I usually think a derivative on GPU prices was a gimmick for traders. Then I read how @injective built SICI, its new Super Intelligence Compute Index, and I think it's a thermometer for a cost most AI teams already carry with no way to hedge it. Here's the design 👇 → The index SICI tracks rental prices for five Nvidia GPUs, with the H100 weighted at 40% and the B200 at 25%. The data comes from compute marketplaces compiled by London-based LO:TECH. → The smoothing It updates hourly using a trailing seven-day average of rental prices, so one noisy hour doesn't move the benchmark much. → The market A perpetual futures market tracks the index, with USDC as margin and up to 10x leverage. → The two sides Injective says developers can hedge rising compute costs, while GPU providers can protect against falling rental revenue. Now the stakes. My read: compute is becoming an input cost for every AI product, yet its price moves with demand. A startup whose costs jump mid-quarter can't easily re-price. A provider whose rental rates fall loses revenue it counted on. Two sides with opposite risks is exactly what a hedging market needs, as long as the benchmark between them can be trusted. The contrast is what stuck with me: Most onchain derivatives track what traders already watch. This one tracks what builders already pay. To be fair, 10x leverage magnifies losses as much as gains, and a hedge only works if it is sized like one. This is not financial advice. My prediction: as AI spending grows, compute will get benchmarks the way oil did, and the benchmark that earns trust will be the one hardest to move.