Practice · Compute
Compute is becoming a commodity. It still trades like a lease.
GPU capacity has a volatile price, a forming forward curve, and a growing list of venues waiting for real flow. What it does not have yet is a derivatives discipline. We bring the one we built across equity volatility, special situations, and digital asset treasuries to the operators who own the capacity, the buyers who depend on it, and the lenders who finance it.
Engagements are limited to accredited investors, qualified clients, and qualified eligible persons.
Where the market stands
A market is being built. The physical sellers have not shown up yet.
Venues are announced. Flow is not.
In 2026, major exchanges and several new venues announced compute futures, perpetuals, and forward curves, and at least three indices 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. An index earns trust from printed transactions with counterparties that can actually deliver.
The underlying already moves.
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. That is real price volatility with almost nowhere deep to transfer it.
The precedent is one asset over.
Hashrate forwards were built on flow from physical sellers, miners monetizing future output. Luxor's over-the-counter forwards launched in late 2022, settled on its Hashprice Index; exchange-listed hashrate futures followed on Bitnomial in 2024. Bitcoin derivatives went through a similar phase in 2017 and 2018, when listed futures were new and liquidity had not yet concentrated.
Pain → Posture
What changes when capacity is treated as a position.
From
Capacity leased for a decade at a fixed escalator: upside in compute pricing traded for downside protection at a price nobody checked, and the tenant's renewal, extension, and expansion options left unpriced.
To
A lease rate tested against a fair forward price, embedded tenant options priced before signing, and a forward program on uncontracted capacity.
From
Lease-backed debt that prices as unenhanced credit, while the credit support that would close the spread sits one layer up the chain.
To
A financing structure designed for the rating it should get, with enhancement mapped before the deal goes to market, not after.
From
Power, rates, and a legacy bitcoin treasury left unhedged through a multi-year build, because the hosting contract was treated as the whole risk program.
To
One governed program covering the risks the lease leaves behind, sized by regime and reported to the board in plain language.
Where we work
Six places the edge still exists.
Each is a problem the compute market has today and the derivatives world solved years ago in other assets.
01
Forward programs on uncontracted capacity
Forwards against pipeline and uncontracted megawatts, referenced to GPU-hour indices where the operator controls the GPUs and to its own colocation pricing where it does not, turning development-stage capacity into contracted revenue earlier in the build cycle.
02
Compute volatility origination
Forwards monetize direction. A volatility program built on the same physical flow goes further: structured hedges collateralized by real capacity, originated by the operator and placed with the buyers who need them.
03
Buyer-side hedging
AI labs, inference platforms, and model developers carry compute as their largest variable cost with no clean way to fix it. We design budget-certainty programs: price limits, term structure, and counterparty selection.
04
Financeability for HPC conversions
Lease-backed notes, project debt, and converts around the pivot from mining to high-performance compute. Convexity and credit review of the structure, including where third-party support could narrow the spread.
05
The risks the lease leaves behind
Power exposure on uncontracted load, floating-rate debt through construction, transmission cost changes, and the bitcoin still on the balance sheet. Each one is hedgeable. Few operators hedge all of them, and most leave several unhedged.
06
Governance for a new derivatives program
Board policy, limits, ISDA and CSA terms, accounting treatment, and disclosure for an operator running its first compute derivatives book, so the program is built to survive its first bad quarter.
Who we work with
Everyone with compute price risk.
Miners converting to HPC
Physical capacity, a build schedule, and a bitcoin treasury, all moving at once.
Data center and neocloud developers
Uncontracted megawatts, tenant concentration, and financing that depends on both.
AI labs and inference platforms
Compute is the cost line. The question is how much of it to fix, and at what price.
Lenders and credit investors
Underwriting lease-backed paper in a market where the reference price is still being invented.
Why Gammon
Gap risk is the same problem at any amplitude.
Twenty-plus years in options, much of it special situations and catalysts: event risk, gap risk, and positions whose payoff is decided by a single discontinuity. A fixed-price offtake against a volatile underlying is that same geometry.
We have watched a derivatives market form around a new asset before. In bitcoin, the participants who understood the physical flow early set the terms everyone else traded on. Compute is at that stage now.
We are paid by the client, not by the spread. Advice is non-discretionary, the client approves every transaction, and custody never sits with Gammon.
Reference library
How the compute market works.
Plain reference pages on the instruments, the indices, and the risks.
What are compute derivatives?
The pillar page: GPU-hour forwards, futures, perpetual swaps, and options, who uses them, and why the market is still early.
Read →
Hedging GPU compute costs
A five-step framework for AI labs and inference platforms that want budget certainty on compute.
Read →
Forward sale vs. fixed-rate lease
What a long-term lease gives away, what a forward program on uncontracted capacity keeps, and how to combine them.
Read →
Compute price indices
How GPU-hour benchmarks are built, why the settlement reference matters, and five tests for evaluating one.
Read →
Miner to AI / HPC conversion risk
The six risks a hosting lease leaves behind for bitcoin miners converting to AI hosting.
Read →
Data-center lease financing
Why similar lease-backed notes price far apart, and how operators capture credit support before going to market.
Read →
Where does your compute risk sit?
Tell us what you own, what you buy, or what you finance. We will tell you where the convexity sits and what it would take to keep it.
Talk about computeEngagements are limited to accredited investors, qualified clients, and qualified eligible persons.
Derivatives, digital assets, and overlay strategies involve substantial risk, including the risk of total loss. Past performance is not indicative of future results.