The problem RAW solves
Compute is becoming one of the most important infrastructure markets in the world. AI inference, GPU clusters, and high-performance compute facilities generate measurable output every hour; enterprises need access to that capacity; and financial markets are beginning to treat compute as a commodity. What’s missing is delivery verification.
Most compute delivery claims today are still proven through invoices, dashboards, screenshots, and trust between counterparties. That works for small bilateral relationships, but it isn’t strong enough for institutional settlement. A buyer needs to know what was actually delivered. A producer needs to prove performance beyond its own billing system. A financial contract referencing compute needs a physical delivery record it can converge on. Without that record, compute contracts stay dependent on self-reported data and manual reconciliation — which creates disputes, delays, trust discounts, and limited financing options for operators who aren’t hyperscalers. RAW turns delivery data into settlement-grade evidence.
How RAW works: connect, contract, deliver, settle
RAW follows a consistent sequence from physical output to settled contract.
Connect
RAW Box installs at the producer’s site and reads delivery telemetry from the inference serving stack, GPU management stack, or mining-pool verification source — as close to the origin as possible, before it becomes a commercial claim. On compatible hardware, readings are signed inside a Trusted Execution Environment, and the keys are generated inside the device and never leave it.
Contract
Buyer and producer enter a delivery contract using Daml templates on Canton Network, defining what’s delivered, how it’s measured, the price, the dispute window, and what happens on under-delivery. If verified data stops flowing, contract issuance and settlement stop with it.
Deliver
Signed evidence accumulates against the contract as compute is delivered — throughput and TTFT for inference, attested GPU-hours for compute, pool-confirmed hash rate for mining.
Settle
Once delivery is verified, settlement happens atomically on Canton Network: delivery validation, payment transfer, and fee deduction complete in one transaction. Payment moves directly from buyer to producer in regulated stablecoins. RAW never holds customer funds, and under-delivery reduces payout automatically.
What RAW verifies today
RAW currently supports three compute-related products, each measured and verified differently but settled through the same rail.
| Product | What’s measured | Verified by |
|---|---|---|
| Verified AI Inference | Millions of tokens per hour (input and output), with throughput and time-to-first-token attested at the edge | RAW Box, edge-signed |
| Verified GPU Compute | The attested GPU-hour: one hour of a specified accelerator, device class, and configuration disclosed at contract formation | RAW Box, signed telemetry |
| Verified Hash Rate | Continuously measurable mining output | Independent confirmation at the mining pool |
Why RAW Box and Canton Network
Compute delivery happens off-chain, and the systems that generate it — GPU management stacks, inference servers, mining pools — are usually controlled by the operator reporting on them. RAW Box changes the trust model by signing delivery evidence at the edge: it’s vendor-agnostic and tamper-evident, doesn’t require replacing existing infrastructure, and turns operator-controlled telemetry into a machine-verifiable record counterparties can use without trusting the producer’s own dashboard.
Settlement then runs on Canton Network for four properties that are difficult to combine elsewhere: sub-transaction privacy, institutional adoption already in production, atomic delivery-versus-payment, and regulated stablecoin settlement. Privacy matters because compute contracts carry sensitive information — buyer and producer identity, pricing, volume, configuration — so counterparties should see only the contracts they’re authorized to see. Atomic settlement matters because delivery validation, payment movement, and fee deduction need to complete together, not as separate steps that could fail independently. RAW is built for institutional counterparties from the start, which is why it settles where that capital already operates.
What this means for producers and buyers
For producers, RAW turns delivered compute into a more legible financial asset. A data center or inference provider may have strong operational performance that’s trapped inside private systems, invisible to buyers, lenders, or institutional counterparties without manual review. RAW gives producers a way to prove delivery under contract — supporting stronger commercial terms, access to institutional counterparties, and eventually financing against verified output. That matters especially for mid-market operators who don’t have a hyperscaler’s default credibility with capital.
For buyers, RAW reduces uncertainty around what was actually delivered, because signed evidence tied to a contract — rather than an invoice or a dashboard screenshot — is what settlement references. And for institutional counterparties more broadly, RAW is building the physical delivery mechanism compute financial markets are still missing: cash-settled compute futures and price indices can exist without physical delivery, but mature commodity markets eventually need a way to connect financial contracts back to real, delivered supply.
Bottom line
RAW Protocol is the verified-delivery settlement infrastructure that lets AI inference, GPU compute, and hash rate become contractable, measurable, and settleable on institutional rails. The core principle is simple — “no data, no contract.” If compute delivery can’t be verified, settlement shouldn’t move. If delivery is proven, payment can move automatically, privately, and non-custodially. That’s the foundation RAW is building for the future of the compute economy.