Tech • AI • Robotics • Game
OpenAI at the heart of the AI bubble: can we avoid the worst?
9.4/10TL;DR
OpenAI’s reported revenue run rate of about $50 billion instead of $70 billion has sharpened fears that the AI boom is colliding with a financing model built on huge fixed commitments, even as demand for the technology remains strong.
KEY POINTS
A lower figure jolts the market
Wall Street reacted sharply after OpenAI confirmed a run-rate revenue figure near $50 billion, not the $70 billion many had expected. The gap mattered less as a simple accounting dispute than as a signal that growth may be slower than required to support the vast spending now tied to the AI buildout.
Technology optimism remains intact
The core debate is not whether generative AI has value. Frontier models are already reshaping software, coding, finance, health and business operations, and capacity remains tight across data centers. The concern is whether the current financing structure can survive the pace of spending needed to sustain that expansion.
The boom is being financed with heavy commitments
The four main hyperscalers are spending roughly $725 billion in capital expenditure this year, with a growing share supported by debt. Around OpenAI, suppliers and partners have signed large take-or-pay contracts, meaning capacity must be paid for whether it is fully used or not. That makes the system more fragile than a standard equity bubble.
A credit shock would be worse than a valuation reset
In the dot-com crash, stocks fell, but much of the damage came through equity repricing. In AI, the larger risk is a credit event. If a major buyer cannot meet its commitments to cloud and chip suppliers, the problem can spread quickly through lenders, infrastructure providers and public markets.
OpenAI sits at the center of that chain
The company is a key customer for a wide network that includes Oracle, CoreWeave, Microsoft and Nvidia. Stargate and related compute projects are heavily exposed to demand from top AI labs. A serious financing problem at OpenAI would therefore hit far beyond one startup and could ripple across hundreds of billions of dollars in planned infrastructure.
Growth is colliding with price deflation
Token prices are falling rapidly as models get cheaper and routing improves. One estimate cited in the debate said token deflation has been 54 times faster than the decline in electricity costs during the industrial revolution. That helps adoption, but it also pressures revenue because more usage is being shifted to cheaper models.
Open source and routing are changing the economics
Cheaper open models, including Chinese and open-source offerings, are taking a growing share of usage. On OpenRouter, those models account for about 60% of tokens, while data from Vercel showed open-source models at 56% of tokens but only 14% of spending. The implication is that premium labs still capture most revenue, but the market is increasingly routing demand away from the most expensive frontier systems.
Only a small share of work may need frontier models
A central question is how much enterprise work truly requires top-end models. Some estimates place that share at just 2% to 5% of workloads. If that is correct, then the extreme revenue projections used to justify trillion-dollar valuations become harder to sustain unless those labs dominate the highest-value tasks and preserve strong margins there.
OpenAI’s obligations may exceed its near-term revenue power
Estimates discussed around the market suggest OpenAI could have around $750 billion in compute commitments, with roughly 80% effectively payable, or about $600 billion over time. Spread across five years, that implies about $120 billion a year before other costs. That is why investors are focused not only on revenue growth, but on whether revenue can rise fast enough to cover fixed obligations.
Anthropic is seen as growing faster in enterprise
Some investors now view Anthropic as having a cleaner enterprise profile, with a larger share of business revenue and a faster growth trajectory. OpenAI remains larger and central to the ecosystem, but questions are intensifying over the quality of its revenue mix, the pace of enterprise adoption and whether future fundraising can keep up with its commitments.
Europe would not benefit from an AI bust
The idea that a US AI financing shock would help Europe was rejected by market participants. A slowdown in American AI investment would likely hit global growth and could push Europe deeper into weakness. More importantly, the technology itself would not disappear. The strategic opportunity for Europe lies less in catching up on model production than in becoming faster at adoption across industry.
CONCLUSION
The AI boom still rests on real demand and transformative technology, but the market is starting to distinguish between technological progress and the debt-heavy structure built around it. The next test is whether leading labs, especially OpenAI, can turn explosive adoption into revenue strong enough to support the infrastructure they have already promised to buy.
Ask a question