flâneur

Could Compute Become a Commodity? Lessons from Oil, Power, and Spectrum

formvfunction.substack.com · 2,613 words · saved by 1 readers

There is a big problem with compute. If you want serious compute capacity today (whether GPUs for training or CPUs for inference), you’re making phone calls. You get a price card from AWS or a specialized provider like CoreWeave, negotiate an SLA that doesn’t port to anyone else, and hope the terms hold when you need to scale. There’s no ticker, no clearing price, no easy way to comparison-shop. It works, but it’s slow, opaque, and hard to hedge. Bloomberg’s Felix Salmon argues that compute needs transparent pricing and liquid markets to facilitate transactions, suggesting that “making compute liquid and adding futures markets will expand investment, increase competition and rationalize AI demand.” And it may no longer be theoretical: Paul Milgrom, the Nobel Prize-winning architect of modern spectrum auctions, is working with OneChronos to build the first tradeable financial market for compute, tackling what they call “the world’s largest unhedged asset.” Meanwhile, startups like SF Co

There is a big problem with compute. If you want serious compute capacity today (whether GPUs for training or CPUs for inference), you’re making phone calls. You get a price card from AWS or a specialized provider like CoreWeave, negotiate an SLA that doesn’t port to anyone else, and hope the terms hold when you need to scale. There’s no ticker, no clearing price, no easy way to comparison-shop. It works, but it’s slow, opaque, and hard to hedge. Bloomberg’s Felix Salmon argues that compute needs transparent pricing and liquid markets to facilitate transactions, suggesting that “making…

related reading