Key Takeaways

  • Evan Conrad argues that pure cash-settled compute derivatives create severe basis risk because spot indexes track thin, unrepresentative transaction data.
  • Physical settlement requires SF Compute to programmatically virtualize data center clusters and subdivide InfiniBand fabrics so buyers can take actual hardware delivery.
  • Conrad compares unbacked compute futures to the LIBOR scandal, where traders manipulated interest rates because contracts referenced self-reported quotes instead of real trades.
  • Physical oil pipelines had to exist before paper oil futures could work, providing an arbitrage backstop that keeps paper prices anchored to reality.
  • Hyperscalers and NVIDIA cannot operate a neutral exchange because their margin models make them direct counterparties with incentives to prevent secondary trading.

Why Paper Compute Markets Fail

Fintech founders regularly propose financial exchanges for GPUs. The pitch sounds clean: create a cash-settled contract against an index of hourly GPU rates, let AI labs hedge their cluster spend, and let funds speculate on demand.

Conrad points out the fatal flaw: without physical delivery, the index drifts away from reality. “If you try to build too much financialization before the industry has had time to build out the physical settlement, then I think what you have is a market that's very easy to manipulate and a market that has a lot of basis risk where the price you may be hedging against actually has nothing to do with the price that you can actually buy the capacity for.”

This failure mode has precedent. Conrad points to LIBOR, the benchmark interest rate that unraveled because it relied on estimates submitted by banks rather than actual borrowing transactions. GPU indexes face the same weakness today. When an index tracks a tiny volume of fragmented spot transactions, a trading desk can move the benchmark with minimal capital while actual reserve contracts trade at completely different levels. If you hedge on a paper index, you might win your financial trade and still fail to secure real machines.

The Pipeline Must Precede the Paper Contract

Commodity markets require logistics before they can sustain paper trading. Conrad uses crude oil to explain why physical delivery determines whether an exchange survives. “In oil, you have to have pipelines. If you would like the oil to be delivered to you, there has to be a way for it to get to you.”

Oil futures work because buyers can demand physical delivery at Cushing, Oklahoma. If futures drop below spot prices, traders buy the contract, take delivery through the pipe, and sell the oil for a profit. That arbitrage forces paper prices to reflect reality.

To make compute trade like oil, SF Compute spent years building cloud software rather than trading screens. “What we spent the last several years doing is building the entire operating capacity of a cloud in order to actually deliver the compute at a reliable level. We call that physical settlement.”

That work requires software to programmatically virtualize private data centers and subdivide InfiniBand fabrics. Dividing an InfiniBand network dynamically lets an exchange carve an 8-way or 64-way H100 block out of a larger cluster, hand it to a buyer, and reclaim it when the term expires. Hyperscalers like AWS cannot offer this neutrality because their margins depend on captive enterprise lock-in.

Cash Settlement Belongs at the End of the Roadmap

Conrad does not oppose cash-settled derivatives forever. He opposes building the roof before the foundation. “So, to be clear, we think cash settlement as a future concept is a great idea. We just think it should be derived from physical settlement. Because, as you said, it's a side bet, and it would be kind of like if you made a cash settlement future before you made physical settlement.”

When physical settlement works, a buyer books a monthly forward contract for 128 GPUs, receives programmatic access, runs workloads, and hands back the cluster. Once thousands of physical trades clear, that tape creates a reliable clearing price. Only then can institutions write cash-settled derivatives that track reality.

What to Do With This

Stop evaluating synthetic compute hedges or unbacked GPU tokens. If you need to manage compute price risk this quarter, audit your provider contracts to confirm they include physical delivery guarantees, automated network reprovisioning, and enforceable SLA penalties for non-delivery.