Injective has launched a perpetual futures market tracking rental prices for five Nvidia GPUs, offering up to 10x leverage against its new Super Intelligence Compute Index.
Summary
- Injective’s SICI perpetual market offers up to 10x leverage, with USDC used as margin.
- H100 rentals carry a 40% weighting in the index’s basket of five Nvidia GPUs.
- LO:TECH calculates the benchmark hourly using a seven-day average of weighted rental-price data.
- Index methodology changes will create separate versions without replacing the existing live series.
Injective said in its Oct. 26 announcement that the SICI market is live on mainnet, uses USDC as margin, and can be integrated into applications across its ecosystem. London-based capital markets firm LO:TECH builds and prices the benchmark using rental listings from compute marketplaces.
The company describes the product as a way for traders to take positions on computing costs and for AI developers and infrastructure providers to hedge their exposure. Its announcement identifies rising rental bills as a risk for developers and falling rental revenue as a risk for equipment providers.
In the same announcement, Injective said it plans to introduce another onchain financial product in the coming days, while describing the SICI perpetual itself as already available for ecosystem integration.
SICI combines five Nvidia GPU rental markets
The benchmark assigns 40% to Nvidia’s H100 SXM 80GB, followed by 25% to the B200 and 15% to the H200, according to the launch announcement. The A100 SXM 80GB and GeForce RTX 4090 each account for the remaining 10%.
In Injective’s description, the basket covers hardware used for large AI training jobs, model serving, fine-tuning, and lower-cost inference. The company includes the consumer-grade RTX 4090 because developers rent those cards to run open-source models and smaller AI workloads.
Each component keeps its stated weight within a methodology version, the announcement says. A change to the methodology would create a separate version alongside the existing series, leaving the live index’s historical calculation intact.
The five-chip product follows Injective’s earlier H100 rental-price contract. As crypto.news previously reported, the network introduced its Nvidia GPU derivative market in August 2025 through a collaboration with Squaretower.
Squaretower supplied hourly H100 rental-price updates drawn from compute providers for that earlier product. The announcement of the original contract described trading exposure to GPU rental rates, with Squaretower available through the Helix decentralized exchange.
Rental listings feed a seven-day index average
For SICI, LO:TECH gathers reference prices in dollars per GPU-hour from compute marketplaces, pooling on-demand and spot listings, Injective says. Each marketplace counts once for each chip, and observations pass validation and outlier checks before entering the calculation.
The methodology rebases each GPU’s reference price to 100 at the July 22, 2026 daily fixing. LO:TECH then combines the five normalized component levels using the stated weights, before publishing a trailing seven-day average of the hourly basket.
SICI is therefore quoted in index points, with the underlying inputs expressed as hourly rental prices, according to the announcement. LO:TECH calculates a reading every hour in UTC and uses the 19:00 UTC reading as the daily fixing.
Injective describes the calculation as reproducible from archived observations, saying “any print can be regenerated from archived data with an identical result.” The announcement names compute marketplaces as the source category for rental observations.
Across its other products, the network has offered exposure to both AI companies and computing infrastructure. Injective dates its AIX index perpetual to January 2025, describing a basket that combined AI-related tokens with equities including Nvidia and TSMC.
U.S. compute proposals address rental and inference costs
In the United States, Hyperliquid Policy Center has sought compute perpetuals through an Oct. 20 submission to the Commodity Futures Trading Commission, earlier coverage reported. The organization asked for regulated contracts tied to GPU rentals and the cost of running AI models.
The policy center argued that perpetuals could address recurring exposure for data-center operators, lenders and AI developers. Its submission supported dated futures for expenses with a known endpoint, such as a planned training run or scheduled capacity purchase.
For benchmark design, the group identified chip type, location, rental period, cluster size, service terms and interruptible access as relevant specifications. It distinguished publicly quoted hourly rentals from reserved capacity and multiyear agreements whose prices generally remain private.
The submission recommended independent price contributors, unusual-observation filters and position limits, alongside funding-rate safeguards for perpetual contracts. Those requests were policy recommendations to the regulator, as described in the earlier coverage.
GPU-backed lending has attracted stablecoin financing
Computing hardware has featured in lending arrangements as well as rental-price derivatives. Bullish announced a $100 million facility for USD.AI on Aug. 28 to finance loans secured by GPUs, according to earlier coverage of the companies’ announcement.
Developed by Permian Labs, USD.AI provides non-recourse financing secured by purchased computing equipment. Bullish said the facility would support middle-market AI infrastructure operators, with the financed hardware serving as collateral and no claim against the borrower’s other corporate assets.
USD.AI had previously announced a $98.1 million loan backed by 2,304 Nvidia B300 GPUs in June, while investors funded a separate $34 million facility secured by 768 B200 units. Together, those disclosed transactions covered 3,072 GPUs and $132.1 million in financing.
Bullish’s arrangement followed its venture arm’s $4 million investment in USD.AI in September 2025. Alongside the debt facility, the companies announced plans for sUSDai trading pairs and a market-making program, without specifying the listing date or trading pairs.
