Injective has opened a perpetual futures market on the cost of renting Nvidia GPUs. The layer-1 blockchain said in a blog post on Monday 26 October that its Super Intelligence Compute Index (SICI), a basket of hourly rental prices for five Nvidia chips, is live on Injective mainnet as a perpetual market, margined in USDC, with up to 10x leverage. The index data comes from LO:TECH, a London market-data firm.
This GPU futures market is thin on public numbers so far. Neither Injective nor Cointelegraph, which reported the launch the same day, gave a ticker, trading volume, open interest or funding rate for the market. Injective also did not say how positions settle or which oracle feeds the price on chain. Cointelegraph’s page showed no INJ price either.
The market widens an older one. In August 2025, Injective listed an onchain market tied to the hourly rental price of a single chip, the H100, built with Squaretower. SICI moves to five chips and a published methodology, and swaps the data supplier for LO:TECH.
How the GPU rental price index is built
SICI sits on top of the LO:TECH Compute Index. Injective describes LO:TECH as a London capital-markets firm that runs market making and institutional market data across more than 80 venues, Coinbase, Kraken and OKX among them. For the compute index, each GPU enters at a LO:TECH reference price in US dollars per GPU-hour. On-demand and spot listings from rental marketplaces are pooled, each marketplace counts once per chip, and observations go through validation and outlier filters, according to the blog post. Neither source names the marketplaces.
Each chip is then rebased to 100 at the daily fixing of 22 July 2026, and the five levels are combined at fixed weights. The H100 carries the largest share.
| Nvidia GPU | Weight in SICI |
|---|---|
| H100 SXM 80GB | 40% |
| B200 | 25% |
| H200 | 15% |
| A100 SXM 80GB | 10% |
| GeForce RTX 4090 | 10% |
Injective says the RTX 4090, a consumer card, is in the basket so the index covers the whole compute curve rather than frontier training alone: it stands for inference and for smaller builders running open-source models and agents. The weights are fixed for each methodology version; a change would come out as a new version published next to the old one, not as an edit of the live index.
The published figure is not the latest hourly price. “The published index is the trailing seven-day average of that hourly basket, quoted in index points,” Injective writes. It prints every hour, on the hour, UTC, and the 19:00 UTC print is the daily fixing. Injective says any print can be regenerated from archived data with an identical result, which would let an outsider check a disputed value.
What a perpetual on a price index means
A perpetual future is a contract with no expiry date. Two sides agree to exchange the change in a reference price for as long as they keep the position open. Here the reference is not a coin but SICI. A trader who goes long gains if the contract price rises, which should happen if GPU rental costs climb, and loses if they fall. A short position works the other way. Nobody rents, buys or delivers a GPU. Profits and losses are paid in USDC.
On most crypto venues a perpetual stays close to its reference through periodic funding payments between longs and shorts. Injective has not published how funding works for this particular market. Our glossary entry on perpetual futures explains funding, mark price and margin in general terms.
Who Injective says the market is for
Injective pitches two uses. AI developers, who often rent computing power from providers such as CoreWeave or Nebius instead of buying chips, could take a position that gains if rental prices rise, offsetting a higher bill. GPU providers could take the opposite side to protect themselves against falling rental income. The blog speaks of “new opportunities for AI developers and infrastructure providers to hedge their economic exposure”.
Those are Injective’s claims. Neither source shows a developer or a provider actually using the market that way, and with no volume or open interest published there is no way to judge from outside whether the market is deep enough to hedge anything. For most users the product is, in practice, a way to bet on the direction of AI compute costs.
Injective has listed similar markets before: an AI index perpetual (AIX) in January 2025, mixing AI tokens such as Bittensor and Virtuals with shares such as Nvidia and TSMC, and, in October 2025, pre-IPO perpetuals on OpenAI, SpaceX, Anthropic and Perplexity priced with data from Caplight and SEDA. The limits of that second kind of product are in our piece on Anthropic’s IPO filing and the crypto venues already quoting it.
Wall Street is building its own GPU futures
Injective is not alone in trying to turn compute into a tradable price. On 19 May 2026, Intercontinental Exchange and Ornn announced a suite of GPU compute futures based on the Ornn Compute Price Index, with series for the H100, H200, B200 and RTX 5090. ICE says that index tracks live-traded spot prices for GPU compute. The contracts would be in US dollars and cash-settled, and the release says they “will be launched subject to regulatory approval”. Five months later, no launch date has been given.
The two projects differ on more than the venue. ICE’s contracts would list on one of its regulated futures exchanges, after approval. Injective’s market went live without any such step: it is open to anyone with a wallet and USDC, and the blog post names no regulator and no country restriction. The baskets differ too. Ornn includes the newer RTX 5090; Injective keeps the older A100 and RTX 4090.
Neocloud warnings and 10x leverage: the risks
The launch comes as warnings pile up about the companies that rent out GPUs. Gartner said in October that more than 80% of specialist AI cloud providers, known as neoclouds, could stop operating independently by 2030, Cointelegraph reported. On Sunday 25 October, Elon Musk wrote on X that “the 200+ neoclouds are going to see guppy-level death and consolidation”, according to a separate Cointelegraph report. His argument was about credit: lenders, he said, have largely stopped financing AI cloud firms without substantial revenue outside AI data centres. In the same post he said SpaceX had secured $40 billion of debt for GPUs, so he speaks as a buyer of compute as well as a commentator.
Failures or mergers among those providers could push rental prices either way: down if their GPUs are dumped on marketplaces, up if capacity disappears. A contract on SICI would carry that uncertainty, with limits of its own. The index is built from marketplace listings collected by one data firm, not from a benchmark traded on an exchange. Its seven-day average smooths the hourly moves, so the published value reacts with a lag to a sudden jump or drop in rental prices. A basket rebased to 100 in July 2026 also has only a few months of history.
Then there is leverage. At 10x, a position worth $10,000 needs about $1,000 of margin, and a move of roughly 10% against it equals the whole margin posted. Venues usually close positions before that point, through liquidation, once the margin falls under a maintenance level. Injective has not published the maintenance level for SICI. Injective’s blog post does not say whether any regulator, in Europe or elsewhere, supervises the market.
What is still unknown
The main gaps are the ones that would tell readers whether the market works: its ticker, how much has traded since Monday, open interest, the funding schedule, the settlement method and the oracle that brings LO:TECH prints on chain. Injective’s recent changes to trading costs on its chain are in our report on fixed gas for exchange orders. Injective also says another product will launch “in the coming days”, without naming it.