The episode explores the creation of a futures market for GPU compute, driven by the immense costs of AI infrastructure. Steve Howe from Silicon Data explains how futures contracts can help AI labs, cloud providers, and investors hedge against volatile GPU rental prices. The discussion challenges assumptions about GPU depreciation and highlights surging inference demand.
Summarized by Podsumo
Silicon Data plans to launch compute futures on the CME (pending regulatory approval), with an H100 and B200 rental index, aiming to bring price transparency and hedging to the GPU market.
Steve Howe argues that older GPU chips like the A100 are seeing strong rental demand due to rising agentic AI, challenging the narrative that AI hardware rapidly depreciates.
The market currently prices a discount for long-term GPU rental contracts, but Steve notes rental rates across all contract lengths have been increasing, signaling robust demand.
Silicon Data's Series A round raised $30 million from investors including Valor, Fidelity, CME Ventures, and Samsung Next, to build an ecosystem for AI compute financing and eventually physical delivery of compute in futures contracts.
The episode highlights the scale of the AI build-out, with hyperscalers spending $750 billion on compute this year and a projected trillion next year.
"I think this year is really what the first time that AI has become useful."
"These things may not depreciate as quickly as people thought and may have actually a longer lifespan as people had previously budgeted for."
"The existence of a futures market will just bring about more pricing transparency and everyone can actually observe and with greater certainty."