Via celestial.ai
The startup is building financial-grade benchmarks for GPU rental costs, and CME Group is already planning futures contracts around them.
Compute is the new crude oil. And just like crude oil, it turns out Wall Street wants to trade it.
Silicon Data, a startup positioning itself as the pricing authority for AI compute, has raised $30.5 million in Series A funding to build out the financial plumbing that lets institutions benchmark, hedge, and eventually speculate on GPU rental costs. The round was led by Valor Atreides AI Fund, with CME Group and Samsung Next among the backers.
From rental rates to reference prices #
The company maintains nine financial-grade indices covering GPU rentals and large language model token expenditure. Those indices pull data daily from roughly 100 platforms across more than 40 countries, covering high-demand chips like NVIDIA’s H100 and the newer Blackwell B200 series.
CEO Carmen Li has described Silicon Data as the “Bloomberg of GPU pricing.” The company introduced forward curves in April 2026, giving market participants a way to see where GPU rental costs are headed, not just where they’ve been.
The Series A follows a $4.7 million seed round closed in March 2025, which means Silicon Data has raised roughly $35 million in total. The new capital is earmarked for expanding its pricing indices, rolling out SiliconMark performance benchmarks, and building derivatives infrastructure.
CME Group enters the GPU futures game #
CME Group, the world’s largest derivatives exchange operator, plans to list cash-settled GPU futures contracts on NYMEX starting October 5, 2026. Those contracts will use Silicon Data’s H100 and B200 Rental Index as reference prices, with each contract covering one month of rental costs.
Cash-settled means no one is actually shipping racks of GPUs at expiration. Buyers and sellers settle in dollars based on the index price, the same structure used for weather derivatives and various energy contracts.
Why transparent compute pricing matters now #
The AI infrastructure market has a transparency problem. GPU rental rates vary wildly across providers, regions, and contract lengths. A spot hour of H100 time on one cloud might cost double what it does on another, and there’s been no standardized way to compare.
That opacity benefits incumbents with negotiating leverage and punishes smaller AI labs that lack the purchasing power to demand discounts. Silicon Data’s indices aim to flatten that information asymmetry by publishing daily benchmarks that any market participant can reference.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our