Quick summary · Chapter 24
Markets for Compute
First editionThe essential ideas, decisions, and diagrams from this chapter. Follow the section links for the examples, calculations, and fuller argument.
First-edition visual companion · Selected from the current chapter · 8x10-inch PDF
The main idea
This book has treated capacity as decisions made inside one organization: what we measure, promise, reserve, and commit to pay. This closing chapter looks outward. Compute already has markets; what 2026 added was an attempt to list futures on the rental price of a GPU, so that buyers and sellers could pass price risk to someone else. As of October 4, 2026, none of those contracts was trading on a U.S. exchange. The work they ask of a buyer, defining exactly what is priced and comparing our own price with an outside one, can start now.
Treat the chapter as advanced reading and a direction of travel; nothing here changes the practice of the previous twenty-three chapters.
A market appears
Why exchange trading has been difficult
The first obstacle to exchange trading is heterogeneity. An H100 and a B200 are different resources, and the same chip in two regions or on two networks may support different work. A token buyer is buying a service, not a GPU-hour. A standard contract has to say which differences it includes, which it adjusts for, and which stay with the buyer; Architect’s proposed delivery network, for example, would publish basis tables by SKU, memory configuration, and location.1
What is available and what is announced
Status as of October 4, 2026:
| Instrument | What is bought | Price set by | Example (status) |
|---|---|---|---|
| Provider spot | Interruptible capacity now | Provider, without bidding | EC2 Spot (operating)2 |
| Prepaid capacity block | Named instances for a future window, prepaid | Provider, from supply and demand at purchase | AWS Capacity Blocks (operating)3 |
| Commitment | Lower rates for a spending or usage obligation | Published or negotiated terms | Chapter 20 (operating) |
| Procurement marketplace | Competing quotes, then a direct supplier contract | Supplier offers | Compute Exchange (operating)4 |
| Physical forward | Delivery in a future period at a price fixed now | The two parties | Compute Exchange (offered); OneChronos–Auctionomics (announced)5 |
| Resale rights | Permission to sublease contracted time | Other buyers | SF Compute (operating)6 |
| Cash-settled index future | Exposure to a monthly average rental index | Exchange trading | CME, ICE, Nodal, Architect (none trading)7 |
| Perpetual future | H100 price exposure, no expiry | Crypto-venue trading | Seven venues (trading, thin)8 |
If suitable contracts start trading, I would take four questions to Finance and Treasury first. How closely has the index tracked our actual rate, month by month? How much trades in the months we would hedge? Who has authority to trade? Who funds the margin when the position moves against us? A hedged shortage is still a shortage.9
If we all want the same GPUs
The question the curve answers
None of this has to wait for an exchange. I would take the next GPU commitment due for renewal and write it as one row: hardware and configuration, region, term, price per GPU-hour, minimum spend, cancellation and transfer rights, and payment timing. Beside it go the nearest outside prices for the same period (the on-demand rate, a Capacity Block quote, a published index, a marketplace quote), each with what differs and roughly what the difference is worth. That page goes to procurement and Finance before the negotiation, with one question for Treasury: if a usable contract existed, could we hedge this exposure, and who would fund the margin? After signing, I would record our actual rate beside the index every month. That record is the basis estimate a future hedge will need, and it strengthens the next negotiation whether or not these exchanges ever become deep markets.
Architect Financial Technologies, “Architect Financial Technologies Partners with Compute Desk to Launch Exchange-for-Physical Market for GPU Compute”, July 8, 2026: the American Innovation Exchange is “forthcoming” and “pending regulatory review”; ComputeConnect would publish “standard basis tables for different SKUs, memory configurations, and locations.” Company announcement, checked October 4, 2026.↩︎
AWS, “Amazon EC2 Update — Streamlined Access to Spot Capacity, Smooth Price Changes, Instance Hibernation”, November 28, 2017, on the change from bidding to smoother provider-set prices.↩︎
AWS, Capacity Blocks and pricing and billing documentation, checked October 3, 2026, for the discount and billing rules described.↩︎
Compute Exchange, marketplace site, checked October 3, 2026: quotes through a “live auction across verified providers,” a services agreement signed directly with the chosen provider, and forward contracts on bilateral terms. Vendor descriptions.↩︎
Auctionomics and OCX Group, “Auctionomics and OneChronos Partner on First Tradable Financial Market for GPU Compute”, Business Wire, July 29, 2025: price discovery “through bi-lateral forwards”; “first” is the companies’ claim. OneChronos, “Compute Markets”, checked October 4, 2026: “Coming soon.”↩︎
San Francisco Compute Company, “About”, checked October 3, 2026: clusters leased “on contracts that let our customers sublease.” Vendor description.↩︎
CFTC, product record 62544, “Approval Pending (90),” checked October 4, 2026; Division of Market Oversight, letter to CME Group, September 21, 2026, extending review of NYMEX Submission No. 26-370 to November 9, 2026, for “novel or complex issues.” An extension is not a disapproval.↩︎
Loris Tools, “H100 Perpetual Futures”, checked October 4, 2026, 21:41 UTC: seven venues, $1.51 million of 24-hour volume, $80,800 of open interest, a $2.647 reference price, and 84.0% of volume on one venue. A third-party, unaudited aggregation of crypto-venue data; an order of magnitude only.↩︎
CME Group, “Learn about Basis: Grains”, checked October 3, 2026, and Self-Study Guide to Hedging with Grain and Oilseed Futures and Options, on basis (cash minus futures), long hedges, and margin. Applying grain mechanics to compute still requires checking the actual contract.↩︎