The making of
a compute market.

Commodity markets are built around physical things with practical constraints. They need a way to describe what is being bought, establish where and when it will be delivered, and determine what happens when an obligation is not met. Price becomes meaningful inside that framework.

Electricity. A power station produces energy, but a working power market also needs transmission, system coordination and rules for buying and delivering it. In much of the United States, regional transmission organizations and independent system operators coordinate wholesale markets around the grid. The physical system and the market operate together. FERC describes that relationship here.

The lesson for compute is that a price cannot be separated from the conditions of delivery. Capacity at the wrong location, at the wrong time, or without the right operating support may have a very different value to the buyer.

Natural gas. The evolution of the US gas market included separating the sale of gas from the service of transporting it. FERC’s 1992 Order No. 636 restructured pipeline services; its capacity-release framework also allowed shippers to make unneeded transportation capacity available to others. These arrangements created ways to reallocate rights around an existing physical network. Order No. 636 and FERC’s capacity-release explanation describe that development.

Henry Hub became a widely used reference for US natural gas prices. A reference gives participants a common point from which to discuss value; it does not make transportation constraints or differences between delivery points disappear. The EIA publishes the Henry Hub price series.

Grain. The Chicago Board of Trade introduced standardized futures contracts in 1865, following a period of trading forwards. A defined contract gave participants a common structure for trading around a physical commodity. That history connects the work of specifying an underlying trade with the development of financial instruments around it. CME Group traces that evolution.

This is what we mean by commoditization: enough common understanding for a wider group of participants to compare, transact, finance and manage exposure. Different markets reach different degrees of standardization. Quality, location and delivery conditions continue to matter.

A GPU-hour is a useful unit of time. It is an incomplete description of a product. The GPU architecture, memory, network fabric, storage, location, availability window, contract term and operating scope all affect what a buyer can do with it.

For the buyer, the requirement is a workload that runs. For the operator, it is a deployment that can be delivered and supported. For the lender, it is equipment and cash flow behind a financing commitment. For the owner, it is an asset whose useful life, utilization and residual value all matter.

The first task is to make those different readings compatible. Define the deployment. Make the performance evidence understandable. Connect the build-out plan to the commercial commitments. Identify who is responsible at each handoff.

That is why we start with the physical transaction. A useful market reference has to reflect the product people actually receive.

An AI company can carry substantial compute-price exposure without owning a single GPU. A change in the price or availability of capacity can change its costs, margins and ability to serve customers. A long-term reservation can secure access while creating a commitment that outlasts the demand behind it.

An operator faces the other side: equipment and site commitments made before every future hour has a buyer. Customer demand can change while financing payments, power arrangements and operating costs continue. New hardware generations can change the economics of equipment already deployed.

A lender’s exposure includes how the cluster gets built, whether it becomes operational on time and whether the demand assumptions supporting repayment hold. An asset owner has exposure to utilization, operating performance and the value of the hardware at exit. Power and site providers have their own commitments tied to the deployment.

These are connected exposures. As more products and businesses depend on AI, more of their economics depend on the price, availability and performance of the compute beneath them. The relevant question becomes: how do participants measure that exposure, and what can they do about it?

Compute Function operates the desk between the people trying to make these deployments happen. A buyer may not know the operator or understand the lender’s conditions. A lender may need help connecting an equipment budget with site readiness, commissioning milestones and an operating plan.

We help facilitate those conversations and organize the transaction so the participants can assess the same underlying facts. Lenders make their own credit decisions; our role is to make the build-out and its dependencies easier to evaluate.

Benchmarking is one part of that work. A number needs a configuration, workload and test conditions behind it. Commercial evidence needs context too: an asking price, an agreed price and an executed transaction are different observations. So are planned capacity and capacity that has been delivered.

The aim is to organize these observations consistently, with their sources, dates and assumptions visible. Over time, and with appropriate permissions, that evidence can become a record that remains useful beyond a single deal.

Our long-term ambition is to build compute indexes and financial products. The desk is our entry into the market; reliable evidence from the physical market is the foundation we want to build on.

A credible index would give participants a reference for a defined segment of compute. It would need representative observations, a transparent methodology and governance people can trust. It would make the boundaries of the reference clear, rather than treating every GPU-hour as the same product.

Financial products could then help participants finance capacity and manage exposure to compute prices. Their usefulness would depend on the quality of the reference, the obligations they define and the market they serve. This is our intended direction; compute indexes and financial products are not currently offered.

This is also the company-building thesis. An operating relationship can create access to evidence that a public listing cannot capture on its own. If that evidence becomes sufficiently representative and reliable, the business can grow from facilitating individual transactions into providing references and products used across many transactions.

That progression has to be earned. It starts with understanding the infrastructure, helping the people around it work together and keeping a record others can read.