Compute Exchange: Carmen Lee builds GPU computing market bigger than oil market

Compute Exchange: Carmen Li is building a GPU computing power market larger than the oil market

In a new episode of the Bloomberg Odd Lots podcast, co-founder and CEO of Compute Exchange and Silicon Data Carmen Li (Carmen Li) told hosts Joe Weisenthal and Tracy Alloway how her team is trying to build a computing power market that could become larger than the oil market, and what is needed to create a GPU index.

Why computing power is becoming a commodity like oil

The generative AI boom has sharply increased demand for scarce GPUs for training and inference of large models. Data center computing power has become a scarce resource that AI labs, cloud providers, and startups compete for — roughly the same way companies have historically competed for oil supplies and other raw materials. GPU shortage in recent years has been one of the main topics in the AI industry: the largest cloud providers spend tens of billions of dollars annually on data centers and Nvidia chips, and Nvidia itself has become one of the most expensive public companies in the world largely because of this demand.

The idea of Compute Exchange and similar projects is to turn this scarcity into a tradeable financial instrument: indices and in the future futures on the price and availability of GPU-hours.

How the Compute Exchange project works

• Guest of the episode — Carmen Li, co-founder and CEO of Compute Exchange and Silicon Data.
• Hosts of the Odd Lots podcast — Bloomberg staff journalists Joe Weisenthal and Tracy Alloway.
• Goal of the project — to build a computing power market that in volume can exceed the oil market.
• Key instrument — GPU index, reflecting the price and availability of computing power on the market.

Why investors and AI companies need such a market

A formalized compute market would give AI companies a way to hedge risks of sharp price swings in GPU rental — similar to how airlines hedge fuel prices through oil futures. For investors, a computing power index would become a new asset class directly tied to the growth rate of the AI industry, rather than individual stocks of chipmakers or cloud providers. The idea of turning access to these resources into a tradeable asset logically continues the trend of financialization of AI infrastructure, which in recent years has already led to the emergence of specialized GPU cloud providers and intermediary platforms for renting compute.

The format of the Odd Lots podcast is one of the most famous financial podcasts from Bloomberg, where hosts Joe Weisenthal and Tracy Alloway regularly break down with guests the structure of new or little-understood markets: from commodity derivatives to complex financial instruments. The choice of the compute market theme for a podcast episode itself shows that the topic has gone beyond a narrow circle of AI infrastructure specialists and has become a subject of interest to a wider financial audience.

What this means

The emergence of tradeable indices and in the future futures on GPU power — is a sign that compute is finally turning from a technical metric of data centers into an independent financial market around which new infrastructure is being built — from exchanges to derivatives.

If such a market takes hold, it will change how AI companies plan expenses: instead of long-term contracts with a single cloud provider, they will be able to buy access to GPU power on the open market, responding to price fluctuations almost the same way industrial companies today manage purchases of energy and raw materials.