27 Aug 2026

A Data Centre Is Not One Asset: How AI Infrastructure Is Becoming a Stack of Securitisable Cash Flows

AI-driven data-centre investment is pushing securitisation beyond a simple real-estate story. US issuance now exceeds $25 billion annually, Europe is developing quickly, and recent US regulatory guidance shows why a structure described as “ABS” in the market may not be an asset-backed security under the statutory definition. The deeper lesson is that one facility can contain several different investable rights, contracts, receivables and financing layers.

Create a structured infrastructure cash-flow record

This starts a temporary private draft. It is not public, listed for sale or shared automatically.

Credit manager Receivable Learn

A data centre can look like a single asset: a site, a building, power and cooling systems, fibre connections, security and rows of equipment. Its financing can nevertheless depend on several different rights. A mortgage lender, an equipment financier and an investor in platform securities may all depend on the same facility while having different collateral, payment sources and enforcement routes.

On 26 August 2026, Dentons described annual US data-centre securitisation issuance exceeding $25 billion, with approximately 20 active issuers. It also reported four public European transactions since the first public transaction in 2024, totalling approximately €2.2 billion. That growth makes the identity of the financed asset increasingly important. Dentons: Data centre securitisation explained—US, Europe and Asia

Some transactions resemble commercial real-estate finance; others finance a platform's contractual operating revenues. Market participants may call a structure “ABS”, yet its statutory classification can depend on a more precise question: what assets produce the cash available to investors?

Two financing structures, two different payment sources

Traditional property finance connects ownership of a property, a mortgage loan and the borrower's debt service. Rental income supports repayment, and the lender may enforce against the property after default. A data centre can be financed in that way, but its operating dependencies extend well beyond those of an ordinary building.

Land access, grid connection, electricity supply, backup generation, transformers, switchgear, cooling, water where relevant, fibre, security, technical operations, permits, capacity allocation, equipment, software, monitoring and maintenance all help sustain the customer service. Customer payments can include occupancy leases or licences, power, cooling, connectivity, maintenance and other services; reimbursement for capital upgrades or customer-specific fit-outs; equipment leasing; managed infrastructure; and, subject to eligibility criteria, subscription or usage-based services such as GPU-as-a-service. Dentons: Data-centre ABS structures

Identify the asset that pays the investor

Mortgage-backed structure

A mortgage loan secured over a stabilised property produces principal and interest payments. Customer revenues support the borrower's ability to service that loan; they are not necessarily transferred to the securities investor.

Platform operating structure

Facilities, customer contracts and operating infrastructure generate contractual revenues. Operating expenses are paid before net cash flow services securities through an issuer or master-trust structure.

Dentons describes data-centre CMBS as typically involving a single-asset, single-borrower mortgage loan secured over stabilised, income-producing property. The securitised financial asset is the mortgage loan, rather than the server hall, electricity system, customer's racks or necessarily the customer contract. Customer payments reach the operator; the operator services the mortgage; the mortgage pays the issuer; the issuer pays investors. Dentons: CMBS structures

That structure keeps familiar property questions central: appraisal and loan-to-value, mortgage enforceability, lease term, tenant concentration and credit quality, debt-service coverage, property condition, replacement cost, reletting risk and enforcement value. Platform finance adds a more direct dependency on the operating system that generates the revenue.

Why a market ABS label may differ from the legal classification

The July 2026 Latham & Watkins request to the SEC described special-purpose issuers with ownership, leasehold or other rights to asset entities holding data centres, physical components and operating contracts. Buildings, data halls, electrical and backup-power systems, networks, service contracts, leases, colocation agreements, insurance and related operating assets could form part of the described structure. Master trusts could support later issuance, refinancing, additions, dispositions and, in some circumstances, asset substitutions. Securities were serviced from net operating cash flow. Latham & Watkins: Request for Interpretive Guidance

Section 3(a)(79) of the Exchange Act refers to securities collateralised by self-liquidating financial assets whose payments depend primarily on those assets' cash flows. Latham argued that the structures it described did not meet that definition. A building, electrical system or cooling plant does not liquidate itself as investors are repaid: the facility can continue operating, appreciate or support later financing. Revenue also has to cover electricity, taxes, insurance, repairs, maintenance, security, staffing, equipment and other operating costs. The operator's ability to generate revenues and manage those costs therefore matters.

On 29 July 2026, the SEC Division of Corporation Finance staff agreed, on the facts and representations presented, that the described securities were not Exchange Act ABS. Reuters discussed the clarification on 10 August 2026 in the context of AI infrastructure's capital requirements. SEC staff response, Reuters

That is a fact-specific staff position, not a Commission rule or an amendment of the law. It does not establish that every data-centre security falls outside the ABS definition. Different structures or facts can require a different analysis.

A financing record should therefore keep the market label separate from the jurisdiction-specific legal classification. Data-centre ABS, CMBS, infrastructure debt, project finance, private credit, whole-business or operating-asset securitisation, equipment finance and lease finance are useful market descriptions. None replaces identification of the issuer, investor instrument, collateral and recourse, primary payment source, and regulatory analysis with its source and date. Cross-border records should preserve differences between jurisdictions rather than force them into one category.

Separate the economic objects inside the facility

One operating facility may connect land ownership or a long lease; the building; grid rights; electrical and cooling systems; fibre; customer occupancy agreements, licences and service contracts; power reimbursements, connectivity and maintenance charges; fit-out reimbursements; equipment leases and GPU or managed-compute contracts; current receivables and future contracted revenue. Its financing can add a mortgage, equipment finance, private-credit facilities, security over assets or accounts, reserves, equity in asset entities, issuer notes and residual equity after debt service.

These objects are connected without being interchangeable. Equipment diligence needs ownership and serial-number identity, depreciation, maintenance, location, insurance, remarketing value and security perfection. Mortgage diligence centres on real estate, lease income, appraisal, title and priority, permitted use, construction quality and foreclosure rights. Platform securities require attention to recurring revenues, expenses, concentration, customer churn and renewal, service performance, pool changes and continued operation.

The same care applies to amounts. A five-year customer agreement can support expectations of future revenue without creating five years of receivables today. Availability of capacity, occupancy, usage and power consumption, service performance, conditions, minimum commitments, escalation, renewal, credits, termination, force majeure and disputes can affect whether payments arise.

Amount or state	What a reviewer needs to distinguish
Contracted future revenue	Future payments subject to the governing agreement and its conditions.
Accrued revenue	Revenue recognised for a relevant period; distinct from billing and collection.
Invoice	The amount billed, including its components and adjustments.
Current receivable	The payment right and balance currently due or outstanding.
Cash collected	Payments actually evidenced as received.

A usage-based compute contract can be valuable even when future monthly charges depend on consumption. A current invoice and a currently due receivable are separate objects again; all these states can coexist.

Power and operating performance belong in the credit analysis

Power affects both infrastructure availability and cash flow. A facility may have contracted grid capacity, staged energisation, backup generation, renewable arrangements or power-purchase agreements, pass-through customer charges, hedges, minimum consumption commitments and peak-demand charges. An “annual rent” field cannot explain the resulting economics.

A financier needs to ask whether power is passed through at cost or includes margin; whether customer pricing can rise with electricity costs; whether interruption produces credits; whether constrained grid capacity delays contracted revenue; whether the model assumes capacity not yet energised; and whether grid rights transfer with the property in enforcement. Occupancy, power, cooling, connectivity and other services should remain separately identifiable.

The operator also has to maintain cooling and power resilience, monitor capacity, respond to incidents, manage contracts and renewals, attract customers, run billing and delinquency processes, maintain insurance, manage capital expenditure, and comply with permits and legal requirements. Isolating assets from wider corporate activities does not remove these operating dependencies. Ownership of a cash flow and production of that cash flow are different questions.

AI infrastructure can shorten parts of the risk horizon. Dentons' August 2026 analysis identified specialist AI-cloud structures with customer contracts sometimes around two to five years, sub-investment-grade or venture-backed counterparties, usage-based revenues, rapid technological change, specialised liquid cooling, costly customer-specific configurations and short GPU economic lives. Dentons: From construction to capital markets

A building may last decades while GPUs become commercially obsolete much sooner. Cooling may suit a particular hardware generation, and converting a customised hall may be expensive. A customer's financial history can also be much shorter than the facility's. A five-year security connected to a twenty-year physical asset can therefore depend on components whose economic lives are shorter than the debt.

Dynamic pools need a dated history, not just a closing list

Master-trust collateral can change through additions, removals, substitutions, refinancing, expansion, partial release and transfers into new asset entities; new securities may be issued as well. Customers arrive, leave or expand. A five-year contract can become a renewed seven-year relationship, while a site receives a new power allocation or undergoes expansion.

For each material asset or contract, an as-of-time record should retain entry into the financed pool, the eligibility basis and supporting valuation, linked contracts, subsequent changes, continuing eligibility, release, competing financing rights and the reporting period in which the event appears. A closing-date list alone cannot establish the current position.

Contractual provenance is equally important. Review the legal customer and guarantor, capacity, facility and hall, commencement and expiry, extension and termination rights, pricing and indexation, minimum commitments, power treatment, fit-out obligations, service levels, credits and outages, insurance, assignment, change of control, confidentiality, security requirements, cross-default and governing law. One customer may have several agreements; an amendment can affect one site, an order form can add capacity, a service schedule can change obligations and a side letter can alter price. The question is which current documents produce the revenue in the financing model.

Component lifecycles should remain separate. Land and buildings may endure for decades; grid rights can be long-lived but capacity-constrained; electrical and cooling systems need periodic investment; customer contracts run for years; hardware may have a much shorter economic life; receivables mature in days or months; and securities have defined tenors. Replacing GPUs should not replace a building's identity. Renewal is not a new facility, refinancing is not necessarily a sale, and paying an invoice extinguishes that receivable without ending the customer relationship.

A proposed evidence record for the financing stack

A cover record such as “Data centre; $1.2 billion; Active” cannot carry the whole analysis. A proposed data-centre evidence model would preserve the following layers and connect them with sources. This is a description of information a reviewer needs, rather than a claim that DaDepo already implements a specialised infrastructure or securitisation engine.

Layer	Information to preserve
Physical facility	Identifier, address and jurisdiction, owner and operator, site area, gross and usable capacity, halls, construction and expansion phases, condition, major equipment and replacement programme, insurance, permits and documents.
Site and infrastructure rights	Title or lease, easements, connected and energised capacity, planned additions, power supply and backup, water where relevant, fibre and access, environmental constraints and transfer restrictions.
Customer relationships	Legal customer, guarantor, contract type, facility and capacity, commencement, expiry and renewal, termination, pricing and minimum commitment, service levels, assignment, change of control and amendments.
Revenues and receivables	Revenue component, invoice, debtor, period, amount and currency, due date, outstanding balance, credits, disputes, payment and collection history, and governing contract.
Operating expenses	Electricity, cooling, maintenance, taxes, insurance, staffing, security, networks, equipment leases and other material costs, with allocation method and as-of date.
Financing	Borrower and lender, principal, interest and maturity, loan-to-value, covenants, security and guarantees, reserves and cash management, default and amendments.
Securitisation	Issuer, SPV or master trust, asset entities, securities, series or tranche, collateral and recourse, waterfall, eligibility, additions and substitutions, releases, servicing, reporting period and supported legal classification.
Provenance	Source document or system, extracted, user-confirmed and externally confirmed fields, reviewer and review status, version and effective date, superseded documents and last update.

Billing illustrates why these layers matter. A $4 million invoice might contain $2.2 million of eligible contracted occupancy revenue, $1.1 million of power reimbursement, $400,000 of equipment rent and $350,000 of installation charges, reduced by a $50,000 service credit. Those components produce the total, but financing documents may treat their eligibility differently. The invoice can also combine cooling, connectivity and managed services or one-off fit-out reimbursement. A reviewer should be able to reconstruct how each component entered the eligible balance.

Ownership, enforcement and valuation require their own objects

A sponsor can own the group, an issuer can own an asset entity, and that entity can own the building. Equipment may be leased or belong to a customer. Different creditors may have security over property, equipment, accounts, contracts or entity shares. “Owner: SPV” obscures the legal owners of land, buildings, electrical infrastructure and computing equipment, as well as lessees, borrowers, security providers, secured creditors, account banks, issuers, noteholder rights and the enforcement agent or trustee.

Distress makes these distinctions practical. Who owns the site, and can the secured party enforce against the asset entity? Can customer contracts be assigned, and can customers terminate on insolvency or change of control? Do permits, licences and grid rights survive or transfer? Who owns equipment, can another operator take over, are cash accounts controlled, and which liabilities remain? Would a hyperscale customer stay, and what investment is needed to maintain competitiveness? Normal-operation value can differ substantially from the value of the enforceable package.

Valuations also need named objects. Land, property, replacement cost, enterprise value, entity equity, mortgage collateral, contracted revenue, receivables, equipment, GPU residuals, loan balance, securitisation collateral, note price, recovery and sponsor equity are different measures. A $1 billion property valuation is not a valuation of the customer contracts; $600 million of notes issued does not establish facility value; a $200 million annual revenue run rate is not $200 million of current receivables; and $50 million of equipment value is not a promise of $50 million realised in enforcement. Retain the object, method, date, currency, valuer, assumptions, scenario, source and relationship to the financing.

Lifecycle reporting should preserve construction completion, grid connection and additional energisation; signing, commencement, capacity changes and pricing amendments; SLA breaches, credits, outages and insurance claims; customer downgrade, default, renewal or termination; equipment installation or replacement; pool additions or removals; refinancing, covenant breach, reserve draw, payment, enforcement, operator replacement and sale. These events can change credit rather than merely add narrative notes.

AI preparation and DaDepo's bounded contribution

AI-assisted document preparation can help identify facility names and addresses, asset entities, customers and guarantors, contract dates and renewals, capacity and power commitments, pricing and indexation, service levels, termination and assignment wording, invoices and payment dates, amendments, equipment schedules, insurance, loan and security terms, reserves and reporting obligations. Each proposed finding needs its source and human review.

A more specialised future review tool could flag conflicting expiry dates, mismatched customer and guarantor names, capacity above energised capacity, unsupported invoices, missing amendments, pricing inconsistent with billing, duplicate pool inclusion, contracts still marked active after termination, paid receivables still outstanding, or investor reporting without an eligibility record. Those are proposed checking capabilities, not a statement that the existing DaDepo workflow performs all these cross-record tests automatically.

DaDepo's existing contribution is upstream preparation of reviewable asset information and documentary evidence. Users can organise the relevant documents and review suggested information; source-described party and contractual roles can be recorded separately for confirmation. A complete graph linking facility, entity, contract, revenue component, invoice, cash, financing, collateral and investor instrument remains a proposed information model. Every connection would need evidence. Such a model could support diligence, refinancing, CMBS or platform-securitisation review, monitoring, servicing, collateral and investor reporting, sale, enforcement and later secondary-market workflows; it would not itself make a facility financeable.

Neither AI preparation nor an Asset Passport establishes contract enforceability, eligibility, security perfection or priority, transferable grid rights, renewal prospects, counterparty creditworthiness, technological competitiveness, equipment value, collateral suitability, regulatory classification, recovery value or investment suitability. Review also does not authenticate every document, verify land title or equipment ownership, confirm energisation or customer performance, inspect engineering condition, or predict occupancy, utilisation and repayment.

DaDepo does not thereby lend, originate mortgages, arrange private credit, issue CMBS or ABS, operate or service a securitisation, operate a data centre, guarantee uptime or recovery, or recommend an investment. Legal, regulatory, engineering, tax, accounting, investment, credit-rating, underwriting and valuation judgements remain with the appropriate reviewers. Technology and information tools do not constitute banking, lending, custody, settlement or other professional services unless a specific service is expressly identified and lawfully provided. Structures and rights depend on the jurisdiction, transaction and facts.

Questions to resolve before adding another financing layer

The following questions retain the practical checks needed before presenting the facility, cash flow or structure to a lender or investor.

Facility and operating basis

Facility: Which exact physical facility or portfolio is in scope?
Ownership: Which entity legally owns the land, building and major infrastructure?
Operator: Which entity operates the facility and under what agreement?
Power: What capacity is contracted, connected, energised and actually available?
Customers: Which legal entities are responsible for the contracted revenue?
Contracts: Which current agreements, schedules and amendments create the customer obligations?
Capacity: What capacity has each customer contracted and what is currently live?

Cash flow, equipment and financing

Revenue: Which amounts are occupancy, power, cooling, connectivity, equipment, managed service or other revenue?
Receivables: Which amounts are currently invoiced and outstanding rather than merely expected future revenue?
Adjustments: Are service credits, disputes, rebates and amendments reflected?
Operating costs: Which material costs must be paid before cash is available for debt service?
Equipment: Who owns material computing, electrical and cooling equipment, and what is its useful economic life?
Financing: Which mortgage, equipment, construction, project-finance or private-credit facilities are outstanding?
Security: Which assets, equity interests, accounts, contracts or receivables secure each financing?

Current rights and reporting

Priority: Do different lenders or investors have competing rights?
Pool eligibility: If a securitisation pool is dynamic, why is each asset or receivable eligible as of the relevant reporting date?
Transferability: Can the relevant contracts, rights, permits and collateral transfer or survive enforcement?
Classification: What is the market label of the transaction, and what legal or regulatory classification applies in the relevant jurisdiction?
Lifecycle: Are additions, releases, substitutions, refinancings, outages, renewals and other material events recorded?
Provenance: Can every material field in the financing model be traced to the correct document, system, party and version?

The same facility can support real-estate, construction, equipment, mortgage, occupancy, service-receivable, operating-platform and investor-security exposures. The July 2026 SEC staff response makes the distinction unusually visible: securitisation techniques, a market ABS label and a statutory ABS classification are separate matters. If the answer to “what exactly supports the investor's payment?” remains unclear, another financing or digital representation layer will not resolve it.