A token used to just be an expense — you paid for it, you used it, it was gone. So it's strange to suddenly see them traded as a form of capital that can be exchanged for ownership in a company.
Last week Altman walked into a YC event and offered $2 million in OpenAI tokens to every startup in the current batch — all 169 of them — in exchange for an uncapped SAFE. Not cash, not advisory, just compute. Same instrument YC invented to do early-stage cash deals, except the asset on the other side of it is API credits.
Tokens are now liquid enough, and load-bearing enough, to substitute for venture dollars in a real deal. That's a category change. The thing you used to budget against is now the thing you can be funded in.
It reinforces something I was writing about a couple weeks ago — token usage is becoming the proxy for productivity. If your output as a startup is increasingly correlated with how many tokens you can put to work, then funding you in tokens is funding the actual fuel of the business. The most direct form of capital you can be given. Skip the cash, skip the conversion to compute, just hand over the compute.
The term floating around is "tokenmaxxing." Startups optimizing internal operations and product around heavy API usage. The minute usage becomes the metric people get measured on, though, the incentive isn't to use tokens well — it's to use a lot of them. A team that burns 10x the tokens to ship the same feature looks more productive on the surface than the team that found the efficient path. The first team also happens to be the better customer. There's a quiet conflict of interest baked into a structure where the infrastructure provider is also the equity investor and the metric of success is consumption of the infrastructure.
A few questions I keep turning over. If OpenAI sees the workflows, the prompts, the data, and the patterns of 169 startups all building on their API, what does that information asymmetry look like in three years? What's the line between platform and competitor when you're funding the platform's next round of training data? And if tokens are capital now, who else issues them — and what does the market for that look like once Anthropic, Google, and the rest run the same play?