The new oil? Inside the effort to turn AI computing power into a tradeable commodity

Futures markets have long been used by companies to manage uncertainty, whether it’s airlines hedging fuel costs or farmers hedging crops. Now, a startup is looking to bring that same financial machinery to the world of artificial intelligence.

Silicon Data, a company that monitors pricing across cloud providers and GPU marketplaces, has teamed up with CME Group to introduce what could potentially be the world’s first futures contracts tied to the computational power required to run AI. This innovative approach would allow businesses to hedge against fluctuations in the cost of training and running AI models, providing a level of financial security in an otherwise volatile market. While these contracts are still awaiting regulatory approval, early indications suggest a growing interest from investors.

Founder and CEO Carmen Li envisions a future where the AI compute futures market could rival some of the largest commodity markets in the world. She believes that the energy demand associated with running artificial intelligence will eventually surpass all other energy uses combined, making it a potentially lucrative market for investors.

The concept behind AI compute futures is simple yet profound. Just as airlines rely on jet fuel to operate, AI companies depend on high-end GPUs to power their systems. However, most companies rent access to this technology through cloud providers and neoclouds, leading to fluctuating costs that can be challenging to predict. This uncertainty is what Silicon Data aims to address with its GPU price indexes, which track the hourly rental cost of specific chips across providers. These benchmarks could serve as the foundation for a futures market similar to how West Texas Intermediate crude oil underpins energy derivatives.

In addition to providing a means for companies to hedge against rising compute costs, the AI compute futures market is also expected to attract speculators. These traders, who may not have a direct need for GPU capacity, can play a crucial role in building liquidity and improving price discovery. While speculation can sometimes lead to volatility, proponents argue that it helps establish prices for the broader industry.

One of the key challenges for the proposed futures market is standardization. Unlike physical commodities like oil or corn, AI compute is not a standardized product. With over 50 different configurations of Nvidia’s H100 chip alone, prices can vary based on various factors such as processors, memory, networking, and data center location. Silicon Data addresses this complexity by normalizing prices to a base H100 case, ensuring that traders have confidence in the benchmark representing these variations.

As the AI compute futures market awaits regulatory approval, the filings for exchange-traded funds tied to these contracts by asset managers like ProShares and Rex Shares indicate a growing interest in AI compute as a potentially tradable asset class. With the support of market participants, regulators, and industry experts, the AI compute futures market could revolutionize the way companies manage their AI infrastructure costs in the future.