Carmen Li knows more about the price of AI computing power than pretty much anyone else on Wall Street. Soon, she hopes to make it an asset class like oil or gas that investors can bet on -- or against.
Li is the 40-year-old founder of Silicon Data, a startup that compiles indexes tracking the rental prices of Nvidia chips. As early as next month, Silicon Data is joining with CME Group to launch futures contracts based on the indexes.
The idea is to bring pricing transparency to what's quickly becoming the most coveted and costly resource powering the AI boom: computing power, or compute, as it is known in the industry. Anthropic, OpenAI and Google are among the AI giants scrambling to secure it, with the country's data-center build-out expected to take years. Li says the futures contracts will give companies a way to hedge a huge business cost and traders a way to speculate on the AI boom.
"The goal of Silicon Data is to become a referee," Li said in an interview.
The contracts are poised to debut as investors worry that AI won't be able to deliver on the hundreds of billions of dollars being poured into it, with prices of stocks and bonds of AI companies recently stumbling as traders try to anticipate where a shakeout might happen.
Compute futures would offer a cleaner way to bet on or against the AI boom by focusing on the rental prices companies would need to pay for the technology. Already, futures contracts have expanded from a way for farmers to lock in prices for corn and other agricultural products to span every corner of the economy, letting traders bet on price moves for everything including gold, interest rates and cryptocurrencies.
Skeptics abound when it comes to compute futures, with some critics comparing the idea to internet bandwidth futures launched in the dot-com era, which failed after a supply glut of fiber-optic networks sent prices plunging.
Regulators are also cautious. CME delayed the Oct. 5 launch of the compute contracts after the Commodity Futures Trading Commission extended its review of them for another 45 days. In its previous request for public comment on such derivatives, the agency flagged a range of concerns.
The agency noted the possibility of market manipulation by dominant players with pricing power, as well as the variance in pricing across providers and regions. It noted that compute may not yet have the characteristics that typically underpin a commodity derivatives market, including "fungibility, standardization, and sufficient liquidity."
A chance to go abroad
Li grew up in Hangzhou, China, a sprawling tech hub on the country's east coast that is home to Alibaba and DeepSeek.
At 15, she got a scholarship to study in Singapore, where she went to an all-girls school. She wanted to go farther afield for college and ended up deciding on the U.S.
At the University of Illinois Urbana-Champaign, she studied finance and computer science, and then got a job as an electronic options market maker at the Chicago trading firm DRW.
Soon after joining, she had an idea for a trade and decided to email the firm's founder and CEO, Don Wilson, to get his thoughts on it. She said she was shocked when he emailed back shortly after, asking about aspects of the trade she hadn't considered.
Over the years, she'd occasionally email him to run other ideas by him, even after she moved on from the firm.
She eventually landed at Bloomberg, leading the company's deals to provide data to outside parties. In her talks with customers, she increasingly noticed companies' frustrations with the volatility of computing costs and the lack of options for hedging them.
In March 2024, she asked Wilson to meet up at a futures industry conference in Boca Raton, Fla., where she said they talked about the opportunity AI computing presented as an asset class. The conversation lasted only about 30 minutes but set her on the path.
By April, she had set up Silicon Data, and Wilson agreed to back her with $1 million in initial funding.
"It escalated quickly," Li said.
Customers of Silicon Data's indexes include hedge funds looking for data sources for their research and financiers providing loans that are backed by GPUs, the chips used for powering AI.
Racing to be first
Others backing Silicon Data include VanEck, the ETF and mutual-fund manager, and the Valor Atreides AI Fund, a venture-capital fund of hedge-fund manager Gavin Baker and SpaceX board member Antonio Gracias.
Another backer is Jump Trading Group, which recently hired a team to launch its compute-trading business when the futures go live. A number of other firms have filed regulatory paperwork to eventually launch exchange-traded funds that track the market for AI computing power.
CME plans to list two contracts representing one month's worth of rent for Nvidia chips, one for the H100 and another for the B200, based on Silicon Data's indexes. Silicon Data says its indexes rely on public and private GPU rental records from more than 40 sources.
Traders can take positions on the price of compute up to three years in the future. When the contract expires, trades will be settled in cash, so sellers won't have to deliver any compute capacity to buyers.
With the delay of the contracts' launch on CME, Silicon Data may no longer have a first-mover advantage. Ornn, a startup founded by two MIT grads and backed by a16z, is partnering with Intercontinental Exchange to launch similar contracts.
Wayne Nelms, one of Ornn's 23-year-old co-founders, said he understands the CFTC's concerns with the prospect of a compute market, given the importance of AI's development for the country and the role derivative contracts might play in helping mitigate risk and bringing more clarity to pricing.
"In order to do that, you have to have trusted benchmarks," he said.