The figures circulating from OpenAI's internal presentations tell two stories at once. Between May and July, the company revised its projected negative free cash flow downward—from roughly $305bn to $278bn across the 2026–2030 period. Yet in the same window, its compute and infrastructure spending estimate climbed sharply, from a publicly stated $600bn target announced in February to approximately $856bn in July materials prepared for a computing partnership. The apparent contradiction dissolves once you understand who is actually paying the bills.

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 same presentation materials forecast revenue reaching $350bn by 2030, up from roughly $36bn in the current year, with compute and infrastructure consuming about $856bn across the five-year span. Samantha Oltman covered the figures for Bloomberg, which issued a correction shortly after its initial publication.

The burn figure is moving in the right direction

OpenAI's May projection showed negative free cash flow of approximately $305bn for 2026–2030. The July revision lowered this to $278bn, representing a $27bn improvement on the company's own prior estimate. The framing of this number as alarming in headlines obscures the fact that it represents a downward revision, not an escalation.

The compute number moved sharply upward

In February, OpenAI publicly communicated a compute target of around $600bn through 2030—a figure presented at the time as a moderation of earlier expectations. Five months later, the July presentation to potential computing partners carried a figure of $856bn, roughly 43% higher than the publicly disclosed number.

One qualification deserves mention: the February figure was labeled as compute, while the July figure encompasses computing power and infrastructure. The categories may not be precisely equivalent, and some portion of the gap could reflect definitional differences rather than genuine cost escalation.

How spending rises while burn falls

The mechanism is straightforward: when capital expenditure lands on someone else's balance sheet, it does not appear in your own free cash flow calculation. OpenAI lacks an investment-grade credit rating, which means its infrastructure financing flows through external partners. Nvidia has been negotiating to guarantee $250bn of data centre debt, enabling lenders to price risk against the chipmaker's creditworthiness rather than OpenAI's.

This pattern extends across the entire buildout. Oracle is deploying more capital on data centres than it generates in quarterly earnings, much of it tied to OpenAI commitments, with the resulting capital expenditure recorded on Oracle's financial statements rather than OpenAI's.

Lease structures achieve similar effects

SB Energy received $5.5bn in OpenAI warrants in exchange for signing a 20-year lease agreement, converting what would be a capital obligation into an operating expense and funding it through equity rather than cash outlay. Free cash flow measures money departing a particular entity; it does not capture obligations incurred, and these two metrics diverge substantially when vendors, landlords and partners shoulder the financing burden.

This structural reality explains how $856bn in compute spending can coexist with $278bn in negative free cash flow. The difference is being absorbed by other parties.

The financing environment shows strain

Credit markets have already signalled where pressure points exist. Oracle required PIMCO to anchor $10bn of a $16.3bn data centre financing after major US banks withdrew from the deal. When institutions of Oracle's stature face bank reluctance, the terms available to smaller borrowers deteriorate accordingly. The guarantees and warrant arrangements are not elegant solutions; they represent what market conditions actually demanded.

The figure that warrants scrutiny

Revenue expanding from approximately $36bn to $350bn by 2030 represents a near-tenfold jump in four years. Every other projection in the presentation flows downstream from this assumption. The burn figure is not an independent forecast; it is the residual after subtracting revenue from spending projections. A shortfall in revenue does not merely reduce burn—it amplifies it.

Treating $278bn as the headline risk inverts the actual logic. The spending is largely locked in through contracts, whereas the revenue target remains contingent on execution. The revenue assumption is what requires validation.

The timeline adds urgency

OpenAI secured $122bn in March at an $852bn valuation and, according to the Financial Times, is projected to deplete those funds by 2028. The projection period extends two years beyond the company's current capital runway. This timeline context explains both the listing timetable and reported discussions valuing the company at approximately $1.2 trillion. A company facing capital exhaustion before 2028 has strong incentive to access public markets before that deadline arrives.

Projections are subject to revision

These figures originate from a presentation designed to secure a computing partnership, not from audited financial statements, and they have already been adjusted twice within the calendar year. OpenAI has also suspended its Stargate development site in the UK due to energy cost concerns and copyright-related regulatory issues. Projects encounter delays and targets get rewritten, which argues for treating any single figure with appropriate skepticism—whether that figure is $856bn or $278bn.

What matters going forward

  • Monitor whether the $856bn figure appears in contexts where OpenAI faces accountability. A number embedded in a deal presentation carries different weight than one appearing in a listing prospectus.
  • Track the guarantees and commitments. If partner balance sheets are carrying the gap between spending and burn, their exposure is the metric worth following.

Source: The Next Web