Injective has launched the Super Intelligence Compute Index, a benchmark that tracks the rental rates of five NVIDIA GPUs and allows traders to take leveraged positions onchain through perpetual futures.

A Stock Index for GPU Rental Prices

The new index rolls five NVIDIA GPU rental rates into a single tradable number. It functions much like a stock index, except its components are hourly chip rental prices rather than company shares. Benchmark indices play the same aggregating role in traditional finance, collapsing many individual prices into one number that traders can track or build products around; here, the underlying is not company shares but the going rate for AI compute itself.

Exposure comes through perpetual futures, commonly known as perps — derivatives with no expiry date, and a contract type that grew up in crypto markets, where it has become one of the most heavily traded instruments. Traders can hold a long or short position for as long as they keep it funded.

The proposition is straightforward: go long if you expect the cost of AI compute to rise, or go short if you expect it to fall. Companies that purchase compute could also use a position to offset swings in what they pay. The underlying signal carries its own weight: as AI developers increasingly rent chips rather than buy them outright, rental rates serve as a live gauge of how tight the compute market is.

The H100 Market Paved the Way

The launch builds on Injective's earlier work in trading compute prices. On August 18, 2025, the network introduced a perpetual futures market tied to tokenized NVIDIA H100 GPU rental rates. The research describes it as the first onchain perps market built around that specific input.

That market runs on Helix, Injective's decentralized exchange, and reportedly allows leverage of up to 5x. By late 2025, it had cleared more than $203 million in cumulative trading volume.

Pricing data for the H100 market is supplied by Squaretower, which provides a custom decentralized oracle that refreshes rental rates every hour. An oracle serves as the bridge between the outside world and a blockchain: smart cannot look up GPU rental prices on their own, so external data must be fed in, and the reliability of the entire market depends on how well that is done.

A Pattern of Exotic Bets

The new index is part of a broader series of specialized products. On January 30, 2025, the network launched the AIX AI Index perpetual, which blends AI-related crypto tokens with traditional equities such as NVIDIA. AIX let traders bet on the AI trade as a theme. The H100 market then narrowed the focus to a single chip's rental price, and the Super Intelligence Compute Index widens the lens again to cover five GPUs instead of one — a progression that keeps circling the same underlying: the price of AI compute.

What This Means for Traders and Compute Buyers

The research frames the H100 product as serving both compute buyers and cloud service providers — two groups positioned on opposite sides of the same price. A startup training models fears rising rental costs, while a provider leasing out GPUs fears falling ones. In theory, a liquid derivatives market lets each lock in some certainty. Compute is one of the defining cost lines of the AI industry, so a tool for managing its price — much as airlines hedge jet fuel — addresses a planning problem that extends well beyond crypto trading.

Whether that theory holds depends on liquidity. More than $20.3 million in cumulative H100 volume is a start, but serious corporate hedging would likely demand far more depth. Spreading the index across five GPUs could help on that front, as a broader benchmark may appeal to more participants than a single-chip market.

The Risks Worth Watching

Oracle design is a key pressure point. Unlike Bitcoin or Apple stock, GPU rental prices are not quoted on a single public exchange, and how the index sources, weights, and updates its five rates will shape how much traders trust it.

There is also the question of whether compute behaves like a commodity at all. GPU rentals can vary by provider, contract length, and region, and compressing those differences into one tradable number involves judgment calls. What happens next will be measured in practical terms: how much volume accrues to the five-GPU index relative to the single-chip market, how the oracle holds up when rental rates move quickly, and whether genuine hedgers begin appearing on both sides of the trade alongside directional traders.