OpenAI is offering every startup in Y Combinator's Spring 2026 batch — roughly 169 companies — $2 million in API tokens apiece in exchange for equity, structured as an uncapped SAFE. Sam Altman pitched the deal at a YC event on Tuesday night, and the implied program value lands at an estimated $338 million at retail token pricing across the cohort.
The critical detail for anyone reading this as a capital event: OpenAI is not writing checks. It is converting inference capacity — its own marginal cost — into equity positions across an entire accelerator class.
The structure
Each company receives $2 million in OpenAI tokens, explicitly framed as AI inference credits rather than cash. The instrument is an uncapped SAFE (Simple Agreement for Future Equity). "It will convert in the next priced round, which is typically the Series A," said YC managing director Jared Friedman. Because the SAFE is uncapped, founders benefit at higher valuations — the equity OpenAI receives shrinks as the conversion mark climbs. One unverified estimate pegged the stake near 2% at a $100 million valuation.
Why credits beat cash
For OpenAI, the economics are asymmetric. Tokens cost it compute, not balance-sheet cash, yet convert into priced equity in the next generation of AI-native products. The playbook rhymes with the cloud-credit programs AWS and Google Cloud ran for a decade to lock early-stage workloads onto their infrastructure — except here the credits buy ownership, not just usage.
The strategic payoff is the API moat. Every team that builds its core product on GPT models inherits switching costs: re-prompting, re-evaluating, and re-architecting against Anthropic's Claude or Google's Gemini becomes progressively harder the deeper the integration runs. OpenAI also gains a real-time window into what 169 funded teams are building.
What builders and buyers should watch
Investor Jason Calacanis cautioned founders to "be careful," flagging that equity surrendered for credits — rather than cash — can quietly erode long-term ownership. Critics raised three concrete risks: vendor lock-in, founder dilution, and OpenAI gaining visibility into emerging ideas it could later compete with directly through its own product surface.
There's also a valuation question with no clean answer: should $2 million in metered inference credits convert into equity at the same terms as $2 million in venture cash? Credits expire, get rate-limited, and are priced at retail — not the discounted rate a high-volume buyer would negotiate.
The takeaway for engineering leaders is architectural, not financial. Free inference is genuinely useful runway in a tightened funding market, but teams that wire business logic straight into one provider's endpoints are trading a short-term subsidy for a long-term dependency. The hedge is boring and effective: an abstraction layer or routing tier that lets you spend the credits without making GPT load-bearing. The startups that treat this as cheap compute — not a foundation — keep their optionality.
