AI has a GPU pricing problem. Silicon Data wants to fix it.
Summary
Steve Howe, Head of Research at Silicon Data, discusses the upcoming launch of GPU rental futures on the CME. He explains how these derivatives allow hyperscalers and AI labs to hedge compute costs, noting that strong inference demand is keeping older chips like the A100 valuable. Howe also details Silicon Data’s Series A funding and its mission to create transparent pricing indices for the AI infrastructure market.
Topics discussed
Introduction: Steve Howe and Silicon Data
Explaining futures contracts and hedging
Why compute can be a tradable commodity
Market participants: Hyperscalers, AI labs, and market makers
Liquidity, financing, and hedging infrastructure risk
Counterparty risk and the 'giant baby' industry
Data sourcing and normalizing GPU rental prices
Rising demand for older chips like the A100
Political pushback and physical constraints on data centers
Term structure trends and NeoCloud vs. Hyperscaler pricing
Series A funding and future physical delivery plans
Listen ad-free on Castria