Layer-1 blockchain Injective has brought a new benchmark for artificial intelligence compute costs onchain. On October 26, 2026, the network activated the Super Intelligence Compute Index (SICI), a perpetual futures market that follows the hourly rental rates of five Nvidia (NASDAQ: NVDA) graphics processing units.
Traders can gain exposure to shifts in the price of the hardware that powers advanced AI models, with leverage available up to 10 times and USDC serving as margin.
The index aggregates rental data for a weighted basket of chips.
The Nvidia H100 SXM 80GB accounts for 40 percent of the composition and remains the dominant component, widely used for training and serving large models.
The B200, a Blackwell-generation datacenter GPU suited to large-scale training and inference, carries a 25 percent weight.
The H200, offering additional memory for long-context workloads, holds 15 percent.
The prior-generation A100 SXM 80GB and the consumer-oriented GeForce RTX 4090 each represent 10 percent.
The inclusion of the RTX 4090 reflects the reality that inference, fine-tuning, and open-source model serving frequently rely on more accessible cards rented by the hour, while frontier training concentrates on higher-end datacenter silicon.
Pricing originates from LO:TECH, a London-based capital markets firm that sources reference levels in dollars per GPU-hour from compute marketplaces.
On-demand and spot listings are combined, with each marketplace counted once per chip and observations filtered through validation and outlier controls. Individual GPU series are rebased to 100 as of the July 22, 2026 daily fixing.
The five normalized levels are then combined at the fixed weights, and the published index value is a trailing seven-day average of that hourly basket, expressed in index points.
Updates occur hourly in UTC, with the 19:00 UTC print serving as the daily fixing.
Fixed weights ensure transparency: any movement in SICI can be traced to changes in one or more of the underlying rental rates, and historical prints can be reconstructed from archived data.
Future methodology adjustments would appear as separate versions rather than overwriting the live series.
This product expands earlier efforts on the same chain.
In 2025, Injective launched a single-chip perpetual tied to H100 rental rates, developed with Squaretower and described as the first onchain market of its kind.
Earlier still, the network introduced an AI-themed index perpetual that blended crypto tokens with traditional equities.
SICI merges the index approach with direct compute pricing and broadens the basket across hardware generations and use cases.
The market addresses a practical gap.
AI laboratories secure GPU capacity years in advance, while smaller developers compete for remaining supply.
Rental prices fluctuate with demand, hardware availability, and competition among cloud providers.
Many organizations developing models rent capacity from specialized providers rather than purchasing chips outright.
Rising rates increase operating costs for developers; falling rates pressure revenue for infrastructure suppliers.
By converting a diversified set of rental prices into a single onchain reference, SICI allows directional speculation on compute costs while creating a potential hedging instrument for both sides of the market.
The perpetual is live on Injective mainnet and open for integration by ecosystem applications, so users can access it through various interfaces without each project building a market independently.
The launch sits alongside other real-world and AI-related offerings on the network, including perpetual exposure to certain private AI companies.
Injective has indicated that an additional onchain financial primitive linked to the theme is expected in the near term.
As AI infrastructure expands and electricity use at related data centers grows, instruments that let participants price and manage compute risk are likely to attract further attention. SICI represents one early step toward treating GPU rental rates as a tradable economic input rather than an opaque operational expense confined to cloud marketplaces.
