Infrastructure assessment guide
How to Run an Infrastructure Risk Assessment in 2026
To run an infrastructure risk assessment, name the decision it supports, reconcile the assets and services in scope, map dependencies, validate assumptions with operators, evaluate likelihood and impact, and assign actions with owners and review dates. Keep evidence gaps visible rather than filling them with a reassuring score.
This workflow covers service continuity and physical and cyber dependencies. It can inform a data center project, but an assessment of an operating estate is not the same as screening a candidate parcel. Sources reviewed November 9, 2026.
1. Write a decision-focused charter
“Review infrastructure” is too broad to guide the work. A decision might be whether to fund a second network route, renew a facility, accept a supplier dependency or retain a site on the shortlist. Name the sponsor, assessment owner, scope, time horizon and due date.
Define critical services and tolerable disruption before assigning scores. Recovery objectives should come from the service and its obligations, not a generic industry number. Record exclusions, such as an untested supplier environment or building system that another workstream will assess.
2. Reconcile the inventory and its exceptions
Gather asset registers, diagrams, contracts, configuration records and operating documentation. Link assets to a service and accountable owner. For important records, capture identity, location, lifecycle status, support arrangements and evidence date.
Do not force contradictory records into a false single truth. Maintain an exceptions list showing the conflict, its owner, the evidence needed and whether it blocks the decision. An inaccessible backup device, an unknown carrier route and an undocumented utility dependency should remain visible.
3. Map dependencies from the service outward
Trace how each service depends on applications, identity systems, facilities, power, cooling, communications, suppliers and people. Mark shared dependencies that can defeat nominal redundancy.
- Two circuits may share a carrier duct or crossing.
- Two applications may rely on the same identity provider or database.
- Redundant equipment may share a switchboard, cooling loop or maintenance dependency.
- Multiple sites may be exposed to the same regional hazard or supplier failure.
Record the plausible failure scenario, detection method, time to impact, workaround and recovery path. A diagram that shows two boxes is not evidence that failover has been tested.
4. Validate with operators and documentary evidence
Ask service owners what actually runs, which records are obsolete, what unofficial workarounds exist and what happened during the last relevant incident or test. Match important answers to configurations, maintenance records, contracts and test results where possible.
Use corroboration proportionate to the decision. A second conversation can expose a misunderstanding, but a fixed rule of two interviews does not guarantee correctness. Record unresolved disagreements and distinguish documented capability from an operator's expectation.
5. Evaluate risk consistently without false precision
Define a time horizon, credible scenario, likelihood basis and impact dimensions. Impacts may include safety, service disruption, financial loss, contractual obligations and environmental consequences. Keep inherent exposure separate from existing controls, residual risk and confidence in the evidence.
A likelihood-by-impact matrix can support a workshop, but its ordinal numbers are not probabilities or expected dollar losses. Multiplying labels does not make different risks quantitatively comparable. Document the rationale and allow a critical safety or legal constraint to override a simple ranking.
Use sensitivity ranges where estimates are uncertain. A risk with weak evidence may require investigation rather than a lower score. Record who can approve acceptance and whether the residual risk stays within the organization's criteria.
6. Use frameworks for their actual purpose
The NIST Cybersecurity Framework 2.0 organizes cybersecurity outcomes through Govern, Identify, Protect, Detect, Respond and Recover. It can structure current and target profiles and discussions with leadership. It is not a general physical-engineering standard or a certification that every infrastructure dependency is safe.
ISO/IEC 27001 concerns information security management systems. Certification within a defined scope is different from validating the availability of a utility connection, structural resilience or every supplier. Select the framework based on scope and obligations; the charter, evidence and operational checks remain necessary.
7. Turn findings into owned actions
| Field | Purpose |
|---|---|
| Service and scenario | Connect the finding to a plausible interruption or harm |
| Evidence and uncertainty | Show source dates, assumptions and gaps |
| Existing controls | Distinguish an installed or tested control from a plan |
| Residual exposure | Describe what remains after controls |
| Response and owner | Reduce, avoid, transfer where possible, or explicitly accept |
| Milestone and closure test | State what proof is needed and when |
Write actions as outcomes: “demonstrate failover over a physically separate route,” not “review networking.” Include estimated cost, dependencies and the approving owner. Insurance or a contract can allocate some financial consequences; it does not remove the underlying interruption or every liability.
Close an action only when its evidence meets the acceptance test. A purchase order is not a successful recovery exercise. A proposed supplier commitment is not an executed agreement.
8. Review changes and test recovery
Set review triggers for new facilities, major load changes, supplier changes, incidents, significant configuration changes and expiring support. Use scheduled reviews appropriate to service criticality rather than assuming an annual review is always enough.
Reconcile the inventory, test relevant failover, review overdue actions and reassess correlated dependencies. Compare actual incidents with the scenarios considered; do not describe every assessment as a prediction or treat a period without incidents as proof that the controls work.
Where AI-assisted site screening fits
Automated tools can organize public evidence and compare candidate locations consistently. They cannot inspect a live plant, validate every dependency, certify cybersecurity controls or replace operator knowledge.
AI Data Center Impact's Site Viability Scorecard is a pre-development location screen. Grid and power, water, climate, regulatory and community evidence help identify questions for deeper diligence. Supplemental emergency-facility and receptor information is context, not proof of response time, acoustic compliance or a site's operating readiness.
Use the site selection checklist for candidate qualification and the power diligence guide for utility dependencies. Keep the operational risk register and the parcel-screening report as related but distinct records.
Frequently asked questions
How is a vulnerability assessment different?
It identifies weaknesses. A risk assessment connects scenarios and weaknesses with likelihood, consequences, controls and decisions. Findings from one can inform the other without making them interchangeable.
Should cyber and physical risks use one register?
They can share governance and decision criteria, but their evidence and analysis methods may differ. Preserve those differences rather than assuming the same numeric score makes them equivalent.
Who owns the report?
Name an accountable sponsor for the decision and an owner for maintaining the assessment. Assign individual findings and acceptance decisions to the appropriate authorized people.
Does a framework replace specialist review?
No. Information-security, electrical, structural, environmental, legal and safety questions may need different qualified specialists and evidence.
This guide is a practical management workflow, not a certification, engineering opinion or assurance of loss prevention.