Analysis
OpenAI expects to burn through roughly $278 billion in cash between 2026 and the end of 2030 as spending on computing infrastructure surges, according to a Financial Times report published Friday and covered by Bloomberg and The Information. The projections, contained in a private presentation prepared in July in connection with a computing deal, show OpenAI expecting to spend roughly $856 billion on compute infrastructure through 2030 while revenue rises from a projected $36 billion this year to $350 billion in 2030. Separately, OpenAI has held early talks with investors about a new funding round that could value the company at more than $1.2 trillion, a roughly 41% premium over its March 2026 valuation.
From $1 Trillion To $1.2 Trillion In A Week
Pulse previously covered Sam Altman telling Fortune on September 12 that an OpenAI IPO in 2026 would be "ill-advised" given unresolved AI-safety questions, with the company still targeting roughly a $1 trillion valuation at the time. Six days later, two things changed: OpenAI now has a specific, self-reported cash-burn number attached to its infrastructure ambitions, and the private valuation target under discussion has climbed to $1.2 trillion -- a bigger number, still entirely outside public-market disclosure rules.
“The $856 billion infrastructure figure is OpenAI's own estimate of what sustaining that pace through 2030 costs.”
What OpenAI Is Actually Spending On
OpenAI, founded in 2015 and the maker of ChatGPT and the GPT-6/Astra model family, has spent 2026 layering compute commitments on top of each other: deals tied to Oracle's Stargate buildout, Broadcom and AMD custom-silicon agreements, and direct capacity purchases from neoclouds including CoreWeave and Nscale, whose $45 billion capacity deal with Anthropic this year set the benchmark for how large a single compute contract can get. The $856 billion infrastructure figure is OpenAI's own estimate of what sustaining that pace through 2030 costs.
The Comparable: Anthropic's October IPO
The most direct comparison sits one filing away: Anthropic is targeting a roughly $2 trillion valuation for a Nasdaq listing it has pushed to mid-October, a number that would nearly double OpenAI's $1.2 trillion private target and make Anthropic's IPO the largest in history if it prices anywhere close to that mark. Both companies are effectively asking investors to underwrite valuations built on projected, not audited, growth -- Anthropic through an actual S-1 that will eventually face public-market scrutiny, OpenAI through a private round that faces none.
The Numbers, Stacked Against Each Other
$278 billion in cumulative cash burn against a 10x revenue climb from $36 billion to $350 billion is a bet that demand keeps compounding without interruption for five straight years. $856 billion in compute spend is more than double the $350 billion revenue target for the same terminal year. Those figures sit inside the same AI-infrastructure capital stack Pulse has tracked all week: Crusoe's $3.9 billion Series F at a $30.9 billion valuation, Nscale's $103.4 billion contracted backlog, and CoreWeave's own $104 billion backlog -- all of it downstream of assumptions about how much compute buyers like OpenAI will actually purchase, and pay for, on schedule.
What the $278 billion headline misses is who produced it and why. This is OpenAI's own five-year forecast, built into a private deck designed to support a specific computing deal -- an interested party's internal projection, not a number any outside auditor or public-market analyst has tested. Five-year infrastructure and revenue forecasts are acutely sensitive to compute-pricing and demand assumptions that can move sharply in either direction, and the same week the company is citing AI-safety concerns as its reason to avoid the disclosure obligations of a public listing, it is shopping a larger number to private investors instead.
For VCs and LPs with indirect AI-infrastructure exposure, OpenAI's own burn projection is effectively the demand-side assumption every supply-side bet -- Crusoe, CoreWeave, Nscale -- is underwriting. If OpenAI's revenue ramp slips even modestly short of $350 billion by 2030, the compute buildout financed against that assumption doesn't have an obvious backstop. What comes next is whether the $1.2 trillion round actually closes at that price, and whether Anthropic's October IPO prices anywhere near its own $2 trillion target -- the first real public-market test of whether outside investors will pay these multiples once the numbers face quarterly disclosure instead of a private deck.
