GPU rent gets a printed price on 5 October

CME Group will begin trading two futures contracts tied to the hourly rental price of Nvidia GPUs on 5 October, subject to regulatory review ^[1]. That converts the central input cost of the AI industry from a number your supplier tells you into a published reference price and a forward curve showing what traders expect that price to be next year ^[8][26].

The mechanics are modest. Each contract represents one month's rent for a single GPU, according to CNBC ^[2]. One tracks the H100, the chip most AI systems run on today; the other tracks the Blackwell B200 that succeeds it ^[3]. Nobody takes delivery of a card: both settle in cash against indexes published by Silicon Data, a New York firm that has spent two years tracking what GPU capacity actually rents for ^[4]. They will be listed on NYMEX, CME's energy exchange ^[5], and the executive fronting the launch is Pete Keavey, CME's global head of energy and environmental products ^[6], who said oil grew from spot trading into a global derivatives market and that these contracts "turn compute into a standardised, tradable commodity" ^[7].

The operator case is straightforward. A data centre owner with rental income can sell futures to lock in revenue; an AI developer paying those rents can buy them to cap costs ^[9]. The more immediate effect is informational. Right now two companies buying identical capacity can pay very different rates and neither will know ^[8], which makes procurement a matter of relationship rather than benchmark.

The financing side arrived first. Nvidia recruited six of the largest names in finance for a $500bn funding package aimed at the AI buildout ^[10], and Silicon Data chief executive Carmen Li told Bloomberg that was a financing layer while this is a risk management layer, arguing a market that size cannot function without somewhere to hedge and discover prices ^[11]. Lenders were already improvising: Lambda sold a $917m leveraged loan backed by chips whose residual value nobody could independently price ^[12], and OpenAI hired a power-trading lead because electricity at least already has a market ^[13]. Gavin Baker of Atreides Management, who led the funding round announced the same day, put it as farm finance: "You can't build against a price you can't see or lock in" ^[14].

Two things should temper the enthusiasm. First, basis risk. Two clusters built from identical chips do not deliver identical output, because networking, topology and configuration change what you get ^[20], so if the index tracks a standard H100 hour and your cluster underperforms it, the hedge stops matching the exposure ^[21]. Silicon Data's answer is a performance product called SiliconMark, funded by the round alongside its benchmarks, an institutional data business and risk infrastructure for derivatives and credit ^[18]; the company says normalising for performance also opens the door to physical delivery later ^[22].

Second, ownership. Silicon Data announced an initial closing of $30.5m led by the Valor Atreides AI Fund ^[15], against $4.7m raised in March 2025 ^[16], roughly 6.5 times the earlier round ^[17]. The investor list includes CME Group itself, plus trading firms DRW, Jump, Wintermute and Tectonic, and VanEck, F-Prime, Samsung and Further ^[23]. So the exchange holds equity in the company whose index its contracts settle against, and several likely participants do too ^[24]; Li told Bloomberg those investors are heavily her clients at the same time ^[25]. The publisher notes this is not unusual in commodity benchmarks ^[27], which is true and also the reason benchmark governance is a standing subject of dispute.

Li's stated ambition is to be "the independent referee" for the compute stack ^[19]. Whether the curve becomes quotable in supply contracts, rather than a screen traders watch, is the test.