The Commodification of Compute: A Misguided Quest for a New Market

The tech industry’s latest obsession is to create a futures market for Nvidia’s AI chips, reducing computing power to a tradable commodity like oil or wheat. CME Group plans to launch futures contracts tied to the hourly rental cost of Nvidia’s H100 and B200 graphics processors on October 5th, pending regulatory approval. This move reflects Wall Street’s ongoing attempt to quantify and control the intangible.

The comparison to hotel rooms is apt but incomplete. While a night at the Ritz and one at a Motel 6 buy different experiences, there’s a fundamental difference between computing power and hospitality. Compute is not just about hardware; it also involves the provider, location, rental term, networking, and availability – all of which can significantly impact its value.

The market for compute has been built on the assumption that AI infrastructure is becoming increasingly commoditized. However, this narrative ignores the complexities of AI development and deployment. Unlike energy or commodities, computing power cannot be simply standardized and traded; it’s a bespoke service requiring expertise, context, and nuance to provide.

Wall Street’s track record in creating new markets is mixed at best. The DRAM attempt failed due to industry disagreement over standards, while the weather futures market faced challenges related to specificity and standardization. Bandwidth shows what happens when a market fails to develop around a commodity: prices plummet as capacity outstrips demand.

The real question is whether the market for compute will follow this pattern or take a different path. The CFTC’s Chair Michael Selig has expressed enthusiasm for a derivatives market for compute, but his agency still questions whether the market is standardized and transparent enough to support one. Yggdrasil Financial Technologies warns of conflicts around privately produced benchmarks, which could reproduce the LIBOR dynamic in miniature.

For investors, the market could become useful even if they never trade it. The futures curve could reveal new insights into expectations months ahead, adding a third dimension to our understanding of AI build-out and demand. However, this will only happen if real buyers and sellers show up, not just managed money looking for a way to be long or short Nvidia.

The divergence between semiconductor stocks and spending plans on the one hand, and H100 and B200 rental prices on the other, could signal that capacity is arriving ahead of demand – but we won’t know for sure unless participation decides whether this is a signal or a sentiment index. In any case, Wall Street’s quest to commodify compute raises more questions than it answers: Can computing power be reduced to a tradable commodity? And what does this say about our understanding of AI and its applications?

The launch of futures contracts on October 5th will be a test not just for the market but also for our collective understanding of the role of Wall Street in shaping the future of technology. Will we see a repeat of past failures, or can we finally tame the complexities of compute and make it a tradable commodity? Only time – and participation – will tell.