Why GPU price discovery is fragmented
AI infrastructure is one of the largest capital build-outs in the economy, with capital expenditure projected on the order of $725 billion in 2026. Yet the hardware at the center of it has no shared reference price.
The problem
Unlike commodities, equities, or even used cars, enterprise GPUs have no public benchmark for what a given configuration is worth. Price lives in private quotes between parties that each see only their own flow.
Why it stays fragmented
Value depends on SKU, condition, location, quantity, timing, and counterparty. The same hardware can clear at materially different levels depending on who is buying and how quickly. Because trade happens bilaterally, no single party sees enough of the market to form a reliable picture.
The cost lands on everyone at once. Buyers and sellers negotiate without a reference. Lenders and insurers underwrite collateral they cannot independently value. Operators plan fleet rotations on incomplete information.
What changes it
Concentrating supply and demand in one venue, and recording where hardware actually clears, turns scattered private quotes into observable market signal. That is the foundation a reference layer for AI hardware is built on.
Source: capital-expenditure figure per AL Capital Advisory, AI infrastructure capex analysis (2026).
Request access to Stoa Markets