OpenAI is preparing for one of the most expensive corporate buildouts in technology history. Internal forecasts reviewed by the FT show the company expecting about $278 billion in negative free cash flow from 2026 through 2030, even as it projects a steep rise in sales.
The gap is driven almost entirely by computing and infrastructure, which the company now estimates will cost roughly $856 billion over that period.
Those figures come from a July presentation prepared in connection with a computing agreement and later seen by the newspaper.
They describe a strategy that treats compute as the scarce resource that will determine who leads the next phase of artificial intelligence.
Training larger models and serving hundreds of millions of users requires vast clusters of chips, power, and data-center capacity.
OpenAI has therefore locked in long-term supply deals and is still expanding that footprint.
Revenue is expected to grow rapidly alongside the spending.
The company forecasts sales of $36 billion in 2026, rising to $350 billion in 2030.
Cumulative revenue over the five years is projected at about $840 billion. In other words, the business would generate nearly as much cash as it plans to spend on infrastructure—and still finish the decade hundreds of billions of dollars short because the capital outlays arrive first.
The latest cash-burn estimate is an improvement on an earlier internal view from May, which put negative free cash flow closer to $305 billion.
Even so, the March fundraising of $122 billion, completed at an $852 billion valuation, is expected to be exhausted by 2028 if spending follows the current path.
That timeline has already prompted discussions of a new round that could value the company above $1.2 trillion.
Competition is tightening the financial picture.
OpenAI has reduced prices to defend share against Anthropic and inexpensive open-weight models.
Those cuts help volume but compress margins at the same time that training and inference costs are rising.
New model releases have lifted annualized revenue—one July jump was reported at about 20 percent—but the company still expects expenses to outrun incoming cash for years.
The scale of the bet is unusual even by Silicon Valley standards.
Few private companies have ever planned to consume hundreds of billions of dollars of capital while remaining unprofitable for most of a decade.
OpenAI’s leadership is betting that demand for frontier models will eventually justify the outlay and that first-mover access to compute will prove decisive.
Investors appear willing to fund that thesis for now, but the projections make clear that additional equity, and possibly more debt, will be required well before 2030.
Chief executive Sam Altman has already delayed a public listing that was confidentially filed in June, citing safety concerns around rapid AI progress.
The cash-flow forecasts reinforce why the company may prefer to stay private a little longer: public markets tend to penalize multi-year losses of this magnitude, even when revenue is growing quickly.
The presentation therefore leaves OpenAI in a familiar but intensified position.
It must keep raising money, keep building capacity, and keep growing revenue fast enough that the enormous compute bill eventually looks like a strategic investment rather than an open-ended drain. Whether that sequence holds will shape not only the company’s future but the broader economics of the AI industry.