Here is what a token is, in plain terms. When you type a message to an AI tool and it responds, every word, part of a word, and punctuation mark gets broken down into small units called tokens. AI companies charge you based on how many of those units get processed. The more complex the task, the more tokens consumed, and the higher the bill.
For a while, this felt manageable. Prices per token were falling fast. But total bills have been rising anyway, because companies are using far more AI than they planned. A Deloitte analysis found that enterprise AI spending has moved from around $1.2 million per year in 2024 to $7 million in 2026. Meanwhile, AI providers are shifting from flat monthly fees to usage-based billing. One developer saw their projected monthly cost rise from roughly €67 to €966 after such a shift, with a single pricing model change.
That kind of unpredictability is what futures markets are built to solve. A futures contract is simply an agreement to buy or sell something at a fixed price on a future date. Airlines do this with jet fuel. Manufacturers do it with steel and copper. The principle is: lock in a price today so you can plan tomorrow.
CME Group, the world's largest derivatives exchange, announced plans in May to launch futures contracts tied to GPU rental costs, partnered with a firm called Silicon Data. Within days, the owner of the New York Stock Exchange announced the same, partnered with a company called Ornn. Both sets of contracts are still pending regulatory approval.
Meanwhile, China's Shanghai Futures Exchange is taking a different angle. Rather than targeting the cost of the hardware itself, it is working on contracts tied to AI token prices: what companies actually pay when they use AI services. That is a more direct hedge for any business whose operations depend on AI tools.
The H100 one-year lease price jumped 38% between October 2025 and March 2026. GPU rental prices vary widely across providers and regions. That volatility flows directly into what AI companies charge for their services, which flows directly into what you pay on your AI invoice.
For most businesses right now, there is no way to lock in AI costs in advance. You take whatever the market gives you each month. These new financial instruments would change that, at least for larger organizations with the sophistication to use them.
There are real complications. Unlike oil sitting in a storage tank, computing power is consumed the moment it is used. You cannot warehouse it, and you cannot take delivery of it through an exchange. The contracts being designed are cash-settled, meaning they pay out based on price differences rather than delivering actual hardware or AI usage. That makes them useful for hedging cost exposure, but imperfect as a direct lock-in.
The fragmented market is another problem. GPU prices vary by chip type, cloud provider, region, and contract length. A single futures contract referenced against one benchmark index may not closely match the specific costs a given company faces. This is called basis risk, and it is a known limitation of any commodity hedge.
The broader meaning is this: AI compute is going through the same transformation that oil went through in the early 1980s, when energy price volatility became serious enough that financial markets built formal infrastructure around it. That process took years, and the early contracts were messy. But once it happened, companies could plan around energy costs in ways they simply could not before.
For a business operator today, the most practical implication is not to rush into these markets, since the products do not even exist yet. The implication is to start treating AI costs the way a good CFO treats energy costs: as a variable with real financial exposure that needs monitoring, forecasting, and eventually, active management.