Pantera Capital says compute could become a hard currency for the AI economy
Pantera Capital partner Jay Yu argues that GPU compute is still in the early stages of financialization, but its market structure may develop in ways that resemble electricity and other physical commodities. In his view, compute is constrained by chip type, time, and geography, which makes it heterogeneous rather than perfectly fungible, yet still capable of evolving into a globally traded asset class over the next five to 10 years. The article maps the sector through layers including hardware providers, compute service operators, clusters, inference platforms, application companies, brokers, OTC desks, index builders, and derivatives venues.
Yu draws repeated comparisons with U.S. power markets, where physical delivery, benchmark formation, and risk transfer tools emerged over time. He suggests compute markets may follow a similar path, with physical GPU delivery venues anchoring index construction and futures trading. The piece also describes Nvidia as a possible “central bank” for the compute economy because of its control over chip release cycles, utilization dynamics, and residual-value support policies. It identifies four broad product categories already taking shape: physical delivery, index products, derivatives exchanges, and financing tools such as lending, treasury structures, synthetic stablecoins, insurance, and other risk-transfer products. At the same time, it says the sector still faces major issues around transparency, basis risk, standardization, and quality verification.
