Skip to content
All insights
MANIFESTO

Why the AI hardware supply chain needs a forward market

May 12, 2026 | 1 min read | Rillor
FWD CURVE · B200 SYS · USD/UNIT

Allocation of every meaningful AI accelerator runs through a small handful of OEMs and an even smaller handful of NVIDIA business-unit managers. Buyers without a direct line in fight for residual capacity bilaterally, each call its own one-off negotiation. There is no public price discovery, no transferable inventory, no transparent forward curve. The market is not illiquid by accident. It is illiquid because the only path to physical delivery is a phone call.

Every commodity market this matters in has the same answer: a forward market with standardized contracts, KYC'd participants, escrowed settlement, and physical delivery. Oil did it. Wheat did it. Memory did it. The thing that's different about AI compute is that the underlying asset depreciates faster than almost anything else of comparable dollar value, which is exactly why the absence of a forward curve is so destructive. Procurement teams are pricing twelve-month buildouts against a number that exists in no public dataset.

Rillor's bet is that the readiness is now. Tier-2 clouds, sovereign AI programs, enterprise AI buildouts, and GPU-as-a-Service operators have crossed the threshold where they can no longer get hardware reliably through bilateral channels, and where their forward demand is big enough to justify standardized contracts. OEMs, for their part, want diversified demand-side risk without giving up margin or violating channel rules. Both sides want the same thing. They just don't have a venue.

This piece is the long version of why we're building one.

Sample preview. Full piece publishes shortly. Subscribe below to get it in your inbox.