OpenAI Is In A Far Worse Position Than I Thought

This is beyond reckless…

Photo by Ryan Braxton on Unsplash

In a previous article, “OpenAI’s Insane Scaling Problem”, I made a mistake. Much to my accountant father’s shame, I confused Annualised Recurring Revenue with Annual Recurring Revenue. As such, I thought OpenAI had claimed to have made $20 billion in revenue last year. I argued that this amount simply wasn’t enough, as it would still mean rapidly growing losses for OpenAI, and that this figure suggests Microsoft is tossing OpenAI a bone, as Ed Zitron had previously found. After correcting this mistake, I discovered that the situation over in Altmanland is catastrophicallyworse than I believed. Let me explain.

The Correction

OpenAI’s $20 billion ARR figure originated from a blog by OpenAI CFO, Sarah Friar. She took the revenue OpenAI made in December 2025 and then multiplied it by 12 to ‘annualise’ it. This means she is claiming OpenAI made $1.66 billion in December 2025.

OpenAI had previously stated they made $5.5 billion in annualised revenue in December 2024 and $10 billion in annualised revenue in June 2025. If we assume linear revenue growth between these points (Dec 2024, June 2025, and Dec 2025), we can estimate that OpenAI’s total revenue for 2025 was actually $11.9 billion.

You can see how my mistake made OpenAI’s position look much better than it actually is.

The Microsoft Loophole Is Still There

Despite this, the revised figure still suggests that Microsoft is tossing OpenAI a bone.

In my earlier article, I mentioned that Ed Zitron found that OpenAI paid Microsoft $454.7 million in revenue sharing during the first half of 2025. As part of their partnership, OpenAI is expected to pay Microsoft 20% of its revenue. So that means OpenAI made $2.27 billion in revenue during the first half of 2025, which is $2 billion less than reported. In Q3, OpenAI paid Microsoft another $411.1 million, meaning they made $2 billion in revenue in Q3. We don’t yet have figures for OpenAI’s Q4 revenue sharing with Microsoft. However, we know OpenAI didn’t experience a dramatic increase in paid or unpaid users during this period, so we can reasonably estimate that the figures would continue to follow a similar upward trend and imply OpenAI had a revenue of under $4 billion for Q4. This would equate to an annual revenue of around $9 billion.

That is substantially less than the $11.9 billion annual revenue that OpenAI suggests. So, why is there a difference?

Well, Microsoft uses OpenAI’s models to run the likes of Copilot and Azure AI, and it pays OpenAI a 20% revenue share for this resale (according to Zitron). This revenue is almost certainly unaccounted for when calculating OpenAI’s 20% revenue share with Microsoft, so it could explain the difference.

As a result, we can estimate that Microsoft paid OpenAI approximately $2.9 billion in 2025, which would indicate that Microsoft generated roughly $14.5 billion from reselling OpenAI models in 2025. But did they really generate that much? I don’t think so.

Microsoft is severely struggling to sell these AIs. For example, we know that their Azure AI based on ChatGPT has sold so poorly that they have slashed sales quotas in half (read more here). Likewise, only 3.3% of Copilot users pay for the service. So, Microsoft has embedded it into existing subscriptions, such as Microsoft 365 (formerly Office), to force it upon its users. In fact, as a new customer in the UK, I cannot buy a 365 subscription without Copilot. Moreover, Microsoft’s 2025 revenue was only 15% higher than in 2024, and the vast majority of that increase came from their Azure cloud computing business, not from AI sales.

It looks to me like Microsoft might be classifying an arbitrary amount of its revenue from these existing subscriptions, which are now embedded with AI, as AI-driven revenue. This would inflate the amount of revenue they generate from AI and allow them to inconspicuously provide OpenAI with some extra cash (though, admittedly, less cash than I previously believed).

The Cash Problem

But why would OpenAI need extra cash? Quite simply, they need every cent they can get!

Multiple analyses have found that OpenAI’s operating costs will be around $28 billion in 2025 (Zitron & The Information). Based on our more realistic $11.9 billion 2025 revenue estimate, that means OpenAI is likely to incur a net loss exceeding $15.6 billion in 2025.

That is nearly double the predicted loss for 2025 and triple their net loss of 2024!

However, even that figure might contain some accounting trickery to hide the true scope of these losses. It could include Microsoft overpaying OpenAI for AI reselling, as discussed. But it could also hide debt obligations (read more here) and overly optimistic data centre amortisation (depreciation) (read more here), both of which are known tactics in the AI industry.

In short, OpenAI is a money black hole, and to keep investor cash flowing in and the AI bubble from crashing, OpenAI (and Microsoft, as the majority holder) needs to do everything it can to polish this turd. Why else do you think they touted their annualised revenue instead of their annual revenue? Don’t be surprised if the books paint the story a little rosier than reality.

The Crunch

Along with that $20 billion claim, Friar also mentioned that OpenAI ended 2025 with 1.9 GW of compute power. This demonstrates that OpenAI is doomed to insane losses. Let me explain.

If OpenAI’s revenue continues to grow at its current rate, it will generate approximately $30.8 billion in 2026 (based on our $11.9 billion annual revenue estimate). Sounds good, right? Well, it isn’t enough.

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. So OpenAI’s 1.9 GW of compute will cost around $53 billion annually.

If OpenAI decides not to build any more data centres and maintains 1.9 GW for the entirety of 2026, it would still incur an annual net loss of approximately $22.2 billion.

But that isn’t what OpenAI is doing — instead, they are significantly ramping up their compute power, with some $1.4 trillion in infrastructure spending being committed between now and 2030. So its 2026 loss is likely to be substantially higher than that!

The AI bubble doesn’t just include the hype around AI as a product but also around the circular funding between AI companies, data centre operators, and chip makers. Indeed, much of OpenAI’s funding comes from this circular nature. Because of this, OpenAI couldn’t halt its compute expansion if it wanted to, as doing so could stop the flow of investor capital that offsets its substantial losses and prevents it from going bankrupt.

So, even though $20 billion in annualised recurring revenue and 1.9 GW of compute power sound promising, in reality, they forewarn a future of catastrophic, ever-growing losses. And we all know where that leads.

But It Gets Worse!

Somehow, Altman has found a way to make this already dire situation even worse.

The Wall Street Journal recently reported that Nvidia’s plan to invest a whopping $100 billion into OpenAI has stalled. OpenAI wanted to seal the deal within a few weeks, but it turns out Nvidia isn’t happy with their business model, and so the deal is reportedly going nowhere.

This $100 billion would have been critical not only for OpenAI’s planned compute expansion but also for covering its losses. Furthermore, Nvidia is a huge part of the circular funding that feeds OpenAI, which suggests a break in this critical chain is forming. Either way, this stall is a monumental problem for OpenAI that makes its mounting losses even more of an existential threat.

Nvidia is right; OpenAI’s business model is deeply worrying, and I’m not talking about the introduction of ads either.

OpenAI’s plan to reach profitability, or at the very least reduce its losses to near-sustainable levels, is heavily reliant on transitioning from selling subscriptions for their plagiarism word calculator to revolutionising the economy to be dependent on their “agentic AI”. In other words, OpenAI will have to develop and sell an AI agent capable of replacing human labour at scale. It isn’t just me or Altman saying that, but Geoffrey Hinton, one of the “Godfathers of AI” (read more here).

It is this pitch that justifies the colossal investment in AI data centres, the circular funding, and the disproportionate valuations of AI companies.

But here’s the problem: these agentic AIs don’t work, and it’s looking increasingly like they never will.

A recent study from Carnegie Mellon University found that even the best “agentic” AIs fail basic tasks 70% of the time. This explains why MIT found that 95% of corporate AI pilots fail, PwC found that only 12% of businesses using AI saw it reduce costs and increase profit, BCG found that only 5% of companies that deployed AI saw value from it, and Forrester Research found that 15% of their corporate survey correspondents reported an increase in profit margins from AI over the past year. It also might explain why The Economist found that the use of AI by large US corporations is actually shrinking, not growing, and why S&P Global Market Intelligence found that the cancellation rate of corporate AI programs skyrocketed from 17% in 2024 to 42% in 2025.

Even worse, it looks more and more like pouring additional compute power or data into these models, as OpenAI is doing, isn’t going to make them any better.

A recent research paper by OpenAI directly admitted this. The paper found that increasing the computing power behind these models, or shoving more data into them, can’t reduce AI “hallucinations” (which caused that 70% failure rate) from their current level. 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.

We may even have a mathematical explanation as to why this is the case. Wired recently covered 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.” Sikka even told Wired, “There is no way [agentic AI] can be reliable.” This paper isn’t peer-reviewed, and I do not have the mathematical skills to even pretend I understand the argument it presents. However, there is substantial research to support it, including work on the efficient compute frontier and the Floridi conjecture.

In short, it doesn’t matter how much cash or computing power OpenAI throws at the problem; agentic AIs will not get much better than they are today. Naturally, their path to profitability, the justification for their insane expenditure, the colossal amount of investment flowing their way, and their vast valuation are all predicated on a falsehood.

Summary

It’s no wonder Nvidia is getting cold feet, and we can clearly see why OpenAI is desperately trying to polish their financial turd — possibly with Microsoft’s assistance. When you look past the industry propaganda and dig a little deeper, the picture is utterly dire. There is only one way I can describe OpenAI: completely f**ked.

Let’s also not forget that thanks to AI-tied debt, AI companies soaking up almost all the investment in the room, and OpenAI’s position as the lynchpin company of the AI bubble (read more here and here), the entire Western economy has effectively bet its health on OpenAI’s success.

This feels more akin to Russian roulette than any kind of safe bet, because, sooner or later, it will end with a bang. But should we really be surprised? This entire bubble was caused by tech bros joining the death-cult-like AI accelerationist movement (which I should write an article about at some point). This huge impending car crash is not a bug but a feature. I just wish these new-age oligarchs hadn’t locked us all up in their boot before slamming the gas to the floor.

---

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!