OpenAI's Leaked Deck Says It Will Burn $278 Billion by 2030. Here's What That Means If Your Product Runs on Its API
A private OpenAI presentation, obtained by the Financial Times and reported September 18, lays out the company's own math on its next five years: $278 billion in cumulative negative free cash flow between 2026 and 2030, funded by a bet that revenue climbs from $36 billion this year to $350 billion by 2030. Compute spending, roughly $856 billion of it, is the single biggest line item behind the burn. If you build anything that calls the OpenAI API, this deck is the closest thing you'll get to seeing the actual bet your dependency rests on.
The numbers, without the spin
The deck reportedly shows OpenAI's cumulative negative free cash flow narrowing slightly from where it stood in May, when the internal estimate was $305 billion through 2030. July's model releases lifted annualized revenue by about 20%, trimming $27 billion off the projected burn. That's a real improvement, and it's worth saying plainly: this isn't a company in freefall, it's a company running an extremely aggressive, extremely expensive growth plan and reporting that the plan is currently tracking slightly better than its own worst case from a few months earlier.
The revenue target is the other half of the story. Going from $36 billion to $350 billion in four years is a roughly 10x jump, and the deck was reportedly part of a private financing conversation tied to a computing deal, with investors reportedly discussing a valuation north of $1.2 trillion. Whether that revenue curve holds is the entire question the next four years will answer, and it's not a question anyone building on top of the platform gets a vote on.
Why this matters more than a typical "startup burns cash" story
Plenty of companies run at a loss while scaling, that alone isn't news. What makes this different is the scale relative to how deeply embedded OpenAI's API already is in other people's businesses. A $278 billion bet isn't a side project you can watch fail from a safe distance if you've built a product, a workflow, or a customer-facing feature on top of GPT models. The compute spend figure, $856 billion, is the number that actually determines what happens to you: it's the cost structure that current API pricing is subsidized against, and it assumes continued access to capital at a scale most companies never touch.
I don't think this deck means OpenAI is in trouble. I think it means OpenAI's leadership is telling investors, in a document meant to be taken seriously by people writing very large checks, that the current growth-at-a-loss strategy has a defined shelf life measured against a specific revenue target. That's useful information. It's also the kind of information that usually doesn't reach the people actually building on the platform until pricing already changed.
What actually happens if the curve slips
Nobody outside OpenAI's finance team can say with certainty what happens if 2027 or 2028 revenue comes in under the deck's projection. But the general shape of what companies do when a growth bet underperforms its financing case is not mysterious: pricing tightens, free and low-tier access gets pruned first, and enterprise contracts get prioritized over the long tail of smaller API consumers who are easiest to reprice without losing much revenue. None of that is unique to OpenAI. It's what happens to any subsidized platform when the subsidy math stops closing.
The comparison that's useful here isn't "OpenAI versus a startup that might fail." It's "OpenAI versus any other capital-intensive infrastructure provider you've ever built on." Cloud pricing, ride-share subsidies, food delivery commissions, they all followed a similar arc: aggressive pricing while chasing growth, then a correction once the growth math needed the correction to keep investors comfortable. The $278 billion figure is the clearest public evidence yet of how big that eventual correction could be if the revenue side doesn't land close to $350 billion.
What I'd actually do
I don't build anything where being wrong about OpenAI's pricing stability in 2028 would sink the business, and after reading this, I'm glad that's true. My actual approach is the same one I use for any single-vendor AI dependency: nothing I ship treats a specific model provider's current pricing as a permanent input. Every place I call OpenAI's API sits behind an interface I could redirect to Anthropic, an open-weight model, or a different provider without a rewrite, and I budget as if API costs for the tier of model I actually need could reasonably double over the next two years, not stay flat.
I'm not moving anything off OpenAI today. The revenue trajectory in this deck, if it holds even roughly, argues for a company that keeps improving its offering rather than one cutting corners to survive. But "if it holds" is doing real work in that sentence, and a $278 billion bet is exactly the kind of number that should make you build the exit ramp before you need it, not after.
Where I could be wrong: OpenAI has beaten skepticism about its financial trajectory before, and a 20% revenue lift from one release cycle suggests the growth curve has real momentum behind it, not just optimistic modeling. If GPT releases keep landing the way July's did, $350 billion by 2030 might look conservative in hindsight rather than aggressive. I'd rather have built the interface layer for nothing than need it and not have it.
Author
Lukas
@lukcombinator