AI computing will be available as a tradable commodity starting October 5, with the price becoming publicly available.

      Renting an Nvidia H100 for one hour costs whatever price your provider sets, with no standard price available. Two companies acquiring the same capacity might end up paying vastly different amounts, and neither would be aware of this.

      This scenario will change on October 5. The CME Group, known for trading crude oil and corn, is set to launch two futures contracts related to the hourly rental price of Nvidia GPUs, pending regulatory approval. Each contract will represent one month's rent for a single GPU, according to CNBC. One contract will track the H100, the most commonly used chip in AI systems today, while the other will monitor the upcoming Blackwell B200.

      Although it is a relatively minor product, it carries significant implications. The essential input for every AI company is about to receive a public reference price as well as a forward curve reflecting traders' expectations for the price in the upcoming year.

      What is being traded

      No one actually takes possession of a graphics card. Both contracts are settled in cash based on indexes published by Silicon Data, a New York firm that has spent two years monitoring actual GPU rental costs. They will be available on the NYMEX, CME's energy exchange. This detail is not trivial. The executive mentioned in the launch announcement, Pete Keavey, serves as CME’s global head of energy and environmental products.

      He explicitly compared this development to oil, stating that oil powered the 20th century economy and evolved from spot trading into a global derivatives market. These contracts, he indicated, will "turn compute into a standardized, tradable commodity.”

      The dynamics of buyers and sellers are straightforward to envision. A data center operator, who owns servers and earns rental income, can sell futures to secure revenue. Conversely, an AI developer, responsible for paying those rentals, can purchase futures to manage their costs.

      Wall Street was the first to establish the financial infrastructure

      This marks the second key development in just two weeks. Previously, Nvidia engaged six prominent financial firms for a $500 billion funding initiative aimed at advancing AI infrastructure. Carmen Li, CEO of Silicon Data, differentiated the two efforts while speaking to Bloomberg. She noted that the Nvidia announcement pertains to financing, while this new initiative serves as a risk management layer. She argued that a market of this magnitude needs a mechanism for hedging and price discovery.

      This past month has illustrated the circumstances from an external perspective. Significant investments have been made in computing resources with bare minimum public data on what those compute costs entail. Lenders have been filling this gap creatively. For example, Lambda facilitated a $917 million leveraged loan backed by chips whose future value remained unquantifiable. OpenAI even sought a lead for power trading because the electricity market already exists.

      Gavin Baker from Atreides Management, who was part of the funding round announced the same day, offered a farming analogy. He said futures markets allow farmers to finance their upcoming seeds and equipment rather than rely on speculation: "You can't plan against a price that is opaque."

      The company setting the price has just raised $30.5 million

      Silicon Data announced it has closed an initial funding round of $30.5 million on the same day, led by the Valor Atreides AI Fund. This round is approximately six times the size of the $4.7 million raised in March 2025, 17 months prior.

      The newly raised funds will support four areas: pricing benchmarks, an institutional data service, risk infrastructure for derivatives and credit, and a performance product known as SiliconMark. Li expressed to Bloomberg the goal is to become "the independent referee" for the compute market.

      SiliconMark is particularly significant, as it addresses an issue that could jeopardize the futures contracts. Even when constructed from identical chips, different clusters can yield distinct performance outputs due to factors like networking, topology, and configuration. This variability poses a challenge for those hedging positions. If the index corresponds to a standard hour of H100 usage but your cluster underperforms, the hedge could diverge from actual exposure. Silicon Data claims that adjusting for performance could also allow for physical delivery in the future.

      Almost everyone in this market is a stakeholder in the referee

      The list of investors warrants careful examination, as it is publicly available. CME Group itself participated, along with trading firms such as DRW, Jump, Wintermute, and Tectonic, as well as financial institutions like VanEck, F-Prime, Samsung, and Further.

      Thus, the exchange possesses equity in the company that produces the index against which its contracts will settle, and several likely market participants also hold stakes. Li acknowledged this relationship during her Bloomberg interview, considering those investors to be among her main clients.

      This situation is not uncommon in commodity benchmarks, where businesses seeking pricing often support the entities that provide it. It is important to emphasize this point nonetheless; a benchmark's reliability depends on its independence, and the market it addresses is currently quite small.

      Regulators have scrutin