CME Group brings forward curve pricing to GPU rental indices as AI compute matures into a tradable asset class

Photo: Nana Dua / Pexels

CME Group brings forward curve pricing to GPU rental indices as AI compute matures into a tradable asset class

Silicon Data's H100 and B200 rental indices are now accessible through standard CME market data feeds, ahead of planned cash-settled futures contracts.

On October 15, 2026, CME Group announced that subscribers can access non-tradable forward curve pricing data for Silicon Data’s GPU rental indices, covering two of the most in-demand chips in AI infrastructure: the H100 and the B200. The data flows through standard CME market data feeds, meaning institutional desks can plug this into existing workflows without custom plumbing.

What the data actually shows

Silicon Data’s forward curves went live on April 20, 2026, and recalculate daily, projecting rates out to 36 months based on actual rental transaction data.

The underlying numbers give a sense of where the market sits right now. As of mid-October 2026, H100 GPU rentals are running between $2.53 and $2.81 per GPU-hour. B200 rentals, the newer and more capable chip, are fetching between $5.45 and $6.01 per GPU-hour.

Advertisement

Both curves are reportedly showing backwardation, meaning the market currently prices near-term supply tighter than future supply. In GPU terms, the implication is that the market expects rental rates to soften as more capacity comes online.

Silicon Data says its forward curves draw from data covering approximately 95% of the neo-cloud provider market, which is what gives the index enough depth to serve as a credible benchmark rather than a survey.

Two specific metrics sit at the core of what CME subscribers are now able to see. Term structure rates reflect the cost of locking in GPU rental capacity across different durations. Implied forward rates translate current spot and term data into what the market implicitly expects on-demand rental prices to look like at a future point in time.

Futures contracts are next in line

CME Group has been preparing to launch two contracts, labeled GPU1 and GPU2, which will reference Silicon Data’s H100 and B200 indices respectively. Each contract is sized at 730 GPU-hours. The planned launch date was October 5, 2026, pending regulatory approval.

Cash settlement is a meaningful design choice here. Rather than requiring physical delivery of GPU-hours, contracts settle against the index value in cash.

Silicon Data raised $30.5 million in a Series A funding round in 2026, with both CME Group and DRW among the investors.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.
CME Group brings forward curve pricing to GPU rental indices as AI compute matures into a tradable asset class
CME Group brings forward curve pricing to GPU rental indices as AI compute matures into a tradable asset class

Silicon Data's H100 and B200 rental indices are now accessible through standard CME market data feeds, ahead of planned cash-settled futures contracts.

Photo: Nana Dua / Pexels

On October 15, 2026, CME Group announced that subscribers can access non-tradable forward curve pricing data for Silicon Data’s GPU rental indices, covering two of the most in-demand chips in AI infrastructure: the H100 and the B200. The data flows through standard CME market data feeds, meaning institutional desks can plug this into existing workflows without custom plumbing.

What the data actually shows

Silicon Data’s forward curves went live on April 20, 2026, and recalculate daily, projecting rates out to 36 months based on actual rental transaction data.

The underlying numbers give a sense of where the market sits right now. As of mid-October 2026, H100 GPU rentals are running between $2.53 and $2.81 per GPU-hour. B200 rentals, the newer and more capable chip, are fetching between $5.45 and $6.01 per GPU-hour.

Advertisement

Both curves are reportedly showing backwardation, meaning the market currently prices near-term supply tighter than future supply. In GPU terms, the implication is that the market expects rental rates to soften as more capacity comes online.

Silicon Data says its forward curves draw from data covering approximately 95% of the neo-cloud provider market, which is what gives the index enough depth to serve as a credible benchmark rather than a survey.

Two specific metrics sit at the core of what CME subscribers are now able to see. Term structure rates reflect the cost of locking in GPU rental capacity across different durations. Implied forward rates translate current spot and term data into what the market implicitly expects on-demand rental prices to look like at a future point in time.

Futures contracts are next in line

CME Group has been preparing to launch two contracts, labeled GPU1 and GPU2, which will reference Silicon Data’s H100 and B200 indices respectively. Each contract is sized at 730 GPU-hours. The planned launch date was October 5, 2026, pending regulatory approval.

Cash settlement is a meaningful design choice here. Rather than requiring physical delivery of GPU-hours, contracts settle against the index value in cash.

Silicon Data raised $30.5 million in a Series A funding round in 2026, with both CME Group and DRW among the investors.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.