Inside Infrastructure: Beyond the Building: The New Frontier of Data Centre Finance

Data centres remain one of the most dynamic and structurally inventive segments of the infrastructure financing market. Capital is deploying at pace, demand continues to accelerate, and financing structures are evolving rapidly in response. The drivers extend beyond compute demand, encompassing how projects are financed, how risk is allocated and how capital is deployed over time.

Against that backdrop, attention is increasingly turning to what comes next. Established financing models are maturing, new structures are gaining traction, and recent transactions point to a further phase of innovation. What is emerging is a market that is not only expanding rapidly, but also re-examining the way data centre assets are financed; from portfolio-level solutions and securitisation through to more integrated approaches that respond to constraints in power supply, the scale of capital required and the complexity of delivering these assets.

The macro backdrop: why financing innovation matters

The scale of capital required to support AI infrastructure buildout is unprecedented. Estimates suggest that close to US$7 trillion of global investment will be required by 2030 to meet demand. Data from the Bank for International Settlements illustrates how quickly that buildout is already progressing: as of mid-2025, data centre and IT manufacturing investment represented approximately 1% of US GDP, with broader IT-related investment reaching 5% exceeding levels seen at the height of the dot-com cycle. 

The physical scale of the buildout is striking. Global data centre capacity stood at approximately 100 GW in 2025 and is forecast to exceed 200 GW by 2030, implying a compound annual growth rate of 14%. AI workloads, which already account for roughly half of all data centre capacity, are expected to represent as much as 75% by 2030, driving demand that existing infrastructure and financing frameworks were not designed to accommodate.

Capital expenditure plans reflect this acceleration. Alphabet, Amazon, Meta and Microsoft have collectively announced approximately US$660 billion of capex for 2026 alone, representing year-on-year increases of between 18% and 102% depending on the company. 

Alongside this expansion, sits a more subtle but significant shift in funding dynamics. Leading technology and AI firms, historically characterised by low leverage and strong internal cash generation, are increasingly finding that cash flows are being outpaced by the sheer scale of capital expenditure. Equity financing presents its own challenges: valuation volatility, constrained issuance windows and dilution concerns sit uncomfortably alongside long-dated, asset-heavy projects.

Debt has therefore emerged as the dominant funding mechanism, with private credit playing an increasingly prominent role. Outstanding private credit exposure to AI-related sectors has grown rapidly from negligible levels to more than US$200 billion, with further expansion expected over the remainder of the decade.

Two maturing structures: portfolio financing and securitisation

Against this backdrop, two financing approaches have gained traction among developers with established, operational portfolios: portfolio-level financing and asset-backed securitisation (ABS).

Portfolio financings consolidate multiple standalone project financings into a single holdco-level structure, with financial covenants tested across the portfolio rather than asset by asset. This smooths the impact of underperformance at individual sites and supports a more corporate style credit profile. Security packages have evolved accordingly, with lenders increasingly favouring share pledges, bank accounts and intercompany receivables over asset-level mortgages. This enables enforcement on a going-concern basis and supports higher leverage, more flexible terms and a broader lender syndicate, opening the door to institutional capital that would not typically participate in traditional project finance.

As portfolios scale, structural flexibility becomes increasingly important. Accordion features and expansion mechanisms allow new assets to be added without repeated refinancing events. Simultaneously, provisions relating to unrestricted or excluded subsidiaries preserve optionality for future refinancings, including capital markets transactions. In respect of the structural mechanics of the financing terms, these vary depending on the nature of the underlying customer contracts. Lenders treat the largest customers such as, Amazon, Microsoft and comparable counterparties differently from wholesale, high-performance computing and other segments, with loan-to-cost requirements, EBITDA adjustments and drawdown conditions all calibrated accordingly. 

For portfolios in active development, drawdown conditions typically require confirmation of signed customer contracts of a defined tenor, technical adviser reports covering construction contracts, cost assumptions and delivery schedules, and certification of permits and licences. The degree of forensic diligence on individual assets is generally lower for portfolio borrowers than for single-asset developers — but lenders will still typically reserve the right to review termination provisions in key contracts and impose minimum reporting standards for significant capital expenditure.

Securitisation represents the next stage of that evolution. The first data centre ABS transaction in EMEA closed in May 2024, with issuance developing steadily since. Structures typically rely on stable rental income from long-term leases with high-credit-quality tenants. Recent transactions have featured highly concentrated tenant bases with investment-grade counterparties, supporting sizeable issuances across rated tranches. 

Tranching remains one of the principal attractions of ABS. Senior notes appeal to insurance and infrastructure investors seeking long-duration, lower-risk exposure, while mezzanine tranches attract higher-yield capital. In the US market, relatively conservative loan-to-value profiles have also created space for additional layers of private credit in the form of mezzanine debt or preferred equity — a dynamic that may become more visible in Europe as the market matures.

Insolvency remoteness is where the structural complexity of data centre ABS is most acute. Transferring operational assets into orphan SPVs (the standard securitisation tool) is often impractical as it can crystallise tax liabilities, cut across operational requirements and, in some jurisdictions, trigger regulatory notification obligations. Structures must therefore work harder to achieve bankruptcy remoteness without severing the financing entity from the sponsor's corporate group, a challenge that is amplified in campus settings where shared infrastructure creates operational interdependencies that are not easily ring-fenced.

A related question is whether European issuance will diverge between ABS and CMBS approaches. Transactions focused on cashflow quality may lend themselves to single-jurisdiction portfolios, while mortgage-backed structures could prove more adaptable for pan-European portfolios where asset-level security is more easily portable across borders.

The DevCo/YieldCo model:

Before turning to the newest structural developments, it is worth pausing on a structural model that underpins much of the above: the “DevCo/YieldCo” or “DevCo/StableCo” bifurcation. Borrowed from the real estate sector, this approach separates development-phase assets from stabilised, income-generating assets into distinct entities, each financed on terms calibrated to its own risk profile.

DevCos are inherently riskier, they carry construction, permitting and market uncertainty and attract lenders and equity investors comfortable with higher risk in exchange for potentially higher returns. YieldCos, by contrast, appeal to more conservative institutional capital: pension funds and insurance companies that prioritise stable, long-term cash flows and are typically restricted from taking construction risk. The structural separation also enables a capital recycling mechanism: as assets stabilise and are transferred from the DevCo to the YieldCo, proceeds prepay development loans which are redrawn for new projects, allowing the platform to reinvest continuously without repeated recourse to new equity.

The model is not without tension. Customers and power providers may require recourse to a creditworthy entity rather than a newly established DevCo. And the competing interests of DevCo and YieldCo creditors, particularly around the terms on which assets are transferred and the incentives each lender group has regarding customer contracting, require careful management. These structural frictions reinforce the point that no single financing model fits all stages of a data centre platform's development.

Integrated structures: when data centres and power converge

Structural innovation continues to accelerate; a notable example, reported under the codename “Project Walleye” involves the financing of a large-scale data centre campus in Ohio. While the project itself is typical of hyperscale deployments, the financing structure is not. 

Rather than separate financings for the data centre and its power infrastructure, both have been wrapped into a single transaction. The driver is grid access, or rather the lack of it. Behind-the-meter power solutions are increasingly a practical necessity for developers unwilling to wait years for a connection. In this case, the project is expected to operate initially as an islanded microgrid, powered by on-site generation, with the potential to connect to the grid at a later stage. 

Similar models are emerging elsewhere, driven by the same constraint. Power has become one of the most consequential bottlenecks in data centre development. The UK's National Grid has indicated that data centre power demand will increase six-fold over the next decade, while The International Energy Agency forecasts that global electricity demand from data centres could double 2022 levels by 2026. Transmission networks are under material capacity pressure, and the mismatch between where power is available and where data centre demand is concentrated is reshaping site selection, development timelines and, increasingly, financing structures.

The core structural question is how to underwrite two fundamentally different assets; a data centre and a power station within a single credit. A solution and ongoing development in the market, sees the tenant's credit doing the work that a guarantee would otherwise perform, resulting in a range of operating-entity credit support mechanisms that perform a similar economic function, with the precise legal form varying transaction by transaction. What sets this structure apart is the dual construction risk: two separate projects being built simultaneously for a common purpose, where difficulties with one have the potential to affect the financing of both.

The pricing reflects the additional risk: a margin of 2.5 percentage points above the secured overnight financing rate, slightly above comparable transactions. That premium has not deterred investors, who appear to have taken comfort from EdgeConneX's delivery track record and the strength of the Meta offtake.

These conditions are, if anything, more acute in Europe. Grid connection timelines in several markets extend into the next decade, and policy developments are reinforcing rather than relieving the pressure: the UK has reformed its connection queue process, Ireland now requires new data centres to provide on-site generation for their full demand, and investment is actively shifting toward secondary cities where grid capacity is more readily available.

The market is shifting from 20–50 MW facilities to campuses of 500 MW to 1 GW or more, with hyperscalers now contracting for entire industrial estates rather than individual buildings. Capital requirements per campus have risen to £750 million to over £1 billion. At that scale, power supply is not merely an operational consideration, it is a financing constraint in its own right, and the ability to bring power infrastructure into the financing structure alongside the data centre building is likely to become an increasingly important tool.

A note on AI-focused data centres

A final consideration is the growing distinction between traditional hyperscale developments and AI-focused data centre models. The established model, where developers provide the shell and hyperscale tenants deploy their own equipment, remains well understood from a financing perspective. Pre-let agreements for lease, typically executed 24–36 months before delivery, underpin the bankability of these transactions, with lease terms of 10–20 years providing the revenue certainty that infrastructure lenders require. Newer AI-oriented models are a different proposition, requiring developers to provide both physical infrastructure and compute capacity through service-based offerings.

The risk profile reflects that difference. Equipment costs more than the building it sits in. Customers are often new entrants with shorter contracts and less established credit. Hardware obsolescence is a genuine financing consideration, not a theoretical one. The neocloud segment (specialist AI cloud providers such as CoreWeave, Lambda Labs and Nebius) is forecast to grow at 82% and reach $180 billion in revenue by 2030, but their capital-efficient, asset-light approach and focus on depreciating compute hardware rather than long-lived real estate infrastructure presents a fundamentally different risk profile for lenders. The market is finding its way through equipment finance, borrowing base facilities and hybrid structures that blend real estate and infrastructure conventions, but the playbook is still being written.

Regulatory and geopolitical considerations add a further layer of complexity. Data centres have been designated critical national infrastructure in the UK, placing them on a par with water and energy with similar classifications emerging elsewhere. This has implications for foreign investment screening, security of tenure and resilience obligations that operators and their lenders must address. Sovereign AI strategies (including the UK government's target of at least 6 GW of AI-capable data centre capacity by 2030) are creating both demand tailwinds and regulatory obligations that will shape financing structures for years to come.

Data centre financing has traveled a long way in a short time. What began as a subsector of real estate lending now encompasses portfolio consolidations, public securitisations and integrated infrastructure transactions that few would have anticipated five years ago. The asset class has matured — but the pace of change has not slowed. The emergence of AI-focused development is accelerating it, introducing capital requirements and risk profiles that existing frameworks were not built to handle. The structures being developed and tested today are likely to define how this market operates for the next decade.