GPU futures bring Wall Street’s tools to compute markets

A new market for AI compute: What GPU futures could mean for energy

Standardized contracts could give data centers, lenders, investors, and energy companies a way to manage volatile compute demand and power costs.

3 key takeaways

1GPU compute remains fragmented and opaque, creating demand for transparent prices, central clearing, and standardized contracts.
2Silicon Data’s indices could support futures settlement as buyers and sellers manage exposure across hardware, power, and financing.
3Chip cycles, supply constraints, and uncertain AI adoption could make compute prices volatile even as the market expands.

Don't miss

Peter Keavey explains why energy remains a volatile marginal input into GPU economics, even though it is not the entire cost of a GPU.

The brief

GPU capacity is mostly traded through reservations, private contracts, platforms, and phone calls, leaving prices fragmented and counterparties exposed. The proposed fix is an exchange-traded market for standardized compute.

Peter Keavey of CME Group and Carmen Lee of Silicon Data explain the infrastructure behind the idea: daily compute indices, verification, central clearing, and forward prices extending as far as 36 months.

The core argument is that GPU economics cannot be separated from energy. Power is only one input, but it is a volatile, continuing marginal cost that can reshape data-center and generation decisions.

Potential participants range from hyperscalers, AI companies, lenders, chipmakers, and memory suppliers to energy producers, with natural buyers and sellers hedging different exposures.

The market’s promise comes with a warning: chip shortages, construction delays, and uncertain AI adoption could produce the persistent volatility that makes hedging valuable in the first place.

What was said on this episode

10 statements · 3 positive · 1 negative · 2 mixed · 4 neutral

Peter Keaveyon GPU marketsNeutral3:55
GPU markets are closely tied to energy markets, especially power.
“GPU. being so closely tied to energy markets, particularly power. It's a natural extension of our customer interest and our customer base.”
Listen at 3:55
Carmen Lion GPU provider contractsNeutral13:13
GPU provider contracts commonly last from over one month to two or three years.
“the contract usually lasts longer than a month, can be two years, three years.”
Listen at 13:13
Peter Keaveyon Global traders and compute consumptionNeutral31:49
Some of the world’s largest traders are also major compute consumers.
“some of the largest traders in the world are also some of the largest consumers of compute”
Listen at 31:49
Carmen Lion Silicon Data clientsNeutral32:38
Silicon Data’s clients are approximately 60% financial services and 40% industry market participants.
“60% financial services can be the lenders, can be whoever have a natural exposure to GPUs, right? Or the underlying asset class, the semi-industries and hyperscalers, right? That sorts. And 40% actually market participants.”
Listen at 32:38
Peter Keaveyon GPU rental ratesNegative33:28
GPU rental rates fell from $7–$10 to $2–$3 per hour as chip supply normalized.
“Rental rates were seven to. $10 an hour. As supply normalized, those came down back to earth, $2 to $3 an hour.”
Listen at 33:28
Peter Keaveyon AI adoption across industriesPositive35:08
AI is beginning to integrate into pharmaceuticals and many other industries.
“it is getting integrated into pharmaceutical, into all sorts of different industries that are just sort of at the beginning stages of putting it into their business chain.”
Listen at 35:08
Peter Keaveyon GPU market trading and risk managementPositive35:29
GPU market trading and risk management retain substantial growth potential even if AI growth moderates.
“even if the growth moderated, there's still a lot of growth potential in trading this market and trying to manage risk in this market”
Listen at 35:29
Carmen Lion B200Mixed36:50
B200 GPU prices initially fell but later exceeded their original listing prices.
“B200 came out to be $5 each last year, right? And it came down a little bit and came back up and then surpassing their listing, whatever, original listing prices”
Listen at 36:50
Carmen Lion GPU pricesMixed37:05
GPU prices should decline only if the supply chain becomes robust.
“The price should come down if and only if we have a very robust supply chain.”
Listen at 37:05
Carmen Lion Large language model adoptionPositive37:36
Greater language-model adoption will require substantially more GPUs.
“adopt token lm token better then they need a lot more gpus right to power the adoption of the large language models”
Listen at 37:36

Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.