On May 20, Sam Altman revealed that OpenAI is offering $2 million in API credits to every company in Y Combinator's current batch, in exchange for equity. With 169 startups in the batch, that's $338 million in tokens flowing through SAFE agreements—a sweeping investment in an entire generation of American startups.
It's the same playbook AWS used years ago.
$2 Million in Tokens—What's the Catch?
The structure is straightforward:
- Each startup gets $2 million in API credits (at retail prices).
- Accepting the credits means signing an uncapped SAFE.
- The SAFE converts to equity in the startup's next round (typically Series A) at that round's valuation.
- The higher the startup's valuation, the smaller the stake OpenAI gets.
The uncapped SAFE is key. It lets OpenAI defer its investment decision until after the startup proves itself. Only if the company raises a Series A does OpenAI get shares at that valuation. If the startup fails, OpenAI only loses the tokens it already issued—and the marginal cost of those tokens is far below retail price.
In other words, OpenAI isn't really spending $338 million. It's spending the electricity, depreciation, and bandwidth of its own data centers for the compute these startups consume.
$338 Million on Paper, But Far Less in Reality
Retail prices are the surface number. The actual cost structure is very different (industry estimates):
- Per million input tokens: retail $5–$15, marginal cost $1–$3
- Per million output tokens: retail $20–$60, marginal cost $5–$12
At marginal cost, OpenAI's $338 million "investment" might actually burn only $60 million to $100 million. The rest is accounting—a price OpenAI set itself.
This is the core business logic of cloud credit subsidies: selling API credits at a discount is essentially using future revenue to offset current marginal costs. AWS ran this playbook for nearly 20 years, giving early startups thousands of dollars in cloud credits.
Locking a Generation of SaaS into Its API
Altman didn't hide the intent. YC is the most critical early-stage startup incubator in the U.S., with 150–200 companies per batch, two batches a year. If all 350 or so startups from this year's spring and summer batches integrate OpenAI's API into their core products over the next 18 months, switching costs become enormous:
- Switching to Anthropic? Prompts and tool calling need retesting.
- Switching to Gemini? Workflows and fine-tuning need redoing.
- Switching to DeepSeek? Compliance for overseas markets must be reassessed.
That's textbook vendor lock-in.
Jason Calacanis, early investor and All-In podcast host, publicly warned YC founders to be careful: OpenAI is their financier, their API supplier, and can see what they're building from API calls—three roles that are inherently conflicting. OpenAI's API terms of service state that enterprise plans don't train on data by default, but quality monitoring accesses metadata, which can reveal product shapes.
A Move Aimed at Anthropic and Google
The timing is telling:
- In early May, Anthropic launched Managed Agents at its Code with Claude event in London, clearly competing for enterprise clients.
- In mid-May, Google I/O unveiled Gemini Spark, with Gemini 3.5 Flash surpassing GPT-5.5 in benchmarks.
- On May 20, OpenAI immediately locked up an entire generation of 169 YC startups.
Instead of fighting for existing enterprise customers, OpenAI is buying the startups that will become enterprise customers in 18 months. Bezos did the same with AWS Activate. Why did AWS win the cloud war? Because from 2010 to 2015, every YC startup started on AWS. By the time they grew and needed to switch, migration costs were astronomical.
Altman is now doing the same for the LLM era—binding the next generation of unicorns to the ChatGPT API.
Should Founders Take the Deal?
In the short term, almost all YC founders will take it—free $2 million in API credits with no reason to refuse. In the long term, if 1–2 unicorns emerge from this batch in 3–5 years, OpenAI wins. The rest will fail before using all their tokens. It's a clear power-law bet.
As for the risk of OpenAI seeing core data, most early founders can't afford to worry about it. How many Calacanis's warning will sway remains to be seen in six months.
Sources: OpenAI offers $2M in API tokens to every Y Combinator startup for equity (Crypto Briefing); CocoLoop; Altman's $2 million OpenAI 'tokenmaxxing' offer to startups raises founder control concerns (American Bazaar); Sam Altman Just Offered Every YC Startup $2 Million in Tokens. Should They Take It? (Medium / MayhemCode)