← Intelligence

Intelligence · Compute · Pillar

What are compute derivatives?

· Michael Mescher, Gammon Capital

Direct answer

Compute derivatives are financial contracts whose value is tied to the price of computing capacity, usually the rental price of a specific class of GPU quoted per GPU hour. The family includes forwards (a fixed price today for capacity delivered later), futures (exchange-listed forwards settled against an index), perpetual swaps, and options. They let the owners of data-center capacity lock in revenue, let AI labs and inference platforms lock in cost, and let lenders underwrite a business whose main input and output both move in price. As of 2026 the instruments have been announced by several venues, but exchange-listed futures are still in regulatory review and the contracts that do trade are small; the market resembles bitcoin derivatives in 2017 and 2018, when listed futures were new and liquidity had not yet concentrated.

Key takeaways

  • The underlying is the price of GPU compute, typically dollars per GPU hour for a named chip class, reported by an index provider.
  • Four instrument types matter: physical forwards, cash-settled futures, perpetual swaps, and options.
  • Natural sellers are capacity owners (miners converting to AI hosting, neoclouds, data-center developers). Natural buyers are AI labs, inference platforms, and model developers.
  • Settlement depends on a credible index, and an index only becomes credible when real physical counterparties transact against it.
  • The closest precedent is the hashrate forward market, built by bitcoin miners selling future output before any listed product existed.

Definition

Compute derivative: a contract whose payoff depends on the future price of computing capacity. In practice the reference price is the hourly rental rate for a specific GPU class (for example, a current-generation data-center GPU), measured by an index that aggregates observed rental prices across providers. The contract either delivers the capacity (physical settlement) or pays the difference between the agreed price and the index (cash settlement).

The four instruments

InstrumentWhat it doesSettlementTypical user
Physical forwardFixes a price today for capacity delivered over a future windowDelivery of GPU hours or reserved capacityOperator selling to a known buyer
Cash-settled futureSame economics, listed and cleared, no deliveryDifference vs. an index at expiryHedgers who do not want delivery risk
Perpetual swapContinuous exposure to the index with a funding ratePeriodic funding paymentsTraders and short-horizon hedgers
OptionRight, not obligation, to buy or sell at a set pricePhysical or cash vs. indexBuyers wanting a price cap; sellers monetizing volatility. No exchange-listed compute options as of October 2026

Why the market exists

GPU rental prices move. Supply arrives in large, lumpy tranches as new data centers energize; demand shifts with each model generation and each new chip release. Prices for a single chip class have moved by tens of percent within a quarter: Ornn's index showed Blackwell spot rental rising 48%, from $2.75 to $4.08 per GPU hour, between mid-February and mid-April 2026. A data-center operator carrying years of construction debt, and an AI company whose largest variable cost is compute, both carry that price risk whether they hedge it or not.

Until recently the only tool was the long-term lease: a hyperscaler or AI lab signs a multi-year contract at a fixed rate with a fixed escalator. A lease is a long-dated forward, usually with options granted to the tenant: renewal at the escalator, extension, and expansion rights. It fixes the price for both sides but cannot be traded, sized, or unwound, and the tenant's options are rarely priced.

Where the market stands

In 2026, several venues announced compute products, including exchange-listed GPU futures, perpetual swaps, and forward curves, and multiple index providers are competing to become the settlement reference. Exchange-listed futures are still in regulatory review, and liquidity on the products that do trade is small; in July 2026 Bernstein described it as nascent and dominated by speculative flows.

That is the normal shape of a new derivatives market. Venues and indices arrive first; liquidity arrives when physical participants, the ones who own or consume the underlying, start transacting. See what a compute price index is for why the settlement reference matters so much at this stage.

Venues and indices (as of October 12, 2026)

VenueProductReferenceStatus
CME GroupCash-settled GPU rental futures (H100, B200)Silicon Data rental indicesAnnounced 2026; planned October 5 launch delayed while the CFTC review runs to about November 9, 2026
ICE Futures U.S.GPU futures (H100, H200, B200, RTX 5090)Ornn OCPIAnnounced May 19, 2026; pending approval, no listing date
ICE Futures U.S.Energy-normalized compute futuresNATIVX COILAnnounced July 1, 2026; pending
KalshiCompute forward curves from event contracts (B200, H200, A100)Ornn OCPILive since July 14, 2026
Architect (AX, Bermuda)GPU perpetual swapsOrnn dataLive offshore, outside CFTC oversight

Status changes quickly in a new market. Check each venue's own notices before relying on a date.

The hashrate precedent

Bitcoin miners faced the same problem with hashrate: a volatile output price and heavy fixed costs. A hashrate forward market formed when miners began selling future output at fixed prices to buyers who wanted the exposure. Luxor's over-the-counter hashrate forwards launched in late 2022, settled on its Hashprice Index, and exchange-listed hashrate futures followed on Bitnomial in 2024. Flow from physical sellers built the market before the listed products existed. Compute is following the same path, and many miners familiar with hashrate hedging are now converting capacity to AI hosting.

Common mistakes

Treating the lease as the hedge. A fixed-rate lease removes price risk on contracted capacity but says nothing about uncontracted megawatts, pipeline capacity, or renewal risk at the end of the term.

Waiting for liquidity. In an early market, the participants who trade first help decide which index wins and where the curve sits. Waiting means accepting a curve someone else built.

Ignoring basis. An index tracks a chip class across many providers. A specific operator's realized price depends on its tenants, location, and contract terms. The gap between the two is a real risk and needs to be sized before any hedge is placed.

Frequently asked questions

When do CME GPU futures launch?

CME planned to list cash-settled GPU rental futures referencing Silicon Data indices on October 5, 2026. The launch was delayed while the CFTC review runs to about November 9, 2026. Check CME's own notices for the current date.

Are there exchange-listed compute options?

Not as of October 2026. Options and price caps on compute are bespoke, bilateral contracts.

What is the difference between a compute forward and a GPU lease?

A lease fixes rent with one tenant for a long term, usually with renewal and expansion options granted to the tenant. A forward fixes the price of a defined amount of capacity for a defined window, and can be sized, layered, and placed with several buyers.

Who are the natural buyers and sellers?

Sellers are capacity owners: bitcoin miners converting to AI hosting, neoclouds, and data-center developers. Buyers are AI labs, inference platforms, and model developers that want to fix or cap their compute cost.

Gammon Capital view

Compute is becoming a commodity that still trades like a lease. The capacity owners who treat it as a position rather than a rental agreement, and who supply the first real flow into the forming market, will have more say in how that market is built than anyone who arrives after the indices settle. We advise operators, buyers, and lenders on that transition. See the compute practice.

Framework, not implementation plan. This page describes how Gammon Capital approaches the topic. It is a general framework; engagement work is specific to each client's contracts and balance sheet. Market facts are stated as of the date shown above.

Canonical citation. Gammon Capital (gammoncap.com), Michael Mescher.

How Gammon works on this

Gammon Capital designs the policy, structure, and counterparty process as a non-discretionary advisor, paid by the client rather than the spread; the client approves every trade and keeps custody.

the compute practice →Service lines →

Related resources

Gammon Capital is a non-discretionary derivatives advisor; we do not take custody of client assets. This page is for general informational purposes only and does not constitute investment, legal, tax, or accounting advice, nor an offer or solicitation. Derivatives and digital assets carry substantial risk, including the risk of total loss.