Pantera's Jay Yu says compute markets are still early, with GPU trading largely off-exchange

News Editor

2026-09-10 00:17:22

Pantera Capital junior partner Jay Yu said in his essay, The Rise of Compute Markets, that compute and data center spending has already become a trillion-dollar category, yet GPU procurement still happens mostly through group chats, over-the-counter brokers, and bilateral agreements. He argued that the financialization of compute remains at an early stage because markets are constrained by SKU differences, time, and location. Over the next five to 10 years, he said, compute could evolve into a commodity-like asset class comparable to electricity or oil. Yu also outlined a possible market structure modeled on the power sector, describing a future stack of hardware, suppliers, and clusters. In that framework, he characterized new cloud providers as structurally short GPUs, while on-demand platforms and the application layer are long. He added that for every $100 spent by the application layer on inference, about $45 goes to the on-demand layer, about $50 goes to new cloud providers or the GPU layer, and about $5 goes to routing layers such as OpenRouter.

ChainCatcher reported that Pantera Capital junior partner Jay Yu, in an essay titled The Rise of Compute Markets, said compute and data center spending has become a trillion-dollar category, but GPU procurement is still handled mainly through group chats, over-the-counter brokers, and bilateral agreements.

Financialization remains in its early stage

Yu said the financialization of compute is still early and is constrained by SKU, time, and location. He added that over the next five to 10 years, compute could develop into a commodity asset similar to electricity or oil.

A market structure modeled on the power grid

He said compute markets may develop a structure similar to the power sector's "grid-operator-node" model, recast as "hardware-supplier-cluster." In his view, new cloud providers are structurally short GPUs, while on-demand platforms and the application layer are long.

Yu said that for every $100 the application layer spends on inference, around $45 flows to the on-demand layer, about $50 goes to new cloud providers or the GPU layer, and about $5 goes to routing layers such as OpenRouter.

NVIDIA as the compute market's "central bank"

He also described NVIDIA as the compute market's "central bank" and said the company has announced residual value support of up to 25%.

Yu added that physical delivery has a more durable moat, while indexes, cash-settled exchanges, and financing tools are likely to appear later. He named SF Compute, Vast AI, Runpod, and Compute Exchange as participants.

He also said the compute market does not face mandatory price transparency requirements, which could leave basis risk higher.

This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan.

200

Disclaimer:

The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.

Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.