Token futures
Enterprise API contracts are starting to look like airline seats. Buy in bulk, lock in a rate, and what you actually use rarely matches what you bought — marketing burns hot in Q4, engineering goes quiet in August. The quota is annualized; the usage isn’t.
The natural result is a secondary market. It’s already happening: Slack posts offering unused credits, the same gray-market activity you see with AWS reserved instances at end of billing cycle. Someone has tokens they won’t burn; someone else is getting rate-limited. They should find each other.
The technical problem is API key sharing — you can’t hand credentials to a buyer. Provider buy-in solves this, and eventually providers have incentive to build an official resale mechanism rather than watch informal markets grow outside their control. AWS Savings Plans are the template: commit to a 12-month block, get a discount, sell back what you don’t use.
Then it gets interesting. Once there’s a spot price for token quotas, you have the inputs for futures contracts. A company burning $500K/month on inference has real budget exposure to model repricing — futures let you hedge that. Buy 100 million tokens for Q3 delivery at today’s price; if Anthropic changes rates, you’re covered. This is how commodity markets always develop. Carbon credits, electricity, bandwidth — anything fungible and measurable at scale eventually sprouts derivatives.
The furthest-out version is cross-provider arbitrage: swap excess Claude tokens for Gemini credits, trade into whatever’s cheapest for your workload next quarter. The settlement layer doesn’t exist yet. The incentive to build it grows every quarter.
None of this is happening next year. But at some point committed quotas and actual usage diverge enough to matter, and when they do, someone builds the market. They always do.