Reviewing the facility behind the loan
Review the operating envelope and the lender's practical access before treating colocation as a settled input.
A facility review in eight questions
A browser-session review aid, not a credit score or approval. Ticking a question records your review; it does not verify the underlying evidence. Inputs are not saved.
Start with the specified system
A facility's headline megawatts do not establish whether it can host a particular cluster. Reconcile the equipment design with usable capacity at the rack, cooling arrangement, floor loading, network paths and the relevant commissioning date. Ask whether the committed capacity remains available under the redundancy conditions the customer expects.
NVIDIA's DGX B200 documentation illustrates why the system matters: it specifies eight B200 GPUs and maximum system power of 14.3 kW. That is a particular server specification, not a universal rack density or an average electricity bill. A facility assessment needs the actual rack population, operating profile and supporting infrastructure. Avoid applying one generic cooling threshold to every design.
Read the facility agreement as a credit document
Map the start of billing, committed power, escalation, service credits, renewal windows and termination costs. Identify deposits and amounts that may remain payable if the customer contract ends. A fixed GPU revenue contract can leave a margin mismatch if facility costs change on a different basis.
Then examine what happens after an unpaid facility invoice. Notice, cure, access, removal and any provider security or lien rights depend on the contract and governing law. A lender's equipment lien is not a complete operating plan. Any negotiated access arrangement should explain the permitted actions and payments rather than relying on a general promise of cooperation.
Separate evidence received from risk resolved
American Compute's due-diligence workbook covers borrower, customer, equipment, facility, insurance, contracts and monitoring. Its completeness calculation counts items marked received or not applicable. That is useful for document collection, but it is not a risk score: receiving an adverse contract does not resolve the adverse term.
Use separate fields for document status, assessment, unresolved issue, owner and approval condition. A material gap should remain visible even when every document has arrived. Facility certification also needs its exact scope checked; it should not be converted into a guaranteed number of outage hours or used instead of operational evidence.
Test the handover
Ask the parties to explain a realistic incident: the operator cannot pay, a component fails or a customer requests a service change. Who can enter, who can instruct technicians, and what permissions are required to continue serving the customer? Compare those answers with the executed contracts and available funds.
The final review should identify what prevents a draw, what can be monitored after funding and what requires a priced or limited exception. This is more useful than a high completion percentage with no indication of which unresolved matter can stop the cluster earning.
Questions for the credit paper
- Is capacity committed at the rack and at the required date?
- What changes under an outage or maintenance condition?
- What can the lender actually do after non-payment?
- Which received documents still contain unresolved risks?
About this guide
Original InfraCompute analysis informed by the American Compute material below. The article structure, examples and interactive visuals are InfraCompute’s work. Source-specific findings remain attributed to American Compute; attribution does not imply affiliation or endorsement.
- American Compute, Data Center Due Diligence Checklist for GPU Loans ↗
October 2026 diligence workbook. Link opens the related appraisal service, not a public workbook download.
Selected primary references
We reviewed the published edition and selected public references. We have not independently verified American Compute’s proprietary datasets, every market claim or confidential financing terms. Frameworks and hypothetical models describe questions to investigate; they do not establish protections in a particular financing. Legal and engineering conclusions depend on the relevant documents, system and jurisdiction.
Read our research methodology →