Who benefits from compute futures?
What do finance bros in Chicago and New York have to do with the future of AI compute?
I’ve written a bunch about compute financialization—turning a standardized unit of GPU performance into a hedgeable financial product akin to oil or other commodity—but I have not really explained who benefits. Why would the various buyers and sellers of compute want to hedge their exposure it, when they can simply buy and sell the product like any other?
If compute becomes a financialized commodity, then the cast of characters would mirror that of any mature commodity or credit market, but with distinctive twists rooted in the physics and economics of compute. There are five sets of parties to keep in mind here:
Structural longs: those consuming compute
Structural shorts: those producing compute
Speculators: trading firms who provide liquidity to longs and shorts
Arbitrageurs: those who profit off spreads between classes of compute
Financial intermediaries: those who manage the financial plumbing that allows a market to function
An important wrinkle in this particular commodities market is that some enterprises will be both long and short compute. Microsoft, for example, both owns its own compute and rents compute from others.
Each party’s role and interest is explained in further detail below.
Structural longs: those consuming compute
These are the industrial users of compute. Their problem is input cost volatility.
AI model developers/hyperscalers/SaaS platforms. Microsoft, Anthropic, OpenAI, etc. They’re long compute exposure: rising GPU rental rates squeeze margins. They’d go short compute futures to lock in price or hedge cost risk.
Enterprises with steady inference loads. Insurance, pharma, finance, etc. They may similarly short futures to secure budget predictability for training or inference workloads.
Data center operators could be structurally long or short depending on contract structure. A colocation operator selling capacity months ahead might be effectively short spot compute (they owe compute in the future), so they’d go long futures to hedge delivery risk.
Think of these participants as analogous to airlines in oil markets or miners in power hedging.
Structural shorts: those producing compute
These are the supply side entities whose revenue depends on selling compute hours.
Cloud providers/GPU lessors. CoreWeave, Crusoe, Voltage Park, Lambda. Their revenues rise with compute prices; they’re long compute naturally and may short futures to lock in sale prices for future capacity.
HPC infrastructure funds or GPU ETFs. Any vehicle whose NAV is tied to GPU use rates or resale value would hedge by shorting futures when prices spike.
Manufacturers and distributors (Nvidia, OEMs). Only indirectly exposed. They could use compute price indices as proxy hedges for inventory valuation or forward contracts for GPUs, but they’re more likely to remain outside direct financialization, at least initially.
This mirrors how oil producers or power generators short forward curves to secure revenue.
Speculators: the liquidity and volatility traders
Financialization always requires participants who want the risk.
Macro hedge funds/CTA desks. Trade compute as a macro-AI exposure, correlated with semiconductor cycles, capex expectations, or broader tech indices.
Quant funds/prop shops. DRW, Jump, Citadel, etc. They’d arbitrage betwen spot GPU rental markets, futures curves, and cross-commodity correlations (e.g., power, chips, or carbon credits).
Crypto-native funds. Financialized compute will likely appear on-chain before it appears in tradfi. DeFi protocols and DAO treasuries become liquidity providers, collateralizing with stablecoins or tokenized treasuries.
Retail speculators. Eventually, AI power ETFs or structured notes would emerge, turning compute prices into a proxy for optimism or pessimism about the AI economy.
Speculators are crucial for depth: they don’t hedge. They seek convexity, volatility, or correlation exposure.
Arbitrageurs and basis traders
Once indices exist, arbitrage layers bloom.
Spot-vs-futures arbitrage. Long GPUs/lease capacity, short futures when futures trade rich.
Cross-region basis. Exploit spreads between U.S. East vs EU hub indices, for example.
Cross-class basis. Arbitrage between A100, H100, and future architectures, much as crude traders arbitrage Brent-WTI.
Power compute spreads. Compute is power plus capital depreciation; some will trade spark-spreads of AI, linking GPU futures to energy futures.
These actors stabilize curves and enforce fungibility.
Financial intermediaries
Compute brokers, clearinghouses, and lenders fill out the ecosystem.
Prime brokers and custodians for margin and collateral management.
Clearing institutions for settlement.
Structured product desks packaging compute exposure into swaps or notes for corporates.
Compute REITs and securitization vehicles. Turning future GPU cashflows into bond-like instruments priced off the same indices.
This tier anchors compute as an institutional asset class.
A note on reflexivity
Financialization would feed back into the physical layer. The curve itself becomes a planning tool for hyperscalers and investors. Expectations embeded in futures prices would guide where and when new data centers are built. That reflexivity is what makes compute uniquely interesting compared to wheat or oil. It is both a commodity and the capital that produces intelligence.
Summary
Shorts: suppliers and producers of compute
Longs: consumers of compute
Speculators, arbitrageurs, and financial intermediaries: hedge funds, crypto protocols, and retail.
Once this collection of particpants matures, you get the true hallmark of commoditization: price discovery precedes production.
If you enjoy this newsletter, consider sharing it with a colleague.
I’m always happy to receive comments, questions, and pushback. If you want to connect with me directly, you can:
"Structural longs: those consuming compute... AI model developers/hyperscalers/SaaS platforms"
I know you know this by now, but it is odd to me that you don't break out the utility of cash settled versus physical delivery futures. Like yes, obviously this group of devs/hyperscalers/SaaS are long compute. That's basically a tautology.
But different types of futures provide substantially different value to them. Cash settled futures don't really do much for these types of businesses because they still need physical delivery.
To not include this fundamental distinction is odd. Clearly, I enjoy your writings. I sure post enough in your comment sections after your articles to show that. But this is just one that I don't get.