Sam Altman offering $2M of OpenAI tokens to every startup in the current Y Combinator class is obviously smart. A powerful adoption loop inside the nursery of future unicorns—the equivalent of placing your currency directly into the bloodstream of the next economy.
The deal structure makes it sharper still: in exchange, OpenAI receives an uncapped SAFE, equity converting at the startup's first priced round. Tokens issued at near-zero marginal cost, traded for real-world stakes in the next generation of AI-native companies. That's something closer to seigniorage than to a marketing move.
Etymology itself has "token" as a symbolic artifact granting access or recognition. Long before crypto, tokens were already quasi-monetary instruments. AI tokens could become something similar—and the analogy with monetary systems starts to hold: inflation (token abundance), exchange rates between models, reserve currencies (GPUs), dependency structures, implicit monetary policies through pricing and quotas, trading across hyperscalers and token marketplaces, and now, computational seigniorage—issuing cognitive currency, collecting equity.
Classical seigniorage worked because the sovereign issued currency at a cost below its perceived value—that spread was profit, and the system held. AI tokens may be running the exact opposite: frontier labs are issuing cognitive currency while burning through investor capital to do it. The "currency" is subsidised. The economics aren't closing.
This matters because entire ecosystems—including hundreds of YC startups—are now building on top of a subsidised exchange rate. The dependencies form before the structure is sound. When pricing normalises, those dependencies don't disappear. They become vulnerabilities.
Classical money is neutral—it represents the value of work, while the work itself exists independently. A dollar paid for a task means a human performs the task. A token spent on the same task means no human does—the token emission is the work, and in performing it, it withdraws that cognitive activity from the economy altogether. Not a medium of exchange—a medium of displacement.
Models become central banks—ones that can't yet close their own economics. Hyperscalers, the commercial banking layer. Agents, compulsive consumers of cognitive currency. Power grids become the gold standard—the truly scarce resource behind everything. Benedict Evans just documented it: US datacenter construction spending is now overtaking office construction, with a multi-year power backlog in most markets. And whoever controls the hardware running on top of them extracts rent from every actor in the system, regardless of which lab wins.
VCs are currently filling the role no monetary system should leave to private capital: lender of last resort. That backstop won't hold indefinitely.
Maybe the monetary analogy is the wrong one. But if it is—what's the right one?
