The Practitioner's Guide to the Compute Future: How to Hedge, Step by Step (Part 1)

Updated September 8, 2026. Index and curve values as of September 7, 2026.

If you run 50 H100s around the clock, you pay a little under $100,000 a month at today's rate, and you have no idea what you will pay next June. From October 5, pending regulatory review, you can find out. That is the practical meaning of the two compute futures contracts NYMEX is listing on Silicon Data's H100 and B200 rental indices, and this piece is about how a company that has never traded a future would actually use one.

The contract, in five facts

The CME notice sets out two contracts: Silicon Data H100 Rental Index Futures, code GPU1, and Silicon Data B200 Rental Index Futures, code GPU2. Five things matter for a user.

One contract is 730 GPU-hours. That number is 365 days times 24 hours divided by 12, the hours in an average month, so one contract is one GPU for one month. Sizing a hedge is counting GPUs.

Prices are quoted in dollars and cents per GPU-hour, the same unit as your invoice and the same unit as the index. The smallest move is one cent; one tick translates to $7.30 of P&L per contract.

Settlement is financial. No GPU changes hands; money does.

Trading in a contract ends on the last business day of its month, and 36 monthly contracts are listed at a time, so you can cover any month up to three years out: stack the monthly contracts for the months you need, and the stack is your hedge.

The exchange fee is $5.50 per contract for non-members. Your broker adds its own.

How it settles, and why an average

As currently announced, GPU1 for a given month settles against the Silicon Data H100 Rental Index averaged over that month, not the index on one day.

The reason for averaging is that compute behaves like electricity, not like oil. A GPU-hour is consumed the hour it exists; it cannot be stored and delivered later. Your bill is not one price on one day, it is the sum of every day's rate across the month. A contract that settles on the month's average tracks that bill.

It is also much harder for one odd day, or one thin quote panel, to move a monthly average than a single print, which is why power contracts such as PJM Western Hub settle the same way. Bandi and Su at Johns Hopkins, in their early asset-pricing framework for compute, make exactly this argument for why a first-generation compute contract should average over the delivery month.

Hedging a year of H100 rental step-by-step

Take a company renting 50 H100s continuously through 2027 from a neocloud provider whose rate runs at $2.78/hour. With the H100 index at $2.63 on September 7, their rental rate is $0.15 above the index (the basis). The company is worried that rental rate may skyrocket next year and decides in early October to hedge that risk.

It buys 50 contracts for each month from January to December 2027, 600 contracts in all. The contracts do not trade yet, so the prices below are illustrative, but they are not invented: they use Silicon Data's published term rate implied forward curve for the H100 on September 7 as guidance, which ran from $2.50 three months out to $2.14 twelve months out. On that curve the strip starts at $2.46 for January and steps down to $2.16 for December, averaging $2.26, well below today's $2.63.

Does that mean the market expects rental prices to fall? In Part 2, we'll unpack how to read this implied forward curve, and whether it reflects fair value for the compute future.

Two things then happen every month of 2027. The provider bills the company at the index plus $0.15. The January contract settles at the January average of the index, and the company receives, or pays, the difference between that average and $2.46, times 730, times 50.

Here is the year under two very different paths for the index, one drifting up to $3.20 by December and one drifting down to $1.80.

The net is identical to the dollar. It is the strip, $2.26 times 730 times 600, plus the $0.15 basis on every hour. Before the hedge, the company's 2027 compute bill had a range of more than $330,000 between those two paths. After it, the bill is a budget line with certainty.

The lower row is worth sitting with. In the falling path the company pays $36,682 into the futures market and its net cost is higher than if it had not hedged. That is not a flaw; it is the deal. A hedge trades the chance of a cheaper year for the certainty of a known one. A company that would rather keep the upside should not hedge, and one that cannot absorb the downside should.

One more thing: you can sometimes keep the upside and limit the downside, if CME lists options on these futures.

What happens day to day

Futures settle every day, not just at the expiration. Suppose the January contract is bought at $2.46 and over the next five trading days settles at $2.49, $2.47, $2.52, $2.50 and $2.55. On 50 contracts the company's account is credited $1,095, debited $730, credited $1,825, debited $730 and credited $1,825. The running total after five days is $3,285, which is exactly $0.09, the move from $2.46 to $2.55, times 730 times 50.

These daily flows are called variation margin. They net to the difference between the final settlement and the price at which you buy, so nothing is gained or lost by the daily accounting; what changes is the timing, and the company needs cash on hand to meet a run of debits in a month when the index falls. Before trading, the exchange also requires a deposit per contract, initial margin, which is returned when the position closes; the exchange sets and publishes the level.

The other side of the story is the same

For a neocloud or compute provider with capacity to lease, the hedging math is exactly the same, just run in reverse. While a buyer fears rising rental rates and buys futures (a long hedge), a provider fears falling rates and sells them (a short hedge). If a provider wanted to lock in the forward curve for their 50 GPUs, they would simply sell those 600 contracts instead of buying them. In the scenario where the index drifts down to $1.80, the provider's physical rental income would fall, but their short futures position would generate an identical offsetting profit of $36,682. The derivative works symmetrically, allowing infrastructure operators to lock in bankable revenue and protect their capital investments from a spot market collapse.

The limitations: where a hedge can fray

No hedge is a magic shield. A cash-settled compute future removes macro price uncertainty, but it does not eliminate operational friction.

The basis risk

In our worked example, the math worked out to the penny because we held the provider's differential static at $0.15 above the index throughout all twelve months. In the physical compute market, that spread, the basis, is not carved in stone.

The Silicon Data index tracks the broad, aggregate equilibrium of the market, the macroeconomic tug-of-war between overall AI compute demand and raw silicon supply. While that broad market tide directly steers the baseline cost of compute, individual cloud providers operate with their own micro-level pricing pressures: regional imbalance, network constraints to name a few.

If your contract sits at Index + $0.15 in January, but your provider raises your rate to Index + $0.35 by July while the index itself stays flat, the futures payout will not absorb that extra $0.20 per hour. Hedging protects you from the macro cycle; it does not protect you from local contract renegotiations.

The hyperscaler mismatch

If your workloads run on a hyperscaler, proceed with caution. GPU1 settles exclusively against an index of independent neocloud providers, and the two ecosystems operate on fundamentally different pricing dynamics.

Execution friction and the cost of liquidity

Entering and exiting a futures position is in itself an art governed by market microstructure. In a nascent contract, liquidity has to develop. Your true transaction cost will depend heavily on the bid-ask spread quoted on the exchange screen:

• If the screen quotes $2.45 bid / $2.47 ask, locking in your 50 January contracts at the ask costs you a one-cent premium over fair value ($365 on the block), negligible to the bottom line.
• If the market is wider, say $2.40 bid / $2.52 ask, executing via a blunt market order could trigger substantial slippage, eating away a notable slice of your planned hedge savings.

As market makers step in to post competitive two-sided quotes, bid-ask spreads will tighten and algorithmic execution will optimize block entry. This is something to watch in the early days of listing and execute your hedge with care. You will need to coordinate closely with your clearing broker to work limit orders or explore block-trade facilities rather than crossing wide spreads on the screen.

Conclusion and up next

Before October the choices for locking in compute were a provider's reserved contract, with its own terms, deposits and specific hardware, or nothing. The future adds a third route that is separate from the hardware: rent wherever you like, and fix the price on the index. Which route is cheaper depends on how the reserve market and the futures market price the same months.

In the sequel, we will break down how to evaluate those two options side by side. We will explore how to compare a provider's physical reserve rate directly against the futures forward curve, and, more importantly, how to execute an arbitrage strategy when those two markets fall out of alignment.

This article is an illustration of contract mechanics, not a recommendation to trade. Futures involve leverage and daily cash requirements; contract terms are subject to the CME rulebook and pending CFTC review. Prices for contracts that have not yet listed are hypothetical. Silicon Data is the benchmark provider, and the futures are the instruments that settle against them.