The Financial Times reported that OpenAI expects negative free cash flow of $278bn between 2026 and 2030, from a July presentation prepared for a computing deal. That figure is an improvement on the roughly $305bn the company projected in May, while the compute and infrastructure line has risen to about $856bn against the roughly $600bn target it gave investors publicly in February. Spending can rise while burn falls because much of the build is financed by partners rather than by OpenAI.

The Financial Times reported that OpenAI expects negative free cash flow of $278bn between 2026 and the end of 2030, citing a company presentation prepared in July for a computing deal. Samantha Oltman wrote up the figures for Bloomberg, which corrected its story shortly after publication.

The same materials put revenue at $350bn in 2030, against roughly $36bn this year, and compute and infrastructure at about $856bn across the period. Three numbers, and the smallest one is getting the headlines.

The burn figure is a downward revision

An earlier projection in May put negative free cash flow for the same period at roughly $305bn. The July version is $278bn.

So the number being reported as alarming is an improvement of about $27bn on OpenAI’s own previous estimate. That does not make it small, and it does make the framing odd.

The compute number went the other way

In February the company publicly reset expectations. It told investors its compute target was around $600bn by 2030, a figure presented at the time as a moderation.

Five months later a private presentation carries $856bn. That is roughly 43% higher than the number given publicly in February.

One caveat belongs here. The February figure was described as compute and the July figure as computing power and infrastructure, so the categories may not be identical, and part of the gap could be definitional rather than real.

How both movements happen at once

Spending rises, burn falls. That combination is possible when the spending does not sit on your own balance sheet.

OpenAI does not hold an investment-grade credit rating, which is why its financing runs through other people. Nvidia has been in talks to guarantee $250bn of data centre debt, letting lenders price against the chipmaker’s credit instead.

The pattern repeats across the build. Oracle is spending more on data centres than it earns in a quarter, much of it against OpenAI commitments, and the capital expenditure lands on Oracle’s accounts rather than OpenAI’s.

Leases do the same work

Structure matters as much as scale. SB Energy took $5.5bn in OpenAI warrants to sign a 20-year lease, which converts a capital commitment into an operating one and pays for it in equity rather than cash.

Free cash flow is a measure of money leaving a specific entity. It is not a measure of obligations created, and the two diverge sharply when a build is financed by vendors, landlords and partners.

That is why $856bn of compute can coexist with $278bn of burn. The difference is being carried by somebody.

The financing conditions are not comfortable

The credit market has already shown where the strain is. Oracle needed PIMCO to anchor $10bn of a $16.3bn data centre financing after US banks stepped back.

When banks retreat from a name like Oracle, the terms available further down the chain are worse. The guarantees and warrants are not elegance, they are what the market required.

The number that should be interrogated

Revenue rising from about $36bn to $350bn by 2030 is close to a tenfold increase in four years. Every other figure in the presentation is downstream of it.

The burn is not an independent forecast. It is what remains after the revenue assumption is subtracted from the spending assumption, so a revenue miss does not shave the burn, it compounds it.

Reporting the $278bn as the headline risk inverts the logic. The spending is largely contracted, and the revenue is the part that has to show up.

The clock is the other story

OpenAI raised $122bn in March at an $852bn valuation, and the FT reports it is on track to exhaust that by 2028. The projection period runs two years beyond the money.

That is the context for the listing timetable and for talks the FT describes as valuing the company at about $1.2 trillion. A company that needs capital before 2028 has a reason to be in the market before then.

Plans do change

These are projections in a document prepared to win a computing deal, not audited accounts, and they have already been revised twice this year. OpenAI has also paused its Stargate site in the UK over energy costs and copyright rules.

What to watch

Watch whether the $856bn figure appears anywhere OpenAI can be held to it. A number in a deal presentation and a number in a listing prospectus carry different consequences.

Watch the guarantees. If partner balance sheets are carrying the difference between the spending and the burn, the exposure worth tracking is theirs.