Oil has Brent, gold has a London fix, and a barrel of wheat has the Chicago pit — but the hottest commodity of the AI age, an hour on an NVIDIA H100, had no public price at all. Two companies could rent the very same chip and pay wildly different rates, with no way to know who overpaid. Silicon Data built the ticker. Let's check them out!
Every frontier AI model runs on GPUs that are rented by the hour, not owned — and for years, what that hour actually cost was anyone's guess. Rates were quoted in private deals, buried in annual contracts, and scattered across dozens of clouds. As founder Carmen Li puts it, "two companies buying the exact same GPU capacity could pay wildly different prices with no way to know who got the better deal." In a market barreling toward ~$1 trillion of data-center capex in 2026, that fog is an expensive place to be flying blind.
Silicon Data is the fix. Founded in 2024 by ex-Bloomberg data executive Carmen Li, the New York startup publishes daily, financial-grade price indices for GPU compute — nine of them and counting, from the H100 to the B200 to a blended LLM token-cost index. Each is built from roughly 150,000 verified pricing records a day, covering 80%+ of the global GPU rental market. The data already streams through Bloomberg and LSEG terminals, and in August the company raised a $30.5M Series A — up from a $4.7M seed just 17 months earlier.
🌐 Website: silicondata.com
📅 Founded: 2024 (New York)
👤 Founder: Carmen Li (CEO), formerly global head of strategic data alliances at Bloomberg
💰 Stage: Series A — $30.5M raised (led by Valor Atreides AI Fund; CME Group, DRW, Jump, Wintermute, Samsung Next & VanEck participating); ~$35M total
📈 Traction: 1,000+ registered users; nine daily indices carried on Bloomberg & LSEG
🏦 Marquee win: the settlement price behind CME's first Compute Futures (H100 & B200), live Oct 5, 2026
🔮 Tech Trend: AI Infrastructure / the financialization of compute
🎯 Target Market: AI labs, neoclouds & hyperscalers, plus hedge funds and trading desks
🏢 Location: New York, NY
📈 You can't hedge what you can't price: The GPU-as-a-service market is projected to more than triple to ~$14.4B by 2033 (16% CAGR), and rental rates are anything but stable — H100 one-year rentals jumped ~40% off their October 2025 low. Scarce, perishable, wildly volatile pricing is the exact condition that makes a benchmark — and the derivatives on top of it — genuinely valuable.
🏦 A startup becoming market infrastructure: This October, CME Group will launch cash-settled H100 and B200 futures that settle against Silicon Data's indices — letting a data center lock in rental revenue and an AI lab cap its compute bill. When the CME chief calls compute "the new oil of the 21st century," being the reference price everyone quotes is a remarkable place for a two-year-old company to sit.
👤 The right founder for a trust business: An index is only worth the credibility behind it, and Carmen Li spent years running enterprise-data partnerships at Bloomberg before setting out to build "the Bloomberg for GPU pricing." The moat is the unglamorous part — a normalization method that turns wildly different lease types, regions, and clusters into one comparable number — and a cap table (CME, DRW, Jump) that doubles as distribution.
The challenge will be trust and liquidity — the two things a benchmark lives on. Silicon Data isn't alone: a week after the CME deal, ICE announced rival compute futures with Ornn, whose index rides the Bloomberg Terminal. And because CME holds equity in Silicon Data while trading-firm backers like DRW and Jump may trade the very contract its index settles, the company has to prove its numbers are beyond reproach — in a thinly traded market where, as one report warns, "a futures contract with few participants produces a price that is easy to move and hard to trust." Own the credibility, though, and you own the plumbing for an entire asset class.
📊 Explore the live GPU indices — H100, B200, H200 and more, updated daily
📰 Read the CME compute-futures announcement (live Oct 5)
🎧 Hear Carmen Li's plan for a compute futures market on Bloomberg Odd Lots
🧠 Learn how compute becomes a tradable commodity on their blog
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.