Silicon Data raises $30.5 million as CME readies GPU futures

Why it matters

CME's planned contracts could turn Silicon Data's GPU indices into financial infrastructure. The Series A funds Li's effort to establish the benchmark before compute derivatives develop liquidity and competing reference prices.

Carmen Li (@carmenli), the former Bloomberg and DRW executive building financial infrastructure for AI compute, has raised a $30.5 million initial Series A closing for Silicon Data, according to an announcement Tuesday.

The Valor Atreides AI Fund led the financing. CME Group, DRW, F-Prime, Samsung Next, VanEck, Further, Jump Trading, Tectonic Ventures and Wintermute also invested. Breed, Hack, Blank VC, Sancus Ventures and SoGal Ventures participated, according to Silicon Data's funding release.

Silicon Data did not disclose its valuation, the amount contributed by the lead fund or the size targeted for the full Series A. Calling the financing an "initial closing" leaves room for Silicon Data to add capital or investors before completing the round.

Li founded New York-based Silicon Data in 2024 after more than 15 years in financial services, including roles at DRW and as Bloomberg's global head of strategic alliances for enterprise data. She holds a degree from the University of Illinois Urbana-Champaign and an MBA from Harvard Business School. Li has described Silicon Data as her first company, writing in a 2025 reflection that the recurring anxiety she felt during her corporate career disappeared after she started building it.

That background explains Silicon Data's approach to the GPU market. Li is treating compute capacity as a financial input that needs benchmarks, forward prices and hedging tools, rather than solely as infrastructure bought from cloud providers. Silicon Data publishes daily rental-price indices for GPUs including Nvidia's H100, H200, A100 and B200, as well as AMD's MI300X. Silicon Data also tracks memory and large language model token prices.

"Every mature market eventually develops independent referees," Li said in the Series A announcement. Silicon Data wants its indices to establish what compute costs, while its SiliconMark product measures what GPU infrastructure actually delivers.

The CME connection

The financing follows a May 12th agreement under which CME Group plans to use Silicon Data's daily GPU rental benchmarks as reference prices for cash-settled compute futures. The contracts remain subject to regulatory review.

If launched, the futures would give AI developers, cloud providers and investors a way to hedge changes in GPU rental prices. They would also put Silicon Data in a position resembling an index administrator: CME's contracts would depend on the credibility, continuity and resistance to manipulation of Silicon Data's underlying pricing data.

CME's participation in the Series A deepens that relationship. The investor list also includes trading firms and asset managers that understand how a benchmark becomes embedded in a market. DRW and Jump co-led Silicon Data's seed financing, while Wintermute and Sancus also participated in that earlier round.

Those backers can provide more than capital. Exchanges need credible settlement prices, traders need liquid instruments, and asset managers need data they can use to value exposure. Silicon Data is assembling those constituencies before the planned futures market opens.

Building the reference layer

Silicon Data said the Series A will support four areas: pricing benchmarks, performance measurement through SiliconMark, institutional and alternative data products, and risk infrastructure for derivatives, credit and insurance.

The performance work addresses a problem that rental prices alone cannot capture. Clusters using the same GPU model can produce different output because of networking, topology, software and configuration. Silicon Data argues that normalizing those differences could let buyers compare effective performance and eventually hedge the risk that rented infrastructure delivers less compute than expected.

Silicon Data says it has developed nine financial-grade indices and has more than 1,000 registered users. Those figures are self-reported, and Silicon Data did not provide revenue or identify the institutions represented in the user count.

Silicon Data previously raised a $4.7 million seed round in March 2025 and announced it publicly on May 20th. The two disclosed rounds total $35.2 million.

The new capital gives Li time to expand Silicon Data's datasets and methodology before CME's planned contracts begin trading. Once a futures market settles against a benchmark, replacing that benchmark becomes operationally difficult. The Series A is financing Silicon Data's attempt to secure that position while the market for AI compute derivatives is still being designed.