Sam Altman Just Offered Every YC Startup $2M in OpenAI Tokens. Here's What That Trade Actually Costs.

What Altman put on the table

Source: Wikimedia Commons

On May 20, 2026, Sam Altman walked into a closed Y Combinator event and made an offer that rippled across Silicon Valley before the night was over. OpenAI would provide $2 million worth of API tokens to every startup in the current YC class in exchange for equity. YC partner Tyler Bosmeny, who was in the room, immediately called it a 'mic drop moment.' Altman confirmed the offer on X the same evening, writing that he was excited to see what would happen with 'tokenmaxxing' startups.

The offer is not cash. It is compute. Specifically, $2 million in OpenAI API credits that startups can burn through to build and run AI-heavy products. For early-stage companies where infrastructure bills can hit five or six figures before the product even ships, that allotment is functionally close to seed capital. The pilot covers both the Spring and Summer 2026 YC batches, meaning the deal touches the entirety of YC's newest cohort at once.

The structure: what an uncapped SAFE actually means

The deal is structured as an uncapped Simple Agreement for Future Equity, or SAFE. SAFE agreements are standard in early-stage startup financing and were originally pioneered by Y Combinator itself. They allow a company to receive investment immediately while delaying the calculation of the investor's actual ownership percentage until a future priced round, typically a Series A.

The 'uncapped' part matters. A cap sets a ceiling on the valuation at which an investor's stake converts into equity. With no cap, OpenAI's percentage is determined entirely by what the startup is worth at conversion. The higher the Series A valuation, the smaller the slice OpenAI ends up with. YC managing director Jared Friedman confirmed to TechCrunch that the conversion would happen at the next priced round. Analysis circulating on X suggests the deal could amount to roughly 2% equity if a startup hits a $100 million valuation, though the exact terms have not been made public.

The current YC cohort includes around 169 startups, which means the implied value of the total commitment lands near $338 million at retail API pricing. That number is large on paper but significantly softer in practice, for reasons that are central to how this deal is actually structured in OpenAI's favor.

The math OpenAI is running

OpenAI is pricing these tokens at retail developer rates, but the actual cost of serving them is a different number entirely. AI inference costs have been falling sharply. According to data from Ramp cited by research firm Artefact, the average cost per million tokens across major providers dropped from roughly $10 to $2.50 in a single year. Epoch AI's research points to even steeper declines when efficiency improvements are factored in. This means OpenAI is distributing something that gets significantly cheaper to produce over time, in exchange for equity stakes that become more valuable as the startups grow.

TechCrunch noted the asymmetry directly: what OpenAI gives away today could cost it very little to deliver six months from now, making the equity it receives in return look increasingly cheap. The deal works on a second level too. Every startup built on OpenAI's tokens is a startup that will not default to a competitor like Anthropic. Sam Altman framed the initiative as support for 'tokenmaxxing,' a term describing small teams that treat AI compute as a core operating resource rather than a line item to minimize. OpenAI is effectively betting that the startups leaning heaviest into AI usage will be the ones that succeed, and structuring the deal so that when they do, OpenAI profits from both the equity and the continued compute consumption.

For a company projecting over $14 billion in losses in 2026, acquiring equity across 169 early-stage companies at a cost that falls over time while simultaneously locking in a development pipeline for its API is a structurally attractive trade. The headline number of $338 million is not what OpenAI actually spends to honor it.

Vendor lock-in and the risks founders are glossing over

Source: Y Combinator

The criticism of this deal is not that it is bad on its face. It is that founders may underestimate what accepting it actually commits them to. The $2 million in credits is redeemable only with OpenAI. Founders who build their entire product stack around OpenAI's API pricing, rate limits, and model capabilities are making a platform bet at the earliest possible stage, before they have meaningful user data, before they know how their unit economics will scale, and before they can evaluate whether a competing model might serve their use case better.

Seed investor Jason Calacanis pushed back hard on X, warning that OpenAI gains visibility into what each startup is building. His concern: that OpenAI could replicate a startup's idea and fold it into ChatGPT. The counterargument is that this risk exists regardless of whether a founder takes the token deal, since building on any public API already exposes your product logic to the provider. But there is a meaningful difference between a supplier and a supplier that also holds equity in your company.

The uncapped SAFE structure also introduces complexity at the Series A. Investors coming into a priced round will see OpenAI on the cap table with an unknown final percentage and a financial interest in the startup's continued use of OpenAI infrastructure. That dynamic can complicate negotiations, particularly for startups in categories where OpenAI itself is an active competitor. Healthcare, legal tech, and coding assistants are all areas where OpenAI has built or is building products that overlap with the startup layer.

The tokenmaxxing bet and what it signals for the broader market

Altman's use of the term 'tokenmaxxing' was not accidental. It reflects a specific theory of how AI-native companies will operate: small teams, high automation, heavy API consumption, products built almost entirely on external model infrastructure rather than custom training. The deal is partly an investment in startups and partly a bet that this operating model is the one that produces the next generation of significant companies.

There is some evidence for the thesis. Meta reportedly had an internal leaderboard tracking employee token consumption, which went viral inside the company before being taken down. Uber burned through its entire 2026 AI budget in four months. ServiceNow hit similar walls. The pattern across large companies is consistent: AI access gets distributed, heavy usage gets celebrated, and the bill arrives before anyone has built the governance to manage it. Altman is betting that the startups who figure out how to build efficiently at high token volumes will be the ones that matter.

What this deal also signals is the degree to which compute is now functioning as venture capital. The broader trend is what some analysts are calling 'compute-for-equity,' where AI infrastructure providers offer credits in exchange for ownership stakes rather than deploying cash. If the OpenAI-YC pilot holds and produces successful outcomes, expect Anthropic, Google DeepMind, and others to respond with their own versions. The race to lock in the next generation of AI-native founders has clearly begun, and the currency being used is not dollars. It is inference.

The bottom line for founders evaluating the deal

• Founders already planning to build heavily on OpenAI should probably take it. The compute is real, the equity given up is modest if the company grows, and the upfront infrastructure relief is meaningful.
• Founders in categories where OpenAI competes directly, including productivity tools, coding assistants, and research interfaces, should think carefully before adding OpenAI to their cap table.
• The uncapped SAFE structure favors founders who raise at high valuations. If a startup struggles and raises a down round, OpenAI's equity stake converts at a more punishing rate.
• Platform risk exists whether you take the deal or not. Building on any external API creates dependency. The difference here is that your API provider will also hold equity and have a financial interest in your continued usage.
• The deal is structured for YC's Spring and Summer 2026 batches. If OpenAI extends it further, the compute-for-equity model may become a standard feature of AI-era startup financing rather than a one-off headline.

Sources