IN 2014, AS cloud computing was taking off, a group of German technologists and financial-platform operators had an idea. They launched the Deutsche Börse Cloud Exchange, on which firms could buy and sell spare access to computing power.
The effort to turn processing power into a liquid asset did not scale well, according to Randolf Roth, its chief executive and veteran of the exchange business. Standardising central processing units (CPUs), the general-purpose chips that predominated at the time, proved too difficult and integrating different types of cloud capacity too expensive. Hopes that futures and options would follow in the wake of the original launch were dashed. The venture shut down in 2016.
But Mr Roth and his partners were early to an idea whose time really may have come. Businesses around the world spent $129bn on cloud services in the first quarter of the year, estimates Synergy, a research firm, more than would be spent in a year a decade ago. Immense investment in data centres is under way. America’s five cloud “hyperscale” giants are poised to spend around $700bn on capital expenditure this year.
The hottest part of the market is not for CPUs but for the graphics-processing units (GPUs) used to train and run AI models. Some companies now spend more on artificial intelligence, and by extension on processing power, than on wages for employees. “Compute”, once a piece of techno-jargon for digital oomph, has entered common parlance. Those who sell it want to turn their assets into cashflows. Those who buy it want to hedge against sharp movements in the prices of a costly, critical input. And middlemen are racing to create financial instruments to match the two groups.
Companies that own and operate data centres certainly want to wring more out of their hardware. Some “neocloud” firms, which focus on high-end GPUs for AI, have used the value of the chips as collateral for loans. In 2024 Macquarie, an Australian investment bank, helped organise a $500m loan to Lambda, a neocloud, with the loan backed by the borrower’s valuable chips from Nvidia, which furnishes most of the world’s AI silicon. The same bank approved another GPU-backed loan to Fluidstack, another neocloud, in April last year for an undisclosed amount.
Debt backed solely by GPUs remains rare, even for neoclouds with masses of chips. Some of what looks like GPU-backed lending is actually more pedestrian. CoreWeave, the largest neocloud, runs 49 data centres. Its compute-related power capacity is one gigawatt, around the same as a medium-sized nuclear reactor. Since 2023 the company has used chips as part of the collateral for a series of loans from alternative-asset managers such as Blackstone and Magnetar Capital. These are backed by chips but also guarantees from CoreWeave’s clients with multi-year contracts for cloud capacity.
The guarantees let lenders take comfort that CoreWeave can pay back its debts while sparing them from having to think too hard about how fast its GPUs will lose their value. This funding structure is also reassuringly familiar, even if the technology is not: it is common in toll roads and other old-economy infrastructure finance.
Although such instruments serve big compute sellers well, big buyers need something different. Locking in a price with a multi-year neocloud contract insures a buyer against compute getting more expensive but not against it getting much cheaper. So as large companies the world over spend ever more on compute, they want to be able to hedge against price volatility just as they insure against changes in energy tariffs, interest rates or foreign-exchange movements—ideally in deep and liquid derivatives markets.
Two startups want to help businesses do this, by turning nascent indices tracking compute costs into a futures market. Silicon Data, founded in 2024 and backed by DRW, a trading firm, has paired up with CME Group, which operates large derivatives exchanges. Ornn, created by recent graduates of the Massachusetts Institute of Technology and run from a flat rather than an office just a few months ago, has paired up with Intercontinental Exchange, the parent company of the New York Stock Exchange, to do the same. Both aim to launch compute futures later this year, to be traded on their partner exchanges.
Each firm has its own indices tracking the hourly rental price of specific advanced chips such as the H100, Nvidia’s workhorse GPU. Silicon Data’s series tracks both transactions and quoted prices, which results in a smoother index. Ornn includes only real transactions, which makes its index harder to manipulate but noisier. Silicon Data’s price for the H100 is now around 10% higher than Ornn’s. Two months ago it was 40% higher. Neither is likely to reflect more than a fragment of the wider market, most of which is private.
Broadening the appeal of compute derivatives requires standardisation. This is, as Mr Roth discovered a decade ago, finicky. Different types of chip are often imperfect substitutes. Even the same chips offered by different cloud providers may vary wildly in performance and price. The way in which hundreds or thousands of chips are connected to one another, the environment in which they are housed, and data centres’ location all weigh on demand.
Such problems lead some industry insiders to suggest that a well-functioning derivatives market could be a decade away. The index-providers and derivative-designers retort, reasonably enough, that no existing market was perfectly standardised at inception. Even in deep and liquid markets, some products will always stray far from benchmarks—just look at oil prices. In March oil futures in Oman and Dubai, mostly for delivery to Asian buyers, climbed to over $150 per barrel. At the same time, the price of West Texas Intermediate, the American benchmark, sat below $100.
There are other reasons to be bullish about compute standardisation. Chips from Nvidia’s Hopper range, which includes the H100, have held much of their rental value even as next-generation Blackwell models entered the market. This flies in the face of a comment in March 2025 by Jensen Huang, Nvidia’s boss, that “when Blackwell starts shipping in volume, you couldn’t give Hoppers away”. It also suggests that compute derivatives’ underlying assets will not suddenly become worthless, the risk of which would slow adoption.
Other middlemen are trying to get in on the action. Architect Financial Technologies, a software firm, has launched an exchange for compute futures in Bermuda. It is also now offering its own futures, in competition with the big exchanges, and planning one licensed for trading in America, called the American Innovation Exchange. Architect’s boss, Brett Harrison used to run FTX US, the American arm of the cryptocurrency exchange which imploded in 2022. He says the AI Exchange (in unsubtle shorthand) aims to make a market not just for compute, but for raw materials, like copper and silver, and sources of energy needed for AI chips and data centres.
Roundhill Investments and ProShares, purveyors of exchange-traded funds, have filed paperwork with securities regulators to create ETFs for compute futures once these exist. Other ETF providers have filed to launch related “2x funds”, which magnify the returns (or losses) with leverage. As compute becomes the currency of the AI era, all these efforts will intensify. As Mr Harrison sums up, “All of a sudden, the race is on.”
For more expert analysis of the biggest stories in economics,finance and markets, sign up toMoney Talks, our weekly subscriber-only newsletter.
