xAI Is Dying
It's about time…
Thanks to SpaceX bailing out — I mean, merging with — xAI, its upcoming IPO is as much about investing in the future of AI, Grok, and the promised terawatts worth of orbital data centres soon to be powering it as it is Musk’s feudal techno slave colonies on the barren red planet. That is the only way the utterly bonkers proposed valuation of $1.75 trillion makes even the slightest bit of sense, and even then, it is still hare-brained. There is just one teeny tiny problem with this narrative. xAI, the cornerstone of the speculative value, is rapidly dying before our eyes. So, in no particular order, let’s go through all the different ways it is failing, and what this means for SpaceX.
Lower Users & Low Utilisation
xAI’s main product is the chatbot Grok. But it sucks, as many users find it wildly inaccurate and almost useless. So, while it does appear to do well in benchmarks, that means nothing as its real-world performance is dire. There is a reason it hasn’t threatened the market share of OpenAI or Anthropic. Indeed, it seems Grok’s main value proposition as a ‘nudifyer’ tool for creepy perverts (more on that in a minute).
This is reflected in the numbers. Forbes recently reported that Grok’s weekly users dropped a whopping 10% from March to April, and the Grok app fell from 2nd place globally in January to 5th place by April, behind Claude and Gemini. In other words, no one is using xAI main product.
Which is a bit of a problem for Musk, as he built a giant, multi-billion-dollar data centre called Colossus to run Grok. However, The Information recently reported that xAI’s utilisation rate was just 11%, way below the typical 40% of its rivals. In other words, so few people are using Grok that this eye-waveringly expensive data centre is mostly sitting idle.
There are two problems with this.
Firstly, Musk’s plan to deploy multiple terawatts of orbital AI data centres to power xAI is a load-bearing promise for SpaceX’s speculative value, and is hinged on the notion that xAI would require a biblical amount of computing power to meet demand. The fact that even xAI’s comparatively small data centres (most likely less than a GW) are already a massive oversupply totally undermines that narrative, and in part, the speculative value it holds up.
But also, xAI went into a vast amount of debt to build these data centres, reportedly around $17.5 billion! That means this oversupply will be a massive drain on xAI’s books, as the huge amount of largely unnecessary debt eats away at its balance sheet. From a business management point of view, this is catastrophic.
The Anthropic Deal
But don’t fear, Elon has a solution! Kind of…
He has rented the entirety of xAI’s 300 MW Colossus 1 data centre to Anthropic. This works out great for Anthropic, which is massively struggling to meet demand with constant outages, as its user base has exploded in recent months, and it doesn’t have enough compute on hand to deal with it. This deal should help solve that. It also means that this data centre will be fully used, and bring in some cash to help cover the debt costs xAI took on to build it.
But let’s not forget that Anthropic is a direct competitor to xAI. Indeed, just a few months ago, Musk was saying Anthropic was “misanthropic and evil.” Handing over this data centre to Anthropic will directly impact xAI’s bottom line, as it will strengthen Anthropic, which in turn will take xAI users. Don’t forget that an xAI user is likely far more profitable to xAI than an Anthropic user using their data centres via this deal, thanks to vertical integration and lack of a middleman.
As such, this deal is not really a solution, as in the long run, it will likely harm xAI more than it will help them. As such, it is more of an admission of just how desperate xAI is.
The Cursor Deal
But this isn’t the first deal like this, Musk has done. In fact, he has kind of double-booked himself.
You see, a few weeks ago, SpaceX/xAI signed a deal that gave them the option to purchase the AI coding company Cursor for $60 billion in SpaceX stock. Oddly, this deal included a breakup clause in which SpaceX would have to pay Cursor $10 billion if it failed to buy them.
This deal appeared to make sense for xAI because Grok is losing the AI coding race badly, so merging with Cursor could help them compete with the likes of Anthropic, as they are faring much better in this race.
This deal made a ton of sense for Cursor, too. For one, they were previously valued at $29.3 billion, so this deal doubles their value. But previously, they just used ChatGPT or Anthropic tokens and sold them on to their customers at a lower price. That is why they were competitive in the AI coding tool race; it was the same AI as the competitor, just cheaper. Terrible business model, but a great way to drum up a customer base. So, they wanted to switch to using their own AI to stem the losses, but they needed the infrastructure to do that. This deal did exactly that, as part of it was for Cursor to run its models on xAI’s Colossus data centre.
You know, the one Anthropic now has exclusive use of… More on that in a bit.
But here is the funny part. You’d think this means that SpaceX/xAI signed this deal to buy Cursor’s own proprietary coding AI tool, right? Or, if not that, power Cursor with Grok and use their expertise to make Grok better at coding. After all, this seems to be what’s so valuable here.
Well, Cursor do not have its own AI. Instead, their current tools are powered by Kimi 2.5, an open-source, free-to-use Chinese AI model. So, SpaceX just signed a $60 billion deal to run an open-source AI in their data centres, which they could have just run themselves for free. It makes no sense.
In other words, SpaceX/xAI didn’t agree to buy Cursor for its technology, expertise, or to expand Grok into its product line. They did so to get more usage out of their desperately under-utilised data centre. It is an admission that Grok is a failure, and they have too much expensive compute on hand, and desperately need to plug that financial hole.
What This Means For SpaceX/xAI Compute
Okay, so what does this all mean for Grok, xAI and SpaceX? Well, the implications aren’t exactly good.
With Anthropic taking up all of Colossus 1, where will Cursor and xAI operate from?
Well, xAI has a second data centre called Colossus 2. It has been widely touted as the world’s first gigawatt-capable data centre, and back in January, that it had reached 1 GW capacity, or more than three times the power of Colossus 1. As always with Mr Pants on Fire, this wasn’t exactly true, with satellite data showing that it was more like 300 MW, or the same size as Colossus 1. It has probably grown in capacity since then, but considering the lack of reporting and the fact that xAI has a compute over supply, it might still be down at the 300 MW mark.
With Anthropic getting all of Colossus 1, Cursor and Grok will have to share Colossus 2. But, Cursors has over a million daily users, and those users are using it for demanding coding tasks. So, they will likely require several hundred MW of compute just to meet that demand. Now, SpaceX/xAI has effectively given them compute for free, so it makes sense for them to use as much of that as possible. There isn’t enough data out there to accurately predict how much of Colossus 2 will be taken up by Cursor, but there is a good chance it is most of the currently installed power.
So, where does that leave xAI and Grok?
Their low utilisation rate shows that they only need a small amount of computing to meet inference (querying the AI) demand. Therefore, I’m sure there will be enough compute space left in Colossus 2 for that.
But Grok needs far more than just inference. In March, Musk said that xAI needed rebuilding from the ground up, as the models were built wrong the first time around. This raised the question as to what the hell SpaceX had just bought a month beforehand. But, in other words, Grok doesn’t just need retraining, but an entire new model built from scratch. That will take an enormous amount of training, which requires a huge amount of computing power.
So, where are they going to get that computing power from? Again, by solving the underutilisation problem, they have shot themselves in the foot.
There is also another issue here. Yes, Anthropic will pay SpaceX/xAI for this deal, and yes, buying Cursor will effectively buy them a ton of paying users. In theory, both deals will bring in revenue that can be used to pay xAI’s vast debts. But, Musk seems to have forgotten that AI data centres are not profitable. You can’t solve xAI financial woes by letting Grok die and selling the compute power you built for it to other competitors.
That is a huge problem because like very other giant AI lab, xAI is bleeding cash. xAI was bleeding $1 billion a month in 2025, and it was projected to lose $13 billion last year. SpaceX posted a $5 billion loss in 2025 likely because it took on xAI’s debt and losses. That’s right, xAI is in such a bad financial position that it took the giant company that bailed it out into the red.
Failed Premise
Musk’s SpaceX $1 trillion payment packet is tied to him hitting certain targets, one of which is deploying 100 TW of orbital data centres. If you want to know just how utterly stupid that is, read my previous article here. But the entire narrative of this promised vast orbital computing power was that the xAI needed it. SpaceX is selling itself as a ‘vertically integrated innovation engine’, where its rockets launch AI data centres that power cutting-edge AI and humanoid robots that would dominate the economy. Maximum efficiency, no middle man, unrivalled impact.
But, SpaceX/xAI is having to sign two massive deals to fill less than a gigawatt of compute capacity, or just 0.001% of Musk’s 100 TW target. That undermines the entire premise of this vertically integrated innovation machine. It shows that the demand for their AI is pitiful, and doesn’t need that amount of power. It shows that, even at this tiny scale compared to where they want to go, they have to literally buy customer bases for twice what they are worth (cursor) and rent their compute power to rivals who are already crushing them.
The narrative of xAI and in turn SpaceX’s speculative value is demolished by this revelation.
The Illegality
All of that is pretty damning evidence that xAI ain’t doing so well. But let’s also not forget that xAI is in deep legal trouble.
Musk has snubbed being summoned by French prosecutors regarding a criminal investigation into xAI over Grok’s rampant production of non-consensual sexual deepfakes, including CP, and other illegal content. In fact, xAI’s French offices were recently raided as part of this investigation, and other countries, such as the UK, have opened their own investigations into xAI. In short, over the coming months, xAI and Musk may be held accountable for the heinous things Musk not only let Grok do, but openly advertised it would do. Legally, ethically and operationally speaking, this is horrific news for both Musk and xAI.
It’s also bad news financially, as it appears that a serious portion of Grok users are solely using it as a ‘nudifier tool’, and in the wake of these investigations, the EU is looking to ban such tools.
This, on its own, could be the death blow for xAI, but everything else we have talked about, it’s more like the last sprinkling of soil on its coffin.
Summary
With all of this in mind, it is obvious that xAI is dying. Its user base is rapidly disappearing, it is having to buy users at exorbitant prices, it is giving its compute away to its main rival, the entire premise that it will one day need orbital data centres is dead in the water, and legally speaking, its main value proposition as a nudifier tool for internet creeps is being banned, and the company is up to its neck in s**t. All of that while it bleeds so much cash that it has sent its newfound parent company into the red, and not the planetary kind. This is not sustainable; this is the death rattle of a failing company.
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!