Silicon Data and CME Plan AI Token Financial Products

Silicon Data founder Carmen Li is pursuing financial products for trading computing prices with CME Group, The Wall Street Journal reported on the 22nd local time. The companies have been preparing futures based on GPU rental rates and now aim to extend trading to token prices, the fees for using AI services. The product structure and launch schedule have not been finalized.

Efforts to make computing an investment target were already underway. In August, Nvidia announced a plan with six Wall Street financial firms, including BlackRock and Goldman Sachs, to raise more than $500 billion for AI infrastructure over the long term. The plan seeks to attract investment based on GPU and data center revenue. Nvidia CEO Jensen Huang said at the time, “Computing is revenue in AI.” OpenAI CEO Sam Altman also said in a December interview last year that doubling computing resources would double revenue.

Securing GPUs also requires major spending on data centers and electricity, while rental rates change with demand. Higher rates can raise costs for AI developers, while lower rates can reduce revenue for data center operators that rent out GPUs. In an official CME announcement in August, Li said companies renting the same GPU faced large price differences depending on their contracts and could not easily tell which terms were favorable. Silicon Data said it surveys rental rates from multiple providers to create a benchmark that helps companies forecast and manage future costs. DRW founder Don Wilson, an early investor, said in a May announcement that tools to manage price fluctuation risks were lacking despite the surge in data center investment.

Futures trading can offset losses from price changes with gains from financial trades, instead of buying or renting actual GPUs in advance. For example, a company worried that a GPU rental rate of $2 per hour could rise to $4 could buy futures to prepare for an increase. A rental provider concerned about rates falling could also trade. If prices move against expectations, traders can lose money. Participants could include banks and specialist investment firms as well as AI developers and data centers. Silicon Data would research the price benchmarks, while CME would operate a market using them.

The companies announced plans for a GPU futures market in May and in August unveiled products based on rental rates for Nvidia H100 and B200 GPUs. Trading had been scheduled to begin on October 5 but was delayed for regulatory review by the U.S. Commodity Futures Trading Commission. CME has set November 9 as the planned first trading day, though the date could change depending on the review.

Bloomberg reported on the 21st that the companies are also pursuing financial products linked to token prices, a benchmark for AI model usage fees. A token is a unit of data processed when AI reads text and generates a response. OpenAI and Anthropic charge business customers based on tokens. Silicon Data operates an average price index combining fees and usage across multiple models and is considering products whose returns would move with that price. GPU rental rates represent the cost of producing AI services, while token prices correspond to the fees charged to customers for those services.

Silicon Data’s average token price was 54% below its late-May peak. An analysis of customer usage records by Ramp, reported by Axios, found token prices were also 48% below their late-February peak. Price cuts by AI companies and businesses choosing less expensive models contributed to lower usage fees. Ramp’s analysis found that companies’ token usage increased while their spending barely changed for two consecutive months. Token-linked products could be used to hedge against price changes or invest in movements in the AI services market, but their structure and launch timing have not been set. GPU values can fall quickly when new products arrive, and rental rates for the same model can vary by contract, making a benchmark price difficult to establish. It is also uncertain whether the market will attract enough participants.

Silicon Data founder Carmen Li is pursuing financial products for trading computing prices with CME Group, The Wall Street Journal reported on the 22nd local time. The companies have been preparing futures based on GPU rental rates and now aim to extend trading to token prices, the fees for using AI services. The product structure and launch schedule have not been finalized.

Efforts to make computing an investment target were already underway. In August, Nvidia announced a plan with six Wall Street financial firms, including BlackRock and Goldman Sachs, to raise more than $500 billion for AI infrastructure over the long term. The plan seeks to attract investment based on GPU and data center revenue. Nvidia CEO Jensen Huang said at the time, “Computing is revenue in AI.” OpenAI CEO Sam Altman also said in a December interview last year that doubling computing resources would double revenue.

Securing GPUs also requires major spending on data centers and electricity, while rental rates change with demand. Higher rates can raise costs for AI developers, while lower rates can reduce revenue for data center operators that rent out GPUs. In an official CME announcement in August, Li said companies renting the same GPU faced large price differences depending on their contracts and could not easily tell which terms were favorable. Silicon Data said it surveys rental rates from multiple providers to create a benchmark that helps companies forecast and manage future costs. DRW founder Don Wilson, an early investor, said in a May announcement that tools to manage price fluctuation risks were lacking despite the surge in data center investment.

Futures trading can offset losses from price changes with gains from financial trades, instead of buying or renting actual GPUs in advance. For example, a company worried that a GPU rental rate of $2 per hour could rise to $4 could buy futures to prepare for an increase. A rental provider concerned about rates falling could also trade. If prices move against expectations, traders can lose money. Participants could include banks and specialist investment firms as well as AI developers and data centers. Silicon Data would research the price benchmarks, while CME would operate a market using them.

The companies announced plans for a GPU futures market in May and in August unveiled products based on rental rates for Nvidia H100 and B200 GPUs. Trading had been scheduled to begin on October 5 but was delayed for regulatory review by the U.S. Commodity Futures Trading Commission. CME has set November 9 as the planned first trading day, though the date could change depending on the review.

Bloomberg reported on the 21st that the companies are also pursuing financial products linked to token prices, a benchmark for AI model usage fees. A token is a unit of data processed when AI reads text and generates a response. OpenAI and Anthropic charge business customers based on tokens. Silicon Data operates an average price index combining fees and usage across multiple models and is considering products whose returns would move with that price. GPU rental rates represent the cost of producing AI services, while token prices correspond to the fees charged to customers for those services.

Silicon Data’s average token price was 54% below its late-May peak. An analysis of customer usage records by Ramp, reported by Axios, found token prices were also 48% below their late-February peak. Price cuts by AI companies and businesses choosing less expensive models contributed to lower usage fees. Ramp’s analysis found that companies’ token usage increased while their spending barely changed for two consecutive months. Token-linked products could be used to hedge against price changes or invest in movements in the AI services market, but their structure and launch timing have not been set. GPU values can fall quickly when new products arrive, and rental rates for the same model can vary by contract, making a benchmark price difficult to establish. It is also uncertain whether the market will attract enough participants.