Sam Altman Just Offered $2 Million in Compute to Every YC Startup. The Lock-In Is the Product.

$800 million in API tokens for ~2% equity across 400 startups. Read that number again.

TechCrunch reports Sam Altman made the offer at a closed-door YC event on May 20. Every startup in YC's Spring 2026 batch gets $2 million in OpenAI API tokens via uncapped SAFEs. The total commitment: roughly $800 million in compute for what works out to about 2% equity per company.

This is the strategic counter-narrative to the cost stories above. If AI compute is expensive (story #1), the winner is whoever can subsidize it longest. And nobody has more capital available than the company that just filed a confidential S-1 targeting a $1 trillion IPO.

The mechanics are simple. Accept the deal, build on OpenAI's APIs, and your biggest early-stage cost disappears. Decline, and you're paying full freight against competitors who aren't. The incentive structure pushes every rational founder toward accepting.

Here's the catch. $2 million in API credits means $2 million in OpenAI API credits. Not Anthropic. Not Google. Not self-hosted. Every line of code these startups write, every agent pipeline they build, every production integration they ship will be wired to OpenAI. When the credits run out, switching costs will be massive. The lock-in isn't a side effect. It's the entire strategy.

I'm not sure how I feel about this. On one hand, eliminating compute costs for early-stage startups is genuinely useful. Building with AI is expensive, and $2M of tokens is substantial. On the other hand, 400 startups building exclusively on OpenAI's stack means 400 startups that can't switch to Claude, Gemini, or open-weight models without significant rework. That's not competition. That's ecosystem capture.

Fortune reports OpenAI's S-1 will reveal a negative 122% non-GAAP operating margin in Q1 2026, roughly $6.95 billion in quarterly losses, and $207 billion in additional capital needed through 2030. At those burn rates, $800M in YC subsidies is a rounding error if it locks in the next generation of AI-native companies.

For builders: if you're a YC founder, take the credits but architect for portability. Use abstraction layers. Keep your prompts provider-agnostic. Build the exit ramp before you need it.