Will OpenAI collapse?
I wouldn’t be surprised to see OpenAI collapse within the next decade or so. The most valuable startup in history is haemorrhaging money at a pace that would make even the most reckless dot-com bubble executives blush. In the first half of 2025 alone, the company burned through $13.5 billion, with another $12 billion lost in the most recent quarter. That’s over $30 billion in a single year for a company whose entire business model rests on the assumption that people and companies will eventually be so reliant on large language models like ChatGPT that they’ll be willing to pay hefty premiums for access. That is a fairly risky assumption.
Sam Altman wants us to believe this is all part of the plan, that OpenAI will generate $20 billion in revenue this year and scale to ‘hundreds of billions’ by 2030. And maybe it will turn out that way. I don’t pretend to know enough to rule it out, but from a lay perspective, the fundamentals just do not work.
ChatGPT has 800 million weekly active users, but only 5% of them pay for the service. Sure, hundreds of millions of people use ChatGPT every week, but when you’re spending billions on computational infrastructure, employing thousands of engineers, and training models that cost $7 billion annually just for the compute, a 5% conversion rate seems catastrophic.
The company is currently valued at around 30 times its revenue, a staggering multiple that assumes exponential growth will continue indefinitely. But where is that growth meant to come from? OpenAI launched its $200-per-month Pro tier expecting to make a profit, but heavy usage pushed it into losses instead. LLMs are a remarkable technical and linguistic achievement, but there is little evidence that there’s a broad market willing to pay premium subscription fees for them, particularly when their obvious limitations mean that they don’t quite fall into the essential category of digital services.
Meanwhile, OpenAI has committed to $1.4 trillion in infrastructure investments over the next eight years. One point four trillion dollars.
The recent Nvidia investment in OpenAI carried a $100 billion headline, but only $10 billion is actually available upfront, with the next $10 billion contingent on OpenAI spending more than $50 billion first. So basically, near-term funding is receding while commitments run wild. This is why OpenAI’s CFO Sarah Friar caused such a furore when she suggested the US government should ‘backstop’ the company’s chip investments. The suggestion revealed what many suspect, that OpenAI doesn’t actually know how it will pay for its own plans.
Tech critic Ed Zitron coined the term that captures this moment perfectly: the subprime AI crisis. Just as banks lent out more credit than they could ever recover in 2007, Wall Street has bet billions on OpenAI reaching a valuation built on extreme speculation rather than sound financial analysis.
The entire tech industry has bought into a technology sold at a vastly discounted rate, heavily subsidised by big tech. What happens when the subsidies run out? What happens when OpenAI is forced to implement the price hikes that its business model requires?
Competition will only make things worse. Google and Meta are all building their own LLMs with far deeper pockets. And they can give their models away for free, driving adoption of their other services, a more realistic path to profitability that isn’t available to OpenAI. Then there’s the open-source movement. The technical differentiation between leading models continues to narrow, potentially reducing OpenAI’s competitive edge. When far smaller startups can release something like Stable Diffusion or when Meta can give away Llama models that rival GPT in capability, what exactly is OpenAI’s enduring advantage? OpenAI may soon learn that first-mover status only lasts so long.
Altman’s defenders typically argue that it’s all just positioning for ‘Artificial General Intelligence’, and that once AGI is achieved, well that will unlock $100 billion in profits, justifying every dollar burned along the way. But AGI is more like a religious belief than a technical roadmap, and the reality is that AGI is just not attainable with current approaches. Problems of reasoning, data bottlenecks, and hallucinations can not be solved through scale alone. But it seems as though OpenAI’s entire financial strategy assumes that if they just spend enough money to build enough compute, AGI will simply emerge.
A 23-year-old former OpenAI employee turned investor published a 165-page manifesto about AGI titled ‘Situational Awareness’. He basically argues that you either see the truth of what’s coming or you don’t, and you don’t need hard facts. Is this the discourse that’s guiding trillion-dollar investment decisions?
And then there’s the Microsoft problem.
Microsoft receives 20% of OpenAI’s revenue and provides crucial cloud infrastructure, creating a dependency that should worry anyone concerned about the company’s long-term autonomy. The partnership that’s currently OpenAI’s greatest asset could eventually become its biggest vulnerability.
Microsoft’s stock has gained 15% year-to-date largely because of its OpenAI partnership, but what happens when Microsoft decides it can build better AI tools internally? The tech giant already has its own AI research division, it already has the compute infrastructure, and it already has the distribution channels through Office and Azure. Why would Microsoft continue subsidising a competitor that’s burning through capital at an unprecedented rate when it could simply build the same capabilities in-house?
When the music stops, OpenAI will either need government intervention or face a fire sale to Microsoft. The latter is the likely outcome, framing bankrupty as absorption. That way, the hype can survive.