A market needs a clear answer to a simple question: what does the buyer actually receive?
For compute, a quoted price is not enough. Device class, interconnect and delivery window determine whether capacity can do the work a buyer needs. A common delivery specification makes those claims comparable.
Start with the good
Gantry’s proposed unit is a dated capacity credit: one GPU-hour of a named device class, redeemable within a named delivery month. The specification defines the obligation before the claim changes hands.
Standardisation does not mean pretending every GPU is identical. It means publishing the differences that matter and making the contract explicit.
Verify capacity and claims
An inventory of machines does not prove that their future hours remain available. The verification problem has two sides: establish the capacity an operator controls, and account for the claims already issued against it.
The design combines capacity audits, issuance limits and ongoing checks. Buyer-signed usage receipts provide evidence of what was delivered. Hardware checks are intended to confirm that the actual device and connection match the specification.
Make the obligation matter
Operators post collateral against delivery. A published enforcement process is designed to draw on that collateral when delivery fails, with replacement compute sought before cash compensation.
The standard, the evidence and the consequences have to work together. That is the foundation for a claim a buyer can transfer with confidence, and eventually redeem for a running GPU.
Gantry is in development. Live operator deployment, runtime hardware probes and production enforcement remain work to be demonstrated.