OpenAI's $2M-tokens-for-equity YC deal, decoded

On May 20, 2026, OpenAI announced it would provide $2 million worth of API tokens to each startup in Y Combinator's Spring 2026 batch—totaling roughly $338 million in inference credits—in exchange for equity via an uncapped SAFE. The deal gives OpenAI portfolio exposure to about 169 companies while creating platform lock-in, as startups building on OpenAI's API face high switching costs to competitors like Claude or Gemini. Critics warn that founders risk giving up equity for tokens that may not lead to product-market fit, though the arrangement benefits token-heavy startups that would otherwise spend significant cash on inference.

On Tuesday, May 20, 2026, Sam Altman told a Y Combinator audience that OpenAI would invest $2 million worth of API tokens into every startup in the current YC batch, in exchange for equity. YC's Spring 2026 directory lists about 169 companies — roughly $338 million of inference, paid as credits rather than cash. TL;DR YC partner Tyler Bosmeny called it a "mic drop moment." OpenAI extends $2M of token credits to every startup in the Spring 2026 batch and the Summer 2026 batch, per follow-up reporting . YC managing director Jared Friedman confirmed to TechCrunch that the instrument is an uncapped SAFE. Three numbers anchor it. $2M per startup. ~169 startups in the current cohort. $338M total implied value at retail token prices. A SAFE is the standard YC instrument for early-stage companies that take money before they have a formal valuation. It converts into equity later, at the next "priced" round — usually a Series A. The word that matters is uncapped. A capped SAFE locks in a valuation ceiling for conversion. An uncapped one does the opposite: the higher the valuation at conversion, the smaller the slice of the company the investor receives. That cuts in the founder's favor. Discussion on X has floated the figure that this would amount to about 2% equity for OpenAI at a $100M conversion. Actual SAFE terms have not been published, so treat that as directional, not confirmed. Two layers. Portfolio exposure is the obvious one — OpenAI now has skin in the success of every company in the batch. The less obvious layer is platform default. A startup that ships on GPT and tool-calls the OpenAI Responses API does not casually re-architect onto Claude, Gemini, or Llama later. By the time $2M of credits run out, the abstraction layer in the codebase is OpenAI-shaped. As inference costs keep falling, the marginal cost to OpenAI of issuing those credits drops over time; the equity it took in exchange does not. Seed investor Jason Calacanis flagged this on X — "be careful, founders" — paired with a warning that OpenAI might observe what gets built and ship a first-party version. That risk is real, but a startup paying cash for OpenAI tokens is exposed to the same observation without the equity counterweight. The framing — "OpenAI invests in every YC startup" — is the marketing. The substance is tokens for equity at no cash outlay from OpenAI. Equity-for-services is not new in venture; the scale and the named platform are. Compare against the existing YC stack. YC takes 7% for $500K cash. Seed investors at the next round typically take ~20%. If OpenAI's stake settles in the 2% range floated on X, the cap-table math is real but not catastrophic — provided the $2M of inference converts into traction. The thing to track is whether OpenAI publishes its SAFE template. Two checks before signing. Take it if the product is token-heavy agentic loops, large-context retrieval, real-time voice and the startup would otherwise spend non-trivial cash on inference in the first 18 months. Trading equity for the AI infrastructure line item — at a stage when cash is scarcer than equity — is the bull case. Wait if the traction loop does not lean on inference, or if multi-model portability is a strategic constraint regulated industries, enterprise customers who require model choice . $2M of OpenAI-specific credits is worth less to a product that needs to run on Claude, Gemini, or local models on day one. The failure mode TechCrunch named: a startup burns its $2M token budget on experimentation, ends the batch without product-market fit, and has given up equity for the privilege. That outcome is worse than paying cash for the same tokens.