OpenAI Is Buying YC Distribution With Compute, Not Cash

Sam Altman announced a $2 million token deal for every Y Combinator startup. The real question isn't whether it's generous — it's what OpenAI is actually buying.

On May 20, at Y Combinator's Demo Day, Sam Altman dropped what sounded like a founder's dream: OpenAI would invest $2 million in API tokens in every startup in YC's current batch — roughly 169 companies — via an uncapped SAFE that converts at the Series A. The headline number was $338 million. Sam Altman announced it. Y Combinator confirmed it. Tyler Bosmeny, a YC partner, put the cohort size on the record. The summer batch got the same offer, with a May 25 deadline.

Every outlet covered it as a distribution play — OpenAI locking up the most concentrated group of early-stage AI companies on the planet.

But look at the structure of the deal and a different story emerges.

The tell is in the fine print

YC's standard SAFE includes a most-favored-nation clause — it guarantees founders that if YC offers better terms to another investor in the same round, the current investor automatically gets those same terms. It's a founder protection. The OpenAI deal does not include an MFN clause. That means OpenAI negotiated actively to exclude it.

Think about what that signals. If you were OpenAI's corporate development team and you wanted purely to build loyalty and goodwill among YC founders, you'd take the standard template. It's a proven trust-builder, it's cheap to include, and it makes founders feel protected. Excluding it means OpenAI wanted flexibility — the ability to structure future deals differently without being locked into parity with other investors.

Why would a company spending $338 million in face-value credits want that flexibility? Because the equity economics are not the point. The credits are near-zero marginal cost inventory. OpenAI has already paid for the GPUs. What it is buying with this deal is something that doesn't appear on any term sheet.

What OpenAI is actually acquiring

Every YC startup that accepts the offer will run its product, its experiments, and — in many cases — its core workflows through OpenAI's API. That gives OpenAI visibility into what the cohort is building: which verticals are attracting founders, which technical approaches are gaining traction, which use cases are generating real inference demand. It is a proprietary intelligence map of the next generation of AI applications, refreshed continuously, delivered by the founders themselves.

Jason Calacanis warned founders explicitly on LinkedIn: taking the deal gives OpenAI information rights. "They watch which ones break out then rip it off, buy it out, or out-compete them." It is a serious warning and it is being cited across founder forums as a reason to think twice.

But Calacanis may have the mechanism wrong.

What the API policy actually says

OpenAI's developer documentation and data policy page state explicitly: "By default, we do not train on any inputs or outputs from our products for business users, including ChatGPT Team, ChatGPT Enterprise, and the API." This has been the company's policy since March 2023. API customers are opted out of model training by default — they must actively opt in to share data for improvement purposes.

That means the specific fear — that OpenAI will study YC companies' prompts and use them to train competitive products — is not supported by the public API policy. If a YC startup accepts the $2 million credit line and does nothing else, their API traffic is not flowing into OpenAI's training pipeline.

The information-rights risk is real. But it is contractual, not data-leakage. It lives in the SAFE — the equity agreement — not in the API terms of service. And that document has not been made public.

The unanswered question

What does the SAFE actually say?

Neither YC nor OpenAI has published the term sheet. No founder who has signed it has quoted the specific data-rights language on the record. The concern that Calacanis and others have raised — that OpenAI negotiated special visibility into how the cohort uses its models — is entirely plausible based on the deal's structure and the exclusion of the MFN clause. But it is currently unverifiable.

The reporter who reads that agreement — or finds a founder or lawyer willing to quote it — will answer the question that every other outlet has left open: what did OpenAI actually buy?

What we know for certain

OpenAI is spending a headline $338 million to place $2 million in near-free compute credits in front of 169 AI startups, simultaneously. The deal converts at Series A — meaning OpenAI gets equity at whatever valuation the next investor sets, not a price OpenAI set itself. The summer batch is included, with a May 25 deadline.

For the startups that take it — and most probably will, because $2 million in credits at seed stage is genuinely transformative for an AI company — the deal is real value. Compute that would cost hundreds of thousands of dollars per year at retail is now free for a meaningful window. Founders can run experiments that were previously cost-prohibitive: large-scale agentic workflows, heavy inference loads, training runs that would have required a dedicated compute budget.

The equity dilution from an uncapped SAFE at Series A is typically modest. For a pre-revenue AI startup, cheap compute may be worth more than the small equity stake given up.

The second-order picture

If this model works — if OpenAI demonstrably gains strategic intelligence value from watching the cohort use its models — expect Anthropic and Google to attempt similar deals. The compute-for-equity swap is a playbook that any well-capitalized AI lab with excess capacity could run. YC is the most visible test case, but it will not be the last.

The story is not that OpenAI is being generous to founders. It is that compute surplus is being converted into ecosystem intelligence at a scale that no traditional VC can match — because no traditional VC has near-free inventory sitting on a GPU cluster.

The question for founders is not whether to take the credits. For most, the economics make sense. The question is what they are signing in the SAFE — and whether anyone is going to read it before the May 25 summer deadline passes.

Sam Altman | Tyler Bosmeny | Y Combinator | Jason Calacanis | Crypto Briefing | OpenAI API Data Policy | TechCrunch