Why GPU price discovery is fragmented
Data-center capex is headed for the trillions, yet the hardware at its center has no shared reference price. Why, and what changes it.

AI infrastructure is one of the largest capital build-outs in economic history. McKinsey estimates data centers will require some $6.7 trillion of capital spending worldwide by 2030, most of it for AI-ready capacity. Yet the hardware at the center of that spend 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.
The asset class is not small. Omdia puts AI data-center chip shipments at $123 billion in 2024, on a path it projects to reach $286 billion by 2030. Markets that size normally have visible pricing. This one does not.
Why it stays fragmented
Value depends on the configuration, its condition, where it sits, how much of it there is, and who is on the other side of the trade. The same hardware can clear at materially different levels depending on who is buying and how quickly. Because trades are struck one pair at a time, no single party sees enough of the market to form a reliable picture.
Fragmentation compounds itself. When price evidence is private, every participant builds a view alone, discounts for what it cannot see, and passes that discount on as wider spreads, slower procurement, and more conservative credit.
What it costs
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.
The market still clears, but slowly and at a premium, and the cost of capital for AI hardware stays higher than the fundamentals warrant.
What changes it
When supply and demand meet in one venue instead of a hundred private conversations, buyers can compare offers, sellers bid against each other, and no participant has to reconstruct the market from its own flow.
Stoa Markets concentrates that flow in one gated venue, where qualified buyers and dealers transact and the view of value comes from real trades rather than estimates.
Sources: McKinsey, The cost of compute (2025); Omdia, AI processors for cloud and data center forecast (2025).