OpenAI Just Added $111B to its Cash Burn Forecast

According to internal financial documents cited by The Information, OpenAI now projects cumulative cash burn of $665 billion through 2030 — roughly $111 billion more than its previous estimate. In 2026 alone, the company expects to burn $25 billion. By 2027, $57 billion. Training costs through the end of the decade will approach $440 billion. Anthropic, for comparison, is targeting break-even by 2028. OpenAI won't get there until 2030, when it projects positive cash flow of $39 billion.

This is not a startup rounding error. This is a structural reckoning.

The Inference Problem Nobody Was Expecting

The single most destabilizing line in this story isn't the topline burn number — it's this: inference costs quadrupled in 2025. The day-to-day expense of simply running these models exploded because demand for ChatGPT and API services outpaced capacity. OpenAI had to buy compute at emergency prices. Adjusted gross margins dropped to 33%, against a target of 46%, and down from 40% the year prior.

Software companies with staying power typically hold gross margins above 70%. OpenAI, even in its most optimistic projections, now concedes it will miss its own 70% margin target by 2029, settling instead between 52 and 67%. That's the margin profile of a capital-intensive industrial business, not a software company.

The main beneficiaries of this spending, assuming OpenAI can sustain it, are Microsoft, Amazon, and Oracle — all of which hold long-term compute contracts. OpenAI is currently negotiating a funding round of over $100 billion at a $750 billion valuation, with SoftBank, Amazon, Nvidia, and Microsoft reportedly participating. The company ended 2025 with roughly $40 billion in cash.

Revenue Is Real — But the Gap Is Growing

To be fair, the revenue story isn't bad. OpenAI more than tripled revenue to $13.1 billion in 2025, beating its own forecast. It's projecting $30 billion in 2026 and $62 billion in 2027. Consumer subscriptions remain the biggest driver, with 910 million weekly active users — impressive, though below the one billion target the company set for itself. By 2030, OpenAI aims to reach 2.75 billion weekly active users and $150 billion in consumer revenue.

Enterprise is the bigger bet: OpenAI wants to grow B2B revenue from $2 billion in 2025 to $70 billion by 2030. API revenue is projected at $47.5 billion. Hardware and new products — including the forthcoming smart speaker — are expected to contribute just $15 billion.

The math isn't impossible. But the gap between the revenue and cost trajectories is widening with each revision cycle, not narrowing. This is the third time OpenAI has revised its projections significantly upward. Each correction has been larger than the last.

What This Means If You're Building on AI

For marketers and growth leaders evaluating AI investments, this story carries a practical warning: the cost structure of frontier AI is not stabilizing. It's compounding. Every tool, every API integration, every AI-native workflow you're building exists on top of an infrastructure that is, at this moment, economically unsustainable without continuous capital injection.

That doesn't mean stop. It means build with awareness of vendor risk. OpenAI's financial health is now a factor in your technology strategy, whether you've thought of it that way or not.

It also raises a broader question that goes beyond balance sheets: if the only viable path to AGI requires burning through hundreds of billions with no governance framework and no verified safety guarantees, at what point does "moving fast" become something society should formally object to? Altman himself this week called for international oversight analogous to nuclear regulation — while simultaneously accelerating.

The dissonance is real. And for anyone building a growth strategy on AI foundations, it deserves more than a footnote.

Winsome Marketing helps growth leaders build AI strategies that account for real-world risk, not just vendor roadmaps. Let's talk.