Investors are increasingly eyeing a new class of tradeable asset: AI compute. Mike Bird, Wall Street editor at The Economist, explained on Yahoo Finance that the sheer volume of money flowing into AI infrastructure is creating pressure to turn rented processing power into liquid financial instruments. The path to a functioning market, however, is far from straightforward.
The push for a compute futures market
AI compute is, at base, processing power sold by cloud companies and specialized providers to labs and businesses that need to train or run models. Bird described a clear buyer-seller dynamic: "If you're a big cloud company or a neo cloud like CoreWeave, you're selling compute. And if you're a big user, maybe an AI lab, you're a buyer." That binary structure resembles the physical commodity markets that gave rise to futures and derivatives, and the appetite to replicate those structures is real.
But the assets are anything but interchangeable. "This is not a homogeneous asset," Bird said. Turning rented GPU time into a financial product requires solving problems that raw materials markets never faced.
Why compute resists commoditization
In traditional commodity markets, a barrel of Brent crude or a bushel of wheat can serve as a benchmark even though regional grades differ. Proponents of financializing compute often point to that model. Bird said the comparison breaks down quickly. "Even if you're renting a chip, like an H100 chip from Nvidia, even renting the same chip from different providers - not just in different parts of the world, but in the same part of the world - can vary in price by two, three, four times."
The culprit is not just geography. Pricing swings on the physical housing of the chip, the reliability of the data center, the software stack, and what power source it's linked to. "All of this matters enormously," Bird said. "It's something that a financial markets investor knows nothing about."
The search for usable benchmarks
A futures market needs clean, simple reference points. Several companies are building indexes on silicon data to create those benchmarks, Bird noted, but the underlying variability makes the task "very, very difficult." Without a standardized unit of compute that traders can price against, liquidity will stay bottled up.
The current wave of experiments may eventually produce something workable, Bird suggested, but he cautioned that it could take "a lot longer to see a sort of financial market to match the real physical market underlying it."
Why this matters for finance professionals
For anyone in finance tracking the emergence of AI as an asset class, the lesson is that technical granularity will dictate deal structure. The price dispersion Bird highlighted - two to four times on identical silicon - signals that knowing a chip model is not enough. Successful products will have to account for power contracts, data center reliability, and stack performance. Professionals who treat compute like a generic input risk mispricing it badly. AI for Finance courses that cover infrastructure economics are becoming part of the due-diligence toolkit for precisely this reason.
Your membership also unlocks:
