Hivemind has published new research examining how compute could develop from a cloud service into a financial market.

The publication comes as the Commodity Futures Trading Commission (CFTC) seeks input on compute derivatives. Its new request for comment asks whether cash markets are large and liquid enough to support derivatives, and how regulators should approach market integrity and customer protection. That consultation brings a question long discussed within the industry into sharper focus: what would a functioning compute market actually require? 

Hivemind’s report, Everything You Wanted to Know About Compute Markets but Were Afraid to Ask, argues that the foundations are already visible. Cloud agreements contain early versions of spot and forward contracts, but opaque pricing makes it difficult to judge value or manage future costs. 

Compute already trades like a market

Major cloud providers sell interruptible capacity at dynamic prices and offer longer commitments that secure access in advance. These arrangements already operate at scale, but each provider runs its own closed system. Buyers have no independent benchmark for the price of compute.

An exchange-traded layer could sit alongside these contracts. Buyers could reserve future capacity with less capital tied up at the outset, while infrastructure owners would gain a way to manage exposure to changing prices and hardware depreciation.

Creating that layer depends on a shared definition of what is being traded. Compute changes quickly, and one unit is rarely identical to another.

A market built for constant change

The challenge for instance is that the value of an hour of GPU capacity depends on the hardware used alongside that GPU chip (the rack) and where that rack is hosted. Any benchmark will also age as newer chips replace the current standard.

Established markets offer a useful precedent. A compute benchmark could move with the technology, while basis pricing accounts for differences between the benchmark and the capacity being delivered. This approach would allow markets to develop without pretending that every unit of compute is interchangeable.

A flexible benchmark would solve only part of the problem. Physical delivery will also matter because buyers often need capacity rather than compensation when prices rise. A lab that secures compute in advance must be able to use it when demand arrives.

Making capacity tradable, however, addresses only one side of the market. Capacity contracts alone will not fund the infrastructure buildout. Standardized debt products could draw more capital into datacenters, with DMBS and DMBX offering potential models for matching investors with appropriate levels of risk.

The takeaway

Compute is already bought and sold through arrangements that resemble a commodity market. A more open structure would bring transparency, flexibility and resilience to activity that is taking place behind closed doors. The CFTC's request shows that regulators are beginning to consider what that market should look like, and Hivemind's research offers a framework for the discussion.

The full research paper, Everything You Wanted to Know About Compute Markets but Were Afraid to Ask, is available to download above.

‍