Physical design, power, facility, compute, contracts, and debt models are reviewed by different teams with different definitions of completion and usable capacity.
Research note 01Method v0.4July 2026
Discuss a financingResearch utility
Share, print, or preserve this version for an internal credit review.
AI infrastructure finance / Technical credit intelligence
Capital is underwriting the buildout. Technical reality still sits between the models.
Independent technical credit assessment and surveillance for AI infrastructure finance.
AI infrastructure is financed as real estate, project finance, power infrastructure, equipment finance, technology credit, and structured finance. Repayment depends on all six operating as one system.
- Find inconsistent technical assumptions before approval.
- Translate evidence gaps into explicit credit protections.
- Preserve the approved case through construction and operations.
Two-minute decision view
Underwrite the causal chain, not six disconnected reports.
One evidence-linked record that shows which technical fact changes completion, billable performance, cash flow, collateral, or a required credit protection.
A lender-facing technical credit assessment before close, followed by construction and operating surveillance against the approved repayment case.
Primary champion
Head of Infrastructure Credit, underwriting MD, portfolio manager, or technical risk lead
Decision authority
Lender credit approves the action; accountable engineers and counsel retain their authority
Diligence budget
The proposal names whether lender, arranger, borrower, sponsor, or advisor pays the transaction cost
Initial coverage
North American AI infrastructure; jurisdiction-specific conclusions require scoped local experts
01 / Market pressure
The capital requirement is scaling faster than the underwriting system around it.
These estimates measure different periods and scopes. They are shown independently to establish the direction and magnitude of the financing challenge, not as additive figures.
AI data-center capital spending through 2030 in a McKinsey scenario
- Scope
- Global
- Period
- Through 2030
- Measure
- Accelerated capex scenario
Estimated financing gap through 2028 in Morgan Stanley research
- Scope
- Global
- Period
- Through 2028
- Measure
- Financing-gap estimate
Average 2026 global shell-and-core construction cost per megawatt
Technology fit-out is excluded from this figure.- Scope
- Global
- Period
- 2026 average
- Measure
- Shell-and-core cost / MW
U.S. data-center credit facilities and bonds identified in 2025
- Scope
- United States
- Period
- 2025 commitments
- Measure
- Facilities and bonds
02 / The missing layer
One asset. Six underwriting systems.
Each lens is necessary. None independently proves whether the financed system can deliver reliable, billable compute at the cost and time assumed by the repayment case.
Commercial real estate
Site, lease, tenant, residual value
Project finance
Completion, contracts, cash-flow waterfall
Power infrastructure
Firm capacity, tariff, interconnection
Equipment finance
Title, deployment, value, recoverability
Technology credit
Utilization, performance, obsolescence
Structured finance
Pool quality, triggers, surveillance
Repayment does not begin with a spreadsheet.
It begins with a causal system.- Physical design
- Cluster performance
- Service delivery
- Customer economics
- Cash flow
- Debt protection
03 / Credit questions
Five questions the technical record must answer.
Can the asset be delivered as represented?
Reconcile power, facility, cooling, network, compute, schedule, cost, and acceptance evidence into one completion case.
Will it perform the intended workload economically?
Test deployable capacity, efficiency, reliability, utilization, service levels, and the cost of delivering usable compute.
Will the economics remain durable through the debt tenor?
Examine hardware competitiveness, customer concentration, contract structure, operating cost, and reinvestment requirements.
Which terms should absorb the uncertainty?
Translate technical findings into conditions precedent, milestones, reserves, covenants, reporting, and intervention rights.
How will the lender know when the case changes?
Define evidence-linked indicators that expose deterioration or improvement before a payment event makes the problem obvious.
04 / Risk architecture
Technical risk is a system of dependencies, not a checklist.
A useful risk model preserves the underlying domain detail while showing how one fact can alter completion, operating economics, covenant headroom, and recovery.
A
Delivery and governance
Can the represented project become an accepted operating asset?
- Sponsor
- Site
- Construction
- Model integrity
B
Power and facility
Is usable capacity firm, timed, resilient, and economically viable?
- Power
- Tariff
- Electrical
- Cooling
- Environment
C
Compute delivery
Can the configured system produce the promised service?
- Network
- Compute
- Software
- Workload
D
Commercial durability
Does technical performance convert into dependable revenue?
- Contracts
- Counterparties
- Concentration
E
Operations and protection
Can performance and value be protected after completion?
- O&M
- Cyber and physical
- Insurance
F
Collateral and recovery
What can creditors control, preserve, or recover under stress?
- Title
- Location
- Portability
- Secondary value
05 / From fact to credit action
The product is not another diligence report. It is a traceable decision record.
Illustrative risk chain
A utility delivery date changes after critical equipment has been committed.
- 01Observed fact
Firm power availability moves beyond the scheduled acceptance date.
- 02Technical consequence
Commissioning sequence, equipment warranty, and deployable capacity diverge.
- 03Financial exposure
Interest carry, minimum utility charges, and delayed revenue consume liquidity.
- 04Credit response
A revised milestone, contingency reserve, reporting trigger, and cure plan.
Required discipline
Root cause → evidence → technical event → commercial effect → financial exposure → mitigant → credit term → residual risk
Illustrative decision record / TC-014Power delivery date versus equipment acceptance
Review requiredSchema v0.3
Evidence cutoffIllustrative snapshot
Source authorityUtility and accountable engineers
Review ownerAssigned by lender credit
Change controlVersioned evidence and disposition log
- Evidence
- Utility milestone letter, integrated schedule, equipment purchase order, acceptance test plan
- Observed fact
- Firm energization is forecast 94 days after the current equipment acceptance milestone.
- Confidence
- Medium, pending utility confirmation and schedule update
- Base-case effect
- Commissioning sequence moves, warranty exposure increases, and first billable capacity is delayed.
- Proposed credit action
- Revised completion milestone, funded carry reserve, monthly utility evidence, and cure plan before the next draw.
- Decision owner
- Lender credit retains approval. Utility and responsible engineers remain authoritative for technical facts.
- Change history
- Prior state, changed field, source, timestamp, reviewer, disposition, and remaining uncertainty are preserved.
- Closure evidence
- Updated utility confirmation, accepted schedule, reserve funding evidence, and decision-owner approval.
This is an illustrative schema, not a customer result, rating, engineering conclusion, or investment recommendation.
06 / Explainable grading
A grade should reveal the repayment logic, not hide it.
The weighting below is an initial blueprint. Every result must remain decomposable to source evidence, confidence, reviewer judgment, and the financial consequence of being wrong.
Current qualityCompletion riskEconomic durabilityRecovery quality
Delivery and completion18%
Power and utility path18%
Compute and network15%
Facility and cooling13%
Revenue technical quality12%
Operations and software10%
Collateral and recovery8%
Governance and reporting6%
07 / Illustrative sensitivity
Small technical variances can become material credit outcomes.
Consider a 100 MW facility operating at an 85% load factor. A $10/MWh variance in effective power cost changes annual economics by about $7.45 million before considering knock-on effects.
Illustrative only. Actual exposure depends on tariff, load shape, curtailment, contract pass-throughs, efficiency, and operating state.Annual energy744,600MWh
Cost variance$10per MWh
Annual exposure$7.4M$620,500 monthly equivalent
Formula: capacity × 8,760 hours × load factor × effective cost variance. This isolates one sensitivity and is not a forecast, valuation, or credit conclusion.
08 / Institutional output
What a capital provider should be able to use.
The output must reduce uncertainty at the decision point and remain useful after close. Every assessment should create a reusable data object, not merely a static PDF.
Evidence-linked assessment
Every conclusion identifies the source, date, confidence, calculation, reviewer, and unresolved evidence gap.
Capacity and economics reconciliation
Contracted, energized, commissioned, deployable, utilized, and billable capacity are separated rather than treated as one number.
Credit-term translation
Findings become proposed conditions, milestones, reserves, covenants, reporting requirements, and escalation thresholds.
Continuous surveillance record
Construction evidence, operating telemetry, contracts, observations, risks, and covenant status remain connected after close.
Pre-closeIndependent technical credit assessment
Screen, reconcile, quantify, structure, and define closing conditions.
ConstructionMilestone and draw surveillance
Test evidence, dependencies, forecast changes, and completion risk.
OperationsContinuous risk monitoring
Connect telemetry, contracts, covenants, exceptions, and intervention.
PortfolioComparable risk intelligence
Normalize exposure, performance, counterparties, and emerging patterns.
09 / Product system
The assessment creates a reusable operating record, not a one-time consulting file.
Doku Infrastructure is the company. Doku Basis is the project operating system. This technical credit application uses the same evidence, dependency, action, owner, and decision model for lender workflows.
Illustrative product surfaceOne financing / evidence-to-action workspace
Schema v0.4 / synthetic example / no customer dataEvidence queue4 items require disposition
- ReviewUtility milestone letterOfficial letter / changed 2 days ago
- ConflictIntegrated project scheduleAcceptance date differs by 94 days
- PendingCluster acceptance planResponsible expert sign-off missing
Decision chain / TC-014Power date changes equipment acceptance
- Fact
Firm energization moves after the accepted delivery case.
- Exposure
Carry, warranty, commissioning, and revenue timing change.
- Action
Reserve, revised milestone, reporting trigger, and cure plan.
Owner: Lender creditConfidence: MediumReview required
Portfolio exceptionsNormalized fields, preserved source context
GPU-01Power pathEscalate
DC-02Cooling acceptanceMonitor
GPU-03Customer concentrationReview
This is an illustrative interface and data model, not a screenshot of a completed customer engagement or a claim of production deployment.
Core data model
Eight connected entities preserve the decision.
- Asset
- Capacity state
- Evidence
- Risk
- Action
- Covenant
- Owner
- Decision
Begin with controlled manual intake. Add structured upload, CSV, API, or source-system access only when the accepted scope permits it.
Extract fields, compare versions, expose contradictions, and preserve lineage without silently changing authoritative evidence.
Accountable reviewers approve conclusions, overrides, owners, conditions, and residual uncertainty. Software never grants approval.
Export an approval memo, structured risk register, CSV/API record, dashboard view, and surveillance update from the same source model.
10 / Existing stack
Doku connects the work already being done. It does not claim the authority of the systems or specialists it depends on.
The gap appears after a source specialist reaches a conclusion but before every affected credit assumption, condition, owner, and surveillance obligation has been reconciled.
| Approach | Best at | Common stopping point | How Doku connects |
|---|---|---|---|
| Engineering report | Design facts, testing, technical opinion | Cross-domain credit consequence and ongoing decision history | Doku references the responsible engineer’s output and preserves its authority. |
| Transaction or Big Four diligence | Broad workstreams, commercial review, transaction process | Persistent technical lineage from source fact to surveillance action | Doku supplies the specialist technical credit record; it does not duplicate the full transaction review. |
| Rating agency | Issuer or instrument credit opinion | Private evidence workflow and lender-owned technical actions | Doku is internal decision support, not a public rating or rating-agency substitute. |
| DCIM and telemetry | Observed equipment and facility operating state | Repayment-case interpretation, covenant effect, and accountable disposition | Telemetry becomes evidence only after context, quality, ownership, and consequence are recorded. |
| Project controls | Schedule, cost, progress, and change control | Credit interpretation across power, compute, contracts, collateral, and debt | Doku converts accepted project-control changes into decision and surveillance records. |
| Market-data provider | Comparable pricing, capacity, supply, and transaction context | Asset-specific evidence, expert judgment, and lender action | Market data is an input to a decision record, never the decision record itself. |
| Internal spreadsheet | Fast ad hoc analysis and familiar review | Versioned evidence, ownership, change history, and reusable portfolio data | Doku preserves the familiar analytical workflow while making provenance and change explicit. |
Supplies high-authority evidence once, with confidentiality rules.
Retains authority for design facts, tests, and technical judgment.
Traces accepted facts into consequence, owner, action, and history.
Approves conditions, reserves, draws, reporting, and intervention.
Drafts accepted credit decisions into controlling documents.
11 / Operating boundary
Trust requires a precise role.
Doku is designed to
- Connect technical facts to credit consequences.
- Expose evidence quality and unresolved uncertainty.
- Support lender decisions before and after close.
- Preserve accountable expert review and overrides.
Doku does not replace
- Stamped engineering design or licensed professional judgment.
- Legal opinions, environmental reports, appraisals, or insurance placement.
- Public credit ratings or an investment committee’s decision authority.
- Source engineering, utility, facility, or operational systems.
12 / Initial application
Built for capital providers with a live, technically complex decision.
The strongest initial fit is a $100 million to $3 billion financing involving a neo-cloud or GPU platform, construction-stage data center, complex power path, liquid cooling, high customer concentration, or debt extending beyond the current hardware generation.
For sponsors and operators, the same record can reduce repeated evidence requests, make closing conditions explicit, and separate authoritative technical facts from lender decisions.
Neo-cloud and GPU finance
Reconcile equipment, facility readiness, performance, contracts, utilization, and recovery.
Construction-stage facilities
Connect power, cost, schedule, acceptance, tenant interfaces, and completion support.
Portfolio and structured credit
Compare operating quality, concentration, reinvestment, technical condition, and triggers.
Sponsor and advisor coordination
Answer one controlled evidence request and expose which accepted facts change closing conditions.
Initial geographyNorth America
Local utility, permitting, legal, and regulatory conclusions remain jurisdiction-specific.
Outside the initial buyer wedgePublic ratings and insurer underwriting
Potential future data users or partners, not represented as current buyers or validated channels.
Explicit exclusionsNo stamped design, legal opinion, appraisal, or investment recommendation
Any regulated or licensed work is separately scoped to an appropriately qualified specialist.
13 / Method and governance
Every conclusion should be inspectable, challengeable, and owned.
Method v0.4 is an initial framework derived from a July 2026 market study that reviewed 96 distinct public-source URLs. It is designed for transaction testing with accountable specialists, not as a public rating methodology.
- 01Define the decision
Name the financing action, timing, threshold, and accountable decision owner.
- 02Build the evidence inventory
Record source, authority, date, version, provenance, confidence, and unresolved gaps.
- 03Reconcile the system
Separate requested, contracted, energized, commissioned, deployable, utilized, and billable capacity.
- 04Trace the consequence
Connect each technical fact to commercial effect, cash flow, collateral, and debt protection.
- 05Translate into action
Propose conditions, milestones, reserves, covenants, reporting, and intervention thresholds.
- 06Preserve reviewability
Record expert review, overrides, residual uncertainty, changes, and the evidence supporting closure.
Evidence authority
A claim is only as strong as the source allowed to support it.
01Executed agreement or official record
Highest authority within its stated scope
02Official letter, certification, or accepted test
Authoritative for the named milestone or condition
03Responsible expert model or report
Expert conclusion with assumptions and version
04Source-system extract or telemetry
Observed state with provenance and quality controls
05Management assertion or working assumption
Visible, challengeable, and never treated as verified fact
Public transaction grounding
Financing documents already connect technical acceptance to capital availability.
These are Doku annotations of public SEC filings, not customer work, legal interpretation, or evidence of Doku performance. The filed source text controls.
CoreWeave credit agreement, SEC Exhibit 10.1, March 30, 2026
- Observed filing logic
- The filed agreement connects written GPU-cluster acceptance and completed testing to the treatment of funding-date capital expenditures.
- Doku workflow implication
- A decision record can preserve the accepted test, the authoritative sign-off, the affected funding condition, and any residual exception.
IREN financing disclosure, SEC Form 8-K, May 29, 2026
- Observed filing logic
- The filing describes delayed-draw and note financing in tranches for GPU infrastructure supporting a dedicated service contract.
- Doku workflow implication
- Infrastructure acceptance, deployment timing, customer obligations, and funding availability must be reviewed as one causal system.
Technical diligence blueprintReview system depth, stress cases, and monitoring fieldsProgressive detail for technical, credit, and model-risk reviewers
- Contracted versus firm capacity, energization sequence, tariff, and curtailment
- Single-line topology, redundancy mode, maintainability, commissioning level, and accepted test
- Air or liquid cooling design state, heat-rejection limit, water dependency, thermal envelope, and failure mode
- Fabric topology, oversubscription, east-west bandwidth, optics, fault domains, and acceptance benchmarks
- GPU mix, cluster homogeneity, yield, reliability, workload performance, and generation exposure
- Orchestration, scheduling, isolation, observability, upgrade controls, incident response, and service acceptance
- Training, fine-tuning, inference, or HPC-specific performance and availability criteria
- Leading indicators, thresholds, frequency, provenance, false-positive review, and accountable response
- Recurring reviews at underwriting, every draw, commissioning, quarterly surveillance, and material change
- Title, location, removal cost, redeployment friction, depreciation, buyer depth, and secondary value
- Energization delay, utilization shortfall, power-price variance, generation change, and customer-loss scenarios
- Trigger, exposure, mitigant, lender control, cure evidence, and residual recovery uncertainty
Reference framework
Standards inform evidence requests. They are not Doku certifications.
- Uptime InstituteFacility topology and operational sustainability references
- ANSI/TIA-942Data-center telecommunications infrastructure reference
- ASHRAE TC 9.9Thermal, energy, and high-density cooling evidence
- IEEE 802.3Ethernet physical and link-layer reference
- NFPA 75Fire-protection reference for information-technology equipment
- NIST CSF 2.0Cybersecurity governance and risk-management reference
01 / Prospective record
Timestamp the evidence, grade, assumptions, reviewer, decision, and predicted failure mode before the outcome is known.
02 / Outcome labels
Record completion, billable capacity, draw, covenant, restructuring, recovery, override, and false-positive outcomes.
03 / Claim threshold
Publish sample size, inclusion rules, calibration error, reviewer agreement, and limitations before claiming predictive performance.
Validation status
The framework has not yet been adopted by a rating agency, approved by a regulator, or validated against a complete default dataset. Transaction evidence and expert review must govern every applied conclusion.
Decision authority
Doku organizes evidence and proposes credit actions. Lender credit, legal counsel, and responsible technical experts retain their respective approval authority and document any override.
Independence and conflicts
The payer, scope, evidence access, permitted reliance, and conflicts are disclosed before work begins. Doku does not issue public credit ratings or provide investment advice.
Review the public role and reliance boundaryData use and model controls
Client evidence is not used for general model training without written authorization. Deployment, access, retention, AI use, export, deletion, and incident handling are defined in the approved scope before restricted evidence is accepted.
Review the public security boundaryDelivery, licensure, and continuity
Delivery is founder-led. Any licensed engineering, legal, insurance, valuation, or other regulated work remains with the client’s advisor or a separately contracted qualified specialist named in the proposal. Reviewer coverage and continuity are engagement-specific.
Reliance and procurement
Public research is non-reliance. Screening, assessment, and surveillance each require proposal-defined scope, permitted users, evidence, liability, reliance, insurance disclosure, and acceptance. Standard vendor materials are provided during qualification.
Request the diligence packetPrepared byJorg Doku, Founder, Doku Infrastructure
PublishedJuly 23, 2026
Current versionMethod v0.4
Source registerDownload 96-source register
Method version history and correction policy
v0.4, July 23, 2026: Added the illustrative product surface, core data model, neutral market comparison, public financing-document annotations, evidence hierarchy, technical diligence appendix, standards references, validation plan, qualified scheduling, and trust-readiness controls.
v0.3, July 23, 2026: Added the public 91-source register, reproducible sensitivity controls, expanded decision record fields, acceptance rules, author context, and qualified financing intake.
v0.2, July 23, 2026: Published the initial lender-focused causal model, illustrative grade architecture, operating boundary, and engagement path.
Material corrections change the version and publication record. Link corrections and non-substantive presentation changes do not change analytical conclusions. Send corrections to jorg@dokuinfra.com.
14 / Engagement path
Start with one live decision and the smallest useful evidence set.
The first conversation requires no project files. Share the asset type, financing decision, timing, and unresolved technical question. Jorg acknowledges by the next business day and replies with fit, clarification, or a scoped next step within two business days.
First-call outcome
A fit decision, the smallest evidence request, the responsible expert boundary, and a proposed next step.
- Typical timing
- 5 to 10 business days
- Indicative fee
- $25K to $60K
A bounded red-flag review of one live financing, its fatal-flaw questions, missing specialist work, and immediate credit actions.
Acceptance rule
Complete when the decision owner receives the evidence gaps, red flags, specialist work plan, and immediate credit actions.
- Typical timing
- 2 to 5 weeks
- Indicative fee
- $75K to $250K
An evidence-linked technical credit assessment, reconciliation, risk register, scenarios, and proposed decision protections.
Acceptance rule
Complete when every material conclusion is traceable to evidence, confidence, consequence, owner, and residual uncertainty.
- Typical timing
- Monthly
- Indicative fee
- $15K to $50K
Ongoing milestone, draw, change, completion, and exception monitoring against the approved credit case.
Acceptance rule
Complete each cycle when changes, exceptions, evidence status, covenant impact, and required actions are reconciled.
Indicative only. The accepted proposal controls final scope, fee, reliance, evidence, timing, expenses, conflicts, and terms. At a $500 million exposure, a $25,000 to $250,000 engagement equals approximately 0.5 to 5 basis points. The illustrative $7.45 million annual power-cost sensitivity equals about 149 basis points of the same exposure; this comparison does not claim the fee prevents that loss.
Typical inclusions
Named decision, evidence inventory, reconciliation, risk record, credit actions, review, and accepted deliverable.
Typical exclusions
Stamped design, legal opinion, appraisal, source-document creation, restricted transfer, travel, and regulated work unless stated.
Working assumptions
One accountable decision owner, timely evidence access, available source experts, and written change control.
Commercial owner
The proposal names payer, budget line, reliance, conversion credit, renewal path, expenses, and final acceptance.
Discuss a live financingUse the contact formView scheduling availabilityNo data room or restricted evidence on first contact.
Current proof status
Public research, source register, method, product schema, and illustrative deliverables are available now. No customer logo, testimonial, completed-transaction outcome, reference, or external endorsement is claimed.
Confidential first-contact briefPrepare a financing review requestEight high-level fields, no project files required
01Acknowledgment
Receipt or prepared-email confirmation by the next business day.
02Fit and boundary
Decision, timing, responsible experts, confidentiality route, and smallest evidence set.
03Scoped next step
Methodology review, rapid screen, proposal, or a clear no-fit answer within two business days.
15 / Selected evidence
Sources behind the market framing.
Commercial estimates are presented as ranges and scenarios, not precise forecasts. The credit architecture is Doku’s synthesis of public market evidence, engineering practice, and financing logic. The references below are used directly on this page. The complete 96-source evidence register is available as a dated, downloadable record.
Download the complete source register01
McKinsey
02The cost of compute: A $7 trillion race to scale data centers
Accessed July 2026Morgan Stanley
03Data centers and the financing gap
Accessed July 2026JLL
042026 Global Data Center Market Outlook
Accessed July 2026S&P Global Market Intelligence
05Banks meeting data-center demand with credit facilities and bonds
Accessed July 2026U.S. EIA
06Electricity prices and the factors that affect them
Accessed July 2026Uptime Institute
Tier Classification System
Accessed July 202616 / Buyer questions
Technical credit diligence should make its boundaries easy to test.
These answers address the most common questions from lenders, sponsors, engineers, counsel, model-risk teams, and procurement.
Is Doku an independent engineer or rating agency?
No. Doku organizes evidence, tests cross-domain consistency, and proposes credit actions. Licensed engineers, legal counsel, lender credit, and rating agencies retain their respective authority.
Does Doku replace DCIM, project controls, or source systems?
No. Those systems remain the source of operating and project facts. Doku preserves how accepted facts change the repayment case, decision, owner, and required follow-up.
How does the first engagement handle confidential evidence?
First contact uses high-level information only. Restricted evidence is accepted only after qualification, signed terms, an approved transfer path, access rules, retention, permitted AI use, and deletion requirements.
What happens after a financing brief is submitted?
Doku acknowledges the request, confirms fit and the smallest evidence set, identifies the responsible expert boundary, and proposes a scoped next step. A scheduling link remains available for buyers who prefer to choose a time directly.
Has the method been validated against defaults?
Not yet. Doku does not claim calibrated default prediction, completed-transaction outcomes, customer references, or external endorsements. Method v0.4 defines how prospective evidence and outcomes should be recorded before performance claims are made.
Live financing review