OpenAI Lifts Compute Spending Forecast to $750 Billion

The number that defines the AI build-out just got bigger. According to a Wall Street Journal report published July 22, 2026, OpenAI has raised its projected spending on compute and cloud infrastructure through 2030 to roughly $750 billion, up from the roughly $600 billion figure the company had set earlier in the year. The revision — about a 25% increase — is being driven by a fresh wave of deals with cloud-computing providers as OpenAI races to lock in the vast computing capacity its models consume.

The disclosure lands in the middle of a debate the market has been having all summer: how large can AI infrastructure commitments grow before the economics buckle. OpenAI’s answer, at least for now, is that they can grow another $150 billion in a single revision.

What’s behind the $750 billion

The higher forecast reflects contracts OpenAI has been signing at an accelerating pace. Among the commitments already on the books, per the reporting:

• A contract with Oracle covering roughly 6 gigawatts of data-center capacity.
• An expanded arrangement with Amazon Web Services reported at $138 billion across eight years.
• A separate pledge of $250 billion in incremental spending routed through Microsoft Azure.

The company is also moving to own more of the physical layer directly rather than renting all of it. Days before the forecast revision, OpenAI unveiled Project Camellia, a data-center campus in Effingham County, Georgia representing at least $20 billion in investment — its first venture as the principal designer and builder of its own site. We covered that campus in detail in our report on Project Camellia; the $750 billion figure is the aggregate spending envelope that projects like it now sit inside.

Taken together, the deals sketch a company trying to secure supply on every axis at once — leased hyperscale capacity, long-term cloud commitments, and self-built campuses — rather than betting on any single provider. That multi-sourcing is a hedge against exactly the concentration risk that dogged OpenAI’s earlier, near-exclusive reliance on one partner.

The losses behind the spending

The spending sits atop a balance sheet that is bleeding heavily by design. OpenAI generated more than $13 billion in revenue in 2025 — reported at about $13.07 billion — while posting a net loss of $38.5 billion.

That headline loss requires an asterisk. The figure was heavily inflated by non-cash accounting charges tied to OpenAI’s conversion to a Public Benefit Corporation, with changes in the fair value of convertible interests and related liabilities generating more than $41 billion in paper losses. On a cash-relevant basis, total costs and expenses reached about $34 billion, with research and development alone accounting for roughly $19.18 billion. Even stripped of the accounting noise, the operating burn is enormous: OpenAI paid Microsoft alone about $17.2 billion in total expenses last year, including $10.59 billion for R&D.

Analysts remain skeptical that the math converges quickly. A Deutsche Bank analysis cited alongside the news projects roughly $143 billion in cumulative negative free cash flow through 2029 before the company’s own model turns sharply profitable in the following years. For context on why loss-making AI firms can still raise on this scale, our primer on hyperscalers’ off-balance-sheet AI debt explains the financing structures now underpinning the sector.

Part of a sector-wide capex surge

OpenAI is not spending in isolation. The forecast revision arrives in the same week that Alphabet raised its 2026 capital-expenditure outlook to $195–$205 billion and rattled its own shareholders, and it extends a build-out that includes OpenAI’s 10-gigawatt partnership with Nvidia and a broader hyperscaler capex boom that has become the defining feature of 2026’s technology tape. The financing has spilled into the bond market as well, as seen in Amazon’s $25 billion AI bond sale — a sign that even the cash-richest players are turning to debt to fund the compute race.

For OpenAI specifically, the spending clock is also ticking against a possible public listing. The company remains one of the most valuable private firms in the world, and — much like Anthropic’s confidential IPO filing — any eventual debut would put these infrastructure commitments in front of public-market investors who will price the losses more harshly than private backers have.

What it means

The lesson of the revision is that AI infrastructure budgets are not stabilizing — they are still ratcheting up. A $600 billion plan became a $750 billion plan in a matter of months, driven not by a strategic rethink but by the simple act of signing more capacity contracts. That tells you demand for compute, at least as OpenAI forecasts it, continues to outrun the supply it has already secured. For chipmakers, power providers, and data-center builders, the read-through is bullish: the anchor tenant of the AI economy is still writing bigger checks.

The risk sits on the other side of the ledger. $750 billion in committed spending only makes sense if revenue scales into it, and OpenAI’s 2025 numbers — roughly $13 billion of revenue against tens of billions in cash costs — leave an enormous gap to close. The company is effectively pre-committing to a decade of capacity on the bet that model demand, enterprise adoption, and pricing power all compound fast enough to justify it. If any of those falter, the commitments become the liability rather than the moat. That asymmetry is why Alphabet’s own capex guidance sent its stock down even on an earnings beat: investors are increasingly asking not whether the spending is large, but whether it will ever pay back.

Watch three things from here. First, whether OpenAI’s next revenue disclosures show the growth needed to service these commitments. Second, how much of the $750 billion is genuinely contracted versus aspirational — the distinction between a signed Oracle or Azure deal and a planning figure matters enormously. And third, whether the financing shifts further toward debt and off-balance-sheet vehicles, which would tell you the equity markets are nearing the limit of what they will fund with cash. The AI build-out has never lacked ambition; the open question in the back half of 2026 is who ultimately pays for it.