Oct 20, 2026 · 37 min
Neutral markets aim to make compute tradable
The Programmable Data Center — Evan Conrad, SF Compute
Reliable physical markets could determine how compute capacity is priced, financed, and eventually turned into financial products.
- 1SF Compute evolved from cluster subleasing into operating the infrastructure required for physical settlement.
- 2Compute contracts remain difficult to standardize because timing, hardware, networking, location, and preemption materially change capacity.
- 3Neutral operators must establish trustworthy physical-market data before compute forwards can support broader financialization.
Don't miss
Evan Conrad describes virtualizing and programming data-center capabilities, including InfiniBand fabrics, so infrastructure can become financeable cloud capacity.
The brief
Evan Conrad explains why Nvidia needs an independent compute market but cannot run one itself: its supply, demand, and financing interests would compromise neutrality.
SF Compute has moved beyond subleasing clusters to operating the infrastructure and backplane needed to settle compute physically, much as physical operations anchor power and oil markets.
Compute is not a uniform commodity: timing, preemption, networking, location, hardware cohorts, burn-in, and degradation all shape what capacity is worth.
The marketplace uses multidimensional orders and flexible queries because customers need both custom GPU specifications and a way to trade before contracts become fully standardized.
Conrad’s central argument is that physical settlement must precede financialization; otherwise forwards risk relying on benchmarks detached from real capacity, as with manipulated LIBOR.
The episode’s defining idea is a programmable data center: neutral operators can turn infrastructure, fabrics, and sovereign-compute projects into financeable cloud capacity.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.