Per-second billing for a globally distributed GPU pool is a payments problem before it's a compute problem. A look at the settlement loop.
Per-second billing for a globally distributed GPU pool is a payments problem long before it's a compute problem. A 1,024-world benchmark touches dozens of nodes for a handful of seconds each; charging and rewarding them through a conventional rail would cost more in fees than the compute itself.
That constraint is why settlement runs on Robinhood Chain, an Ethereum L2 with ~100ms blocks. Sub-cent fees and fast finality let us aggregate millions of GPU-second events into operator payouts without the fee drag that would make micro-metering pointless on a dollar-fee chain.
Each completed job emits an execution receipt — node, duration, verified completion — that feeds the settlement loop. Operators are paid per GPU-second served, weighted by uptime. Builders see the exact spot price they paid, down to the microdollar.
The next milestone is verifiable execution proofs, so a buyer can confirm the work they paid for happened as billed without trusting the scheduler.