An airline dispatcher knows why carrying extra fuel can make sense: buy it where the price is better, carry it to where it is needed, and weigh the saving against the cost of carrying it.
The analogy for an AI company is input-cost discipline. Compute is consumed to produce a service, while its price and availability can change independently of the price charged to customers.
Price certainty and access are different
A financial hedge can offset a price move. It does not, by itself, put a GPU at the buyer’s disposal when supply is tight. A delivered forward commitment addresses access as well as the agreed purchase price.
That distinction is central to Gantry’s proposed market: a specified device class and delivery window, attached to a claim that can be redeemed for machine time.
Every commitment has a carry cost
Buying forward is not free of risk. Demand may fall. Hardware requirements may change. A buyer can commit to capacity it later does not need. The operator can also fail to deliver.
Transferable claims offer a way to move a commitment to another holder, but a secondary market requires actual liquidity. Collateral and delivery rules address performance risk; they do not eliminate it.
Make the decision explicit
The useful lesson from aviation is to distinguish operating policy from a bet on prices. Understand the workload, the delivery specification and the period of exposure before deciding which capacity to commit to.
Gantry is building infrastructure for that decision: comparable claims, a common venue and a path from purchase to verifiable delivery.