Two verified products. One settlement rail.
RAW settles delivery across two asset classes — AI inference and GPU compute — both attested by the same on-site device and settled on the same Canton Network rail.
Built for asset classes that already generate continuous, measurable output
RAW settles two kinds of compute delivery — AI inference and GPU compute — using one attestation device and one settlement rail. Each product answers a different "did this actually happen" question for its buyers; both settle atomically in regulated stablecoins once verified.
Verified AI Inference
Measured in millions of tokens per hour — input and output — with throughput and TTFT attested at the edge.
Verified GPU Compute
The attested GPU-hour: one hour of a specified accelerator, device class, and configuration, delivered under contract by a named, attested asset. Heterogeneity is disclosed and priced at contract formation — not discovered after delivery.
| Product | Settlement unit | Verified at | Verification method |
|---|---|---|---|
| Verified AI Inference | 1M tokens (MTph) | Edge (RAW Box) | Signed telemetry; throughput and TTFT attested at inference |
| Verified GPU Compute | Attested GPU-hour | Edge (RAW Box) | Signed telemetry; device class and configuration disclosed at contract formation |
The world's largest derivatives exchanges launched compute futures. Verified physical delivery remains unclaimed.
Both futures markets are cash-settled against price indices. Neither specifies a delivery-verification mechanism. RAW is the physical leg.
Inference market figures: Grand View Research, 2025–2030 forecast.
Who settles on RAW
Monetize idle GPU capacity
Serve verified inference under your own brand while every delivery claim is independently attested at the edge before payment moves.
Turn utilization into a financial asset
A verified track record of delivered GPU-hours becomes portable, financeable proof — visible to lenders and institutional buyers, not trapped in private billing systems.
Buy output, not a price proxy
Get programmable, verified exposure to compute output instead of a cash-settled price index — on the same regulated rail your counterparties already use.