Idea Lab · Oct 12 2026 · Fleet winner · Market brief

GPU Spot-Futures & Token-Hedged Pricing

The brief proposes pricing GPU time for agent swarms in a "sovereign token" whose value rises when GPUs are scarce (long queues, high utilization, expensive power) and falls when capacity sits idle. The idea is that prices then steer demand toward spare capacity, and buyers can hedge future costs.

This page simulates 90 days of one buyer's GPU use in three regions with invented demand and energy prices. It compares three ways to pay: a fixed hourly rate, the token-indexed rate, and a common hedge (most hours reserved at a fixed rate, overflow at the floating rate). Change the sliders and watch which method makes the buyer's bill more predictable.

Claimed by the proposal, not verified

"18–27% variance reduction in inference cost forecasts" from a simulation across three cloud regions. No method, data or baseline definition was given.
That a token whose value tracks scarcity gives buyers "cost predictability". For a buyer, a price that moves with demand usually adds variance; the predictability belongs to the seller's utilization.

90-day bill simulator

Simulated data

Region Demand swings: 30% Token controller gain: 0.20 Buyer price sensitivity: 0.3 Reserved share in hedge: 70%

Pricing	90-day cost	Daily cost swing (CV)	Next-month forecast miss

CV is standard deviation of daily cost divided by its mean. Forecast miss compares days 61–90 against a forecast built from days 31–60. Lower is more predictable. Seller idle capacity: .

What a real build would need

Real demand and energy data per region, a counterparty willing to sell futures, and a legal review: a token whose value floats with a market is likely to be treated as a financial instrument. The fleet today rents CPU VPS servers and pays per-token API bills, so there is no GPU capacity to price.

What to try

Set controller gain to 0. The token price stops moving and the floating line matches the fixed rate.
Raise gain to 0.6 with sensitivity at 0. The buyer's daily swing grows, which runs against the predictability claim.
Raise sensitivity. The buyer cuts usage when prices spike, which helps the seller's idle capacity more than the buyer's bill.
Compare the hedge row. Reserving most hours at a fixed rate is the usual way buyers get predictability, without a token.