How it works
A structured verification process for AI infrastructure
VerifyCompute follows a defined sequence: intake, evidence collection, technical review, reporting, and lifecycle recordkeeping.
How it works
How VerifyCompute Works
Each stage produces evidence that can be reviewed by buyers, sellers, institutions, consultants, and internal stakeholders.
Step 1
Submit Asset Details
Provide the hardware, quantities, location, and transaction context under review.
Step 2
Upload Technical Evidence
Supply invoices, serial lists, diagnostics, firmware and driver output, thermal logs, and photographs.
Step 3
Review Configuration and Risk Signals
Evidence is reviewed against expected configuration and known risk signals for the hardware class.
Step 4
Generate Verification Report
A structured report documents findings, risk flags, confidence level, limitations, and next steps.
Step 5
Maintain Lifecycle Record
Verification results and subsequent events are retained as an ongoing record for the asset.
VerifyCompute reports are evidence-based and do not guarantee future performance, financing approval, insurance approval, resale value, useful life, or transaction outcome.
Assurance levels
A tiered verification model
Assurance depends on what the hardware itself can prove. We work from the highest available level down, and every finding is labeled with the tier that produced it.
Hardware attestation
Where the device supports it, we collect a cryptographically signed attestation report and device certificate chain directly from the hardware, validate the certificate chain, and compare runtime measurements against the manufacturer's published reference measurements. This establishes that the silicon is genuine and running authentic firmware, anchored in a hardware root of trust.
Signed collector telemetry
Where hardware attestation is unavailable, we collect thermals, error counts, utilization, throttling events, and benchmark results through a signed collector, then cross-check the results for internal consistency against the declared configuration.
Physical inspection
Where neither method can establish component authenticity, on-site inspection is available as an add-on.
Every report states which tier was used for each finding. Evidence that cannot be validated at any tier is reported as Unverified.
Architecture coverage
AI infrastructure increasingly runs mixed silicon from multiple vendors. Our verification scope covers NVIDIA hardware today, with attestation support where the device and driver provide it. Coverage for additional architectures is on the roadmap, and every report states which methods were available for the hardware submitted.
Boundaries
Where attestation ends
Hardware attestation confirms that a device is genuine and running authentic firmware at the moment it is checked. It does not record who has owned the device, where it has been, whether it was removed from a fleet without authorization, or whether surrounding components were substituted. Attestation answers identity. VerifyCompute records history.
Scope
What verification covers, and what it does not
Clear boundaries are part of the product. Reports describe the evidence reviewed and state the limits of that review.
Included in review
- Asset identity and declared configuration
- GPU and memory module inventory against supplied documentation
- Firmware and driver versions as evidenced
- Thermal and diagnostic output supplied by the holder
- Documented risk signals and recommended follow-up
Not covered
- Valuation, appraisal, and residual value estimates
- Price guidance or resale forecasts
- Generational performance comparisons or economic fitness for a given workload. A VerifyCompute grade describes condition and authenticity. Whether a unit suits your workload is a separate question.
- Manufacturer warranty claims
- Guarantees of future performance or useful life
- Financing or insurance approval decisions
- Findings beyond the evidence supplied at assessment time
Evidence validation
Submitted evidence is validated against device attestation reports and certificate chains where the hardware supports it, and against collector signatures and internal consistency checks where it does not. Results that cannot be validated are reported as Unverified rather than accepted.
Disclaimer
VerifyCompute provides evidence-based review, verification workflows, and reporting for AI infrastructure assets.
VerifyCompute does not guarantee future hardware performance, useful life, resale value, financing approval, insurance approval, or transaction outcomes.
All reports are based on available evidence, documentation, diagnostic data, and review procedures at the time of assessment.