Silicon Data CEO Carmen Li wants to turn GPU prices into a futures market

The AI arms race has a new front, and it has nothing to do with who builds the fastest chip. The real fight, according to Carmen Li, is over who gets to define how AI compute is priced, traded, and hedged.

Li is the founder and CEO of Silicon Data, a company she launched in April 2024 with a fairly audacious premise: GPU rental prices are volatile enough, and economically significant enough, that they deserve the same financial infrastructure as oil or interest rates.

Building the Bloomberg terminal for compute

Silicon Data’s core product is a set of pricing indices, most notably the H100 Rental Index and a suite of LLM token indices, that aggregate hundreds of thousands of data points to track what AI compute actually costs in real time.

When Li says prices fluctuated from roughly $1.20 to $1.55 per million inference tokens in early 2026, that range matters as much to an AI startup’s unit economics as crude benchmarks matter to an airline’s fuel budget.

The indices feed into Compute Exchange, Silicon Data’s trading platform for GPU capacity. The exchange has reported hundreds of millions in notional trading volume since launch, and the company has accumulated more than 1,000 registered users spanning AI developers and financial institutions.

The funding trajectory reflects the market’s appetite. Silicon Data closed a $4.7 million seed round in 2025, then followed that with a $30.5 million Series A in August 2026, bringing total capital raised to $35.2 million.

CME Group enters the picture

The most consequential development is a partnership with CME Group, the world’s largest derivatives exchange. CME plans to reference Silicon Data’s benchmarks for cash-settled GPU futures and options products, with a targeted launch in fall 2026, pending regulatory approval.

Li has framed the addressable market in eye-catching terms. The potential derivatives market for compute resources, she has suggested, could reach into the range of $6 trillion to $30 trillion, dwarfing the hundreds of billions already flowing into data center capital expenditure.

Tokenomics, redefined

Li has deliberately borrowed the word “tokenomics” to describe Silicon Data’s framework, though the tokens in question are AI inference tokens, not blockchain assets. The economics of how AI systems consume and price compute share structural similarities with how crypto networks meter resource usage, and Li is positioning Silicon Data at that conceptual intersection.

Silicon Data is taking a path of centralized benchmarks, institutional partnerships, and regulated derivatives, in contrast to crypto-native infrastructure projects that have spent years trying to build decentralized compute markets. The bet is that TradFi plumbing scales faster than token incentive mechanisms when the clients are Fortune 500 procurement teams.

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