also means this record-breaking investment will only keep the fires going for a few years.

But, there is something telling here. OpenAI posted a loss of $5 billion in 2024 and $8 billion in 2025, meaning its losses are growing 60% each year. If you assume losses keep growing at the same rate of 60% per year, then between 2026 and 2029, OpenAI would rack up $118 billion of losses, which is almost identical to OpenAI’s own estimates.

Why is that telling? Well, the entire AI industry operates on the idea that scaling AI will eventually lead to sustainability and profitability. In other words, the industry assumes AI revenue will scale faster than the cost to develop and run AI models, so eventually, at a certain scale, they will break even. This argument of scale to profitability is foundational to the current frenzied AI investment drive.

But OpenAI’s assumption that their losses will keep growing at the same rate, rather than at a slower rate as before, strongly implies they know the scale-to-profit idea simply isn’t real. They know that scaling this way will actually push them further away from break-even.

In other words, once this $110 billion is used up, they will need even more to keep the ash bonfire burning. But, more on that in a second.

Does reducing spending to $600 billion by 2030 make any of this better? At the very least, does it mean this $110 billion will last longer?

Take a wild guess!

OpenAI’s $1.4 trillion in commitments were for roughly 26 GW of compute power, suggesting this $600 billion is for roughly 11 GW of compute power by 2030. How much would that cost?

Well, as I discussed in my previous article, IBM CEO Arvind Krishna has stated that a single GW of AI-capable data centres costs $80 billion to build. However, we also know that a GW of AI data centres costs around $1.3 billion in energy costs each year and has a realistic operational lifespan of three years. The annual cost of a single GW of AI compute power (including the annual spread build cost and energy cost) is $27.97 billion.

OpenAI ended 2025 with 1.9 GW of compute, which will cost around $53 billion a year to operate! Let’s assume OpenAI adds 2.75 GW of compute power each year between now and 2030, for a total of 11 GW. That would add another $76 billion a year to their operational costs! As such, by the end of 2027, OpenAI would have blown through all of this $110 billion investment.

But it would also mean that by the end of 2028, OpenAI would have a $205 billion compute bill to pay. But, OpenAI itself only predicts that it will have $100 billion in revenue in 2028, and even that is highly optimistic. In other words, in just two years, OpenAI will have to raise this colossal amount of cash yet again just to avoid bankruptcy.

Now, these are very rough estimates. For example, they don’t consider how data centre operators soak up some of the data centre losses, or how some of OpenAI’s partners give it discounted compute. So this isn’t 100% accurate, but it gives a sense of the size of the hole OpenAI has dug itself, and how little this $110 billion will help it.

But, speaking of OpenAI’s “partners”, this $110 billion “investment” round has a huge cannibalistic problem to it.

You see, Amazon didn’t invest $50 billion into OpenAI. It invested $15 billion, with another $35 billion “in the coming months when certain conditions are met.” But this deal also locks OpenAI to buying 2 GW of Amazon-owned compute, which, as we have seen, will likely cost more than $50 billion. That sounds like a bit of a sour deal for OpenAI. It’s not so much of an investment as an extraction. Likewise, Nvidia’s investment in OpenAI is explicitly intended to lease chips from Nvidia.

So, OpenAI is selling off its share, its equity, to lease the very cloud computing it needs to operate. That is like chopping off and selling your body parts to buy food. Sure, it might work for a little while, but eventually, you won’t have any body parts left to chop off. OpenAI didn’t want to do it this way. It wanted to build and operate its own data centres, but struggled to raise the funds needed to do so. The only way it could raise enough was to enter these arguably predatory agreements.

And that is the huge issue here. The fact that OpenAI has to sell off such a large portion of itself to lease cloud computing from these very “investors” is terrifying. What kind of circular payday loan economics is that? It is not only a huge red flag that this is all a giant bubble, but it also leaves OpenAI high and dry. At some point, OpenAI won’t have enough equity to keep its cash bonfire going, and it will burn to the ground.

So, no. $110 billion is nowhere near enough, and lowering their projected expenditure to $600 billion by 2030 won’t make a difference. What’s more, how OpenAI raised this cash is a sign of just how desperate they are for funds, and just how much they have backed themselves into a corner. In my opinion, you would have to be blind to think any of this is “good news” for OpenAI.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share

AI is a new economic revolution! It will separate economic growth from labour, enable unlimited growth, and lead to a new age of prosperity. This narrative has been used to justify a truly gargantuan investment in the technology. Last year, Big Tech spent $400 billion on AI. That is enough to eradicate world hunger for a decade! But this year, they are poised to spend $700 billion on AI. Indeed, economists have said these investments have boosted the US economy. Deutsche Bank found that almost all of US GDP growth was from AI investment, and without it, the country would be in a functional recession. Harvard economics professor Jason Furman has backed this up, saying that AI drive 92% of US GDP growth in the first half of 2025. So, is AI the economic miracle it promised to be? Well, no. Economists are starting to actually dig a little deeper past the AI propaganda and have found that this biblical investment has driven ‘basically zero’ economic growth.

This comes from Goldman Sachs Chief Economist Jan Hatzius. In an interview with the Atlantic Council, Hatzius said that AI spending had ‘basically zero’ contribution to the US GDP growth in 2025. He even said “We don’t actually view AI investment as strongly growth positive” and that “there’s a lot of misreporting, actually, of the impact AI investment had on U.S. GDP growth in 2025, and it’s much smaller than is often perceived.”

Subscribe now

So, why does Hatzius have such a different opinion from many of his peers?

Well, unlike them, he actually looks at where all that cash is going. You see, all these billions of dollars are mostly being used to import computer chips. So, Hatzius argues that these investments are adding to the Korean and Taiwanese GDP, not the US GDP. So, in fact, these investments are draining the US of capital it really should be using to grow its own economy!

But Hatzius had a, quite frankly, odd omission. You see, AI is meant to increase productivity through augmentation and automation, right? And, increasing productivity creates GDP growth. So, surely, if AI investment itself isn’t increasing GDP, the AIs created by this investment, and deployed into the US economy, will be driving economic growth. So, why didn’t Hatzius take this into account?

Well, because AI isn’t increasing productivity, and so its deployment isn’t increasing GDP.

Analysts have known this for quite a while.

Goldman Sachs itself estimates that AI will increase US productivity by only 15%. But even that might be wildly optimistic. Back in 2024, ING estimated that AI would only increase productivity by 1%. For some context, the computer and internet revolution accounted for roughly 45% of the US’s productivity gains since 1980, and from 1980 to 2019, US productivity increased by roughly 60%, meaning PCs and the web drove a 27% productivity increase. Even during the peak IT craze in 2000, the industry’s annual capex was $164.2 billion, or $310 billion in today’s money. So, even optimistically, AI is costing more than twice what the IT revolution did, and will only deliver half the productivity gains. More realistically, it is set to deliver less than 4% the productivity gains.

But I think ING’s predictions are actually too optimistic, as recent data makes it clear AI is not at all the productivity tool we once thought it was.

A study from Carnegie Mellon University found that even the best “agentic” AIs completely fail basic tasks 70% of the time. A recent study found that the best current AIs failed 97.5% of realistic real-world freelancing jobs given to them. METR found that generative AI slowed down experienced developers by 19%, as they have to spend more time correcting mistakes the AI had made than the time the AI saved them. All of this explains why a recent Harvard Business Review report found that AI is not boosting productivity, but instead intensifying work. In short, AI is more like a burnout machine than a productivity tool.

We can see these findings reflected at the business level too. Famously, MIT found that 95% of corporate AI pilots fail. This is backed up by PwC, who found that only 12% of businesses using AI saw it reduce costs and increase profit, BCG, who found that only 5% of companies that deployed AI saw value from it, and Forrester Research, who found that 15% of their corporate survey correspondents reported an increase in profit margins from AI over the past year.

As such, it looks like businesses are pulling back from AI use, rather than rapid adoption. The Economist found that the use of AI by large US corporations is actually shrinking, not growing, and S&P Global Market Intelligence found that the cancellation rate of corporate AI programs skyrocketed from 17% in 2024 to 42% in 2025.

Again, we can see all of this reflected in economic data. Despite what mainstream media might have told you, there are currently zero layoffs caused by AI automation or augmentation, according to Oxford Economics.

So, we shouldn’t be surprised that Hatzius didn’t even bother to discuss how AI deployment could increase productivity and, therefore, GDP, because the data clearly shows that simply isn’t happening.

But, it is actually worse than that, because despite the colossal amount of cash being poured into AI, it isn’t going to get much better than it already is.

OpenAI’s recent research paper found that increasing the computing power behind AI, or shoving more data into them, can’t reduce AI “hallucinations” or make AI more accurate than it is today. In fact, they found there is no viable way to reduce AI hallucinations, heavily implying these models are doomed to stay as unreliable as they currently are. This is corroborated by a paper written by the eminent Vishal Sikka and his son Varin Sikka, which claims to mathematically prove that these AIs “are incapable of carrying out computational and agentic tasks beyond a certain complexity.” All of this also backs up research into the efficient compute frontier, the scaling problem, and the Floridi conjecture, which detail these limitations of AI.

So, just because big tech is shoving $700 billion into AI over the next year, it doesn’t mean it is going to get any better at all. That is why I think ING’s 1% productivity estimate is actually unrealistically optimistic.

But this means that the very justification for that insane $700 billion investment is totally wrong. It isn’t going to make AI properly useful, it isn’t going to boost productivity, it isn’t going to drive GDP growth, it isn’t the economic silver bullet it was promised to be.

So, what does that say about where this is all going? This is a dead end. Shoving that cash into it is just going to make this economic bubble even bigger and make the fallout when it pops even worse.

The more I look at the AI industry, the more it is starting to look like a boys’ club. They are just passing money between themselves to make each other richer, and giving blatantly crap and hollow excuses to the rest of us to explain it away. There are other words for this, like circular financing or an oligarchical economy. But one thing is for sure, it isn’t going to end well.

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTubechannel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

Share