📊 GPU rental rates got leveraged first; the regulated version is still stuck in a drawer

On October 26, Injective, a Layer 1 blockchain, launched SICI perpetuals on its mainnet. The contracts track rental rates for five NVIDIA GPUs, use USDC as margin, and offer up to 10x leverage. On the same day, CME’s computing-power futures were still marked “Pending Approval.”

How is the index compiled? H100 SXM 80GB accounts for 40%, B200 25%, H200 15%, and A100 SXM 80GB and RTX 4090 each account for 10%. London-based LO:TECH pulls the listed hourly rate for each GPU from computing-power marketplaces, taking just one quote per GPU from each marketplace. It runs validation checks, removes outliers, normalizes rates to 100 using the daily settlement price on July 22, 2026, then combines them according to their weights. It’s calculated hourly; the 19:00 UTC reading is the daily settlement price, while the figure reported publicly is a seven-day moving average. Any change to the methodology gets its own new series; the old one remains unchanged. Every reading can be reproduced exactly from archived data.

CME’s version follows a familiar playbook. Each contract represents 730 GPU hours—the equivalent of renting one GPU for a full month—and is cash-settled against Silicon Data’s rental-rate index. The minimum price increment is $0.01 per GPU hour, or $7.30 per tick. The contracts, coded GPU1 and GPU2, were originally set to launch on October 5. On September 21, the U.S. Commodity Futures Trading Commission extended its review to November 9, citing the market’s fragmentation and vulnerability to manipulation.

Here’s what the two have in common: nobody gives you a GPU. Physical delivery doesn’t exist in the regulated version either.

So this isn’t a contest about compliance. It’s about pricing power. Whoever first turns rental rates into a settleable number gets to set the benchmark. The on-chain version has run ahead by half a lap.

Why are people rushing in? Cash flows at computing-power providers are coming under scrutiny. Gartner’s October 14 forecast was that more than 80% of neoclouds won’t make it to independent operation by 2030. On October 25, Elon Musk said more than 200 neoclouds would face “guppy-level death and consolidation.” Lenders are starting to favor companies whose revenue doesn’t depend solely on data centers. Those renting GPUs fear prices rising; those renting them out fear prices falling. Both want to lock in rates.

The gap is in what the contracts track. This index bets on this month’s rental rate. It says nothing about what that GPU will be worth three years from now. Yet it’s residual value plus debt maturities that could crush neoclouds—and nobody has built an index for that yet.

The regulated route is still moving, too. On October 20, Hyperliquid’s Policy Center filed a petition with the CFTC, asking it to allow contracts linked to GPU rental prices and inference costs. It was signed by CEO Jake Chervinsky and senior counsel Brad Bourque.

As I write this, six BTC price feeds range from $79,689 to $79,826, up 0.1% over 24 hours. ETH is at $2,310.

For an asset to become an asset class, two things matter: someone has to be willing to bet on its price, and someone has to be willing to quote a price for it every day. GPUs have the first part. The quote sheet lists rental rates and nothing else. The depreciation column is blank.

$BTC $ETH

#中本聪国际社区Baoluo币商资本 #宏观经济 #AI computing