A data centre that loses money is rarely undone by a faulty chiller or a delayed fibre connection. It is undone by the way it was financed. The sequencing of equity, debt, guarantees, and contractual controls determines which parties bear stress, when they bear it, and how much of it reaches the operational layer at all. Get the capital stack right, and a project can absorb construction delays, demand volatility, and technology shifts without breaking. Get it wrong, and even a well run facility becomes a write down.
This matters because the sums involved are now very large. Data centre debt issuance nearly doubled to $182 billion in 2025, according to S&P Global. Dealmaking in the sector surpassed $61 billion. McKinsey's proprietary data centre demand model projects $6.7 trillion in cumulative global capital expenditure will be needed by 2030 to keep pace with demand for compute power, with $5.2 trillion of that attributed to AI related capacity alone. Brookfield has launched a $100 billion AI infrastructure programme backed by NVIDIA and Kuwait's sovereign wealth fund, targeting the full value chain from power to compute.
At this scale, capital structure is not a back office concern. It is the primary instrument of risk management.
Infrastructure projects are not financed like corporations. They are ring fenced in special purpose vehicles (SPVs) that impose a strict hierarchy, often called a waterfall, on every dollar of cash flow. Operating costs are covered first. Senior lenders are paid next. Junior creditors follow. Equity holders take what remains.
This is not merely convention. It is a contractual architecture that forces risk downward through the capital stack before it can disrupt the underlying asset. Each layer absorbs a different category of stress.
sits in the first loss position. It bears the cost of overruns, shortfalls, and early stage execution failures. In European data centre project finance, equity contributions for pre let, investment grade structures typically fall between 20 and 30 per cent of total spend, according to Norton Rose Fulbright. For higher risk profiles, including AI native facilities during construction phases with novel cooling requirements or less established operator track records, equity requirements can reach 40 to 50 per cent. In either case, the equity cushion means a project must absorb significant losses before senior debt service is threatened.
Is secured against physical assets and contracted revenue. Unlike corporate lending, where collateral is often intangible, infrastructure debt is backed by real property and long term leases. Macquarie Asset Management reports that recovery rates for senior secured infrastructure debt average 70 per cent, versus 52 per cent for direct lending and 38 per cent for corporate credit.
Occupies the space between senior tranches and equity. In data centre structures, it often defers interest through payment in kind mechanisms and may carry warrants or other equity kickers, as BCLP has documented. This preserves cash during construction while compensating lenders for the higher risk.
Provide a further layer of absorption. Parent company guarantees, performance bonds, and completion undertakings are standard in data centre project finance. Fixed price, date certain construction contracts backed by these instruments give lenders confidence that the asset will be delivered on specification. Lenders will not fund projects with uncertain grid connection dates or uncapped construction costs.
The combined effect: operational risk, the risk that the data centre will not perform, is substantially absorbed by the capital stack before it can manifest as a real world failure.
No single variable matters more to the capital stack than the offtake agreement. An uncontracted data centre is a speculative asset. A contracted one resembles a utility: revenue is visible, cash flows are forecastable, and the structure can be optimized accordingly.
Lenders understand this clearly. As practitioners noted in Norton Rose Fulbright's 2025 financing roundtable, a project becomes financeable when land is secured and lease discussions are advancing. The creditworthiness of the tenant and the clarity of the lease terms directly determine the cost and availability of debt. The lender is not underwriting a building. It is underwriting a contracted revenue stream.
Cloud data centres pre let to investment grade hyperscalers under ten to fifteen year leases are, as Norton Rose Fulbright's European team describes, highly bankable. In practice, financing for pre let hyperscale facilities typically involves limited recourse to the sponsor during the construction phase, transitioning to full non recourse once completion tests are satisfied and contracted cash flows are established. The offtake contract, backed by one of the world's most creditworthy counterparties, converts construction risk from a commercial uncertainty into a bounded engineering problem.
This conversion is the mechanism by which offtake agreements absorb risk. They do not eliminate it. They redistribute it. Construction risk is mitigated by completion guarantees. Power risk is managed through grid agreements and on site generation. Technology obsolescence is allocated through lease terms and hardware refresh provisions. At each point, a contractual instrument assigns the risk to the party best positioned to manage it.
Without contracted demand, this architecture collapses. Capital deployed against speculative capacity, against the hope of tenants rather than signed leases, is exposed across the full spectrum. It should be noted that some operators and markets have supported uncontracted development where demand fundamentals are sufficiently strong and the developer's balance sheet can absorb the risk.
Listed data centre operators with established hyperscale relationships have financed speculative capacity successfully in supply constrained markets such as Northern Virginia and Singapore. But the offtake first model is most relevant in markets with higher execution complexity, less established demand histories, and capital structures that require institutional or sovereign participation. In those settings, the principle that offtake precedes capital is not a preference. It is a structural requirement.
The claim that well structured capital stacks absorb risk is not conceptual. It is empirical, though the evidence base requires careful interpretation.
The most robust data covers infrastructure debt broadly, including transport, utilities, and social infrastructure. Data centre specific default and recovery histories are materially shorter, reflecting the asset class's relative youth. The broader infrastructure dataset is cited here as the closest available proxy, with the caveat that it should not be mapped mechanically to data centre project finance.
Moody's infrastructure default and recovery data, cited by Brookfield, shows the cumulative five year default rate for non financial corporates at 9.6 per cent. For infrastructure corporate and project finance, the figure is 2.4 per cent: roughly a quarter of the corporate rate. The differential holds when controlling for credit rating.
Scientific Infra and Private Assets (SIPA) corroborates this. Between 2019 and 2023, the average annual probability of default for infrastructure debt stabilized at 1 to 1.3 per cent, with recovery rates consistently above 75 per cent. In the power sector, the closest analogue to data centre infrastructure in terms of asset profile and revenue structure, AEW reports recovery rates of 92 per cent.
Three structural features explain these numbers:
SIPA's data also reveals a maturation effect. Young infrastructure firms in their first five years carry default probabilities around 1.8 per cent. Mature firms drop to 0.6 per cent. Once construction is complete and contracted cash flows are established, the risk profile fundamentally shifts.
The implication for investors is direct: in infrastructure, the capital stack is not merely a financing choice. It is a risk management tool with a measurable, if still maturing, track record.
One feature of infrastructure capital stacks that institutional investors often underweight is that they are not static. The stack evolves as the project moves through distinct phases, and each transition resets the risk profile and the instruments used to manage it.
Risk is concentrated in site selection, permitting, and power procurement. Equity absorbs nearly all of it. Sponsor guarantees may be required. The capital stack is thin and the cost of capital is high.
Execution risk dominates. The stack absorbs it through fixed price EPC contracts, performance bonds, and completion guarantees. Senior debt becomes available but subject to stringent conditions and typically on a limited recourse basis. Grid connection certainty and cost caps are non negotiable for lenders.
The data centre is operational and tenants are in place. Completion guarantees fall away. Sponsor recourse extinguishes. The financing transitions to a non recourse infrastructure regime. Norton Rose Fulbright notes that this shift is typically designed into the documentation from the outset, with sponsor guarantees expiring on customer contract execution.
Stabilized assets access lower cost permanent capital through asset backed securities (ABS) or commercial mortgage backed securities (CMBS). Data centre ABS and CMBS issuance exceeded $20 billion in 2025, according to Morgan Stanley data cited by Brandywine Global, surpassing the 2024 total by 50 per cent. Developers refinance construction debt at materially lower cost and recycle equity into the next project.
At each phase, the stack reshapes itself. The instruments change, the participants change, and the cost of capital falls as uncertainty is resolved. This progression is the fundamental purpose of structured finance.
For sovereign wealth funds and quasi sovereign investors, the capital stack serves an additional function: it enforces the discipline their mandates require. But discipline is only part of the picture. GCC sovereign capital, in particular, operates at the intersection of financial return and national strategic interest. AI infrastructure investment for these entities is not purely a portfolio allocation. It is a component of national digital sovereignty, economic diversification, and strategic positioning in an era of increasing technological competition.
According to industry estimates compiled by Global SWF and reported by AGBI, GCC sovereign funds collectively deployed a record $119 billion in 2025. Mubadala was estimated to have committed approximately $12.9 billion across all digital assets, including AI, data centres, and digital infrastructure, according to Global SWF data cited by Khaleej Times. Kuwait's Investment Authority directed an estimated $6 billion to related themes. Deloitte estimates that GCC sovereign funds now manage nearly $6 trillion in aggregate, with infrastructure and AI among their fastest growing allocation categories.
These investors are not passive capital providers. They demand co investment rights, governance representation, and ESG grade compliance. They will not accept speculative exposure or opaque deployment structures. Equally, they evaluate investments against national strategic priorities, not solely against financial benchmarks.
A properly sequenced capital stack meets these requirements. Clear subordination defines who absorbs losses and in what order. Contracted offtake provides demand visibility before capital is committed. Staged deployment triggers link drawdowns to verified milestones. Step in rights preserve the option to intervene.
When Brookfield launched its AI infrastructure programme with NVIDIA and the Kuwait Investment Authority, it explicitly prioritized investments backed by highly creditworthy counterparties and contracted cash flows. The logic is the same: capital follows demand, not the reverse.
For platforms operating across the GCC and multiple jurisdictions, presenting a capital stack that has already absorbed demand risk through offtake, execution risk through qualified operators, and regulatory risk through jurisdictional compliance is not a competitive advantage. It is the entry condition.
Three categories of risk sit partly or wholly outside the capital stack, and intellectual rigour requires naming them.
GPU architectures evolve on eighteen to twenty four month cycles. Over a fifteen to twenty year facility lifecycle, several replacement cycles will occur, each requiring significant capital expenditure. Capital structure can allocate refresh costs between parties, but it cannot eliminate the risk of misjudging the pace of change.
The regulatory environment for AI infrastructure is evolving rapidly. US export controls on advanced computing hardware have undergone multiple revisions since October 2022, most recently through the Biden administration's AI Diffusion Rule of January 2025, which introduced a tiered country framework directly affecting GCC procurement of advanced GPUs. That rule was rescinded by the Trump administration in May 2025, with a replacement regulation pending. Data sovereignty requirements, shifting energy policy, and evolving AI governance frameworks across the GCC, EU, and ASEAN add further layers of uncertainty. No contractual framework fully anticipates these dynamics. Jurisdictional diversification mitigates but does not remove this risk.
The Bank for International Settlements has flagged that some AI financing structures may mask leverage through off balance sheet arrangements, and that the long term viability of the investment cycle depends on meeting the expectations currently embedded in valuations. A meaningful gap between expected and realized AI adoption would pressure the entire asset class.
These risks do not argue against structured capital. They argue for it. The projects best positioned to withstand a correction are those with contracted demand, conservative leverage, and clear allocation of who bears what. The capital stack does not eliminate uncertainty. It ensures that when uncertainty materializes, the consequences follow an orderly, predictable sequence.
The current AI infrastructure cycle presents a tension. Capital is plentiful: over $200 billion was raised across more than 80 infrastructure funds in 2025, according to MetLife Investment Management. Demand is strong: the major hyperscalers collectively spent an estimated $320 billion on AI infrastructure in the same year. Yet the binding constraint is not capital or demand in isolation. It is the structured alignment between them.
Uncontracted capacity cannot access favourable financing terms. It cannot benefit from the risk absorbing mechanics of project finance. It cannot attract sovereign and institutional capital on terms that justify the deployment. Every layer of its capital stack sits exposed.
Contracted capacity unlocks the full architecture. Equity cushions absorb execution risk. Senior debt is secured against physical assets and visible revenue. Guarantees protect against delivery failures. Control rights provide lenders with tools to intervene before distress cascades. And the entire structure can be refinanced into lower cost permanent capital once the asset stabilizes, recycling equity into the next phase.
As the asset class matures, the distinction between projects that can demonstrate this discipline and those that cannot will sharpen. In a market that will require trillions of dollars in investment over the coming decade, the capital stack is not a technicality. It is the mechanism that determines which projects attract capital, which withstand stress, and which endure.
This article is published by Claymont Equivator Infrastructure for informational purposes only. It does not constitute a financial promotion, an invitation to invest, or investment advice under applicable regulations in any jurisdiction. Claymont Equivator is a strategic demand and capital enabler for AI data centre infrastructure across the GCC and selected European jurisdictions. For institutional enquiries, visit claymont-equivator.ai.
1. S&P Global / CNBC, "Data center deals hit record $61 billion in 2025," December 2025 Link
2. McKinsey Quarterly, "The cost of compute: A $7 trillion race to scale data centers," April 2025 Link
3. JLL / Data Centre Magazine, "Financing the Future: Trends in 2025," October 2025 Link
4. Morgan Stanley / Brandywine Global, "Brave New World of AI Capex," 2025 Link
5. ZISHI, "Data Centre Financing: Inside the $3T Race," 2025 Link
6. Brookfield, "$100 Billion AI Infrastructure Program," November 2025 Link
7. Norton Rose Fulbright, "Data Center Financing Structures," June 2025 Link
8. Norton Rose Fulbright, "Data Centre Financing: A European Perspective," 2025 Link
9. Norton Rose Fulbright, "Lender Considerations on Data Centre Financings in Europe" Link
10. BCLP, "Financing Data Centre Developments," 2025 Link
11. Macquarie Asset Management, "Infrastructure Debt: First Among Equals," 2025 Link
12. Brookfield, "Capturing the Benefits of Infrastructure Debt," July 2025 Link
13. MetLife / Markets Group, "Infrastructure Debt: A Compelling Addition," November 2025 Link
14. Top1000Funds / SIPA, "Infrastructure Debt Out of the Shadows," December 2025 Link
15. AEW, "Private Infra Lenders Target Solid Margins," 2025 Link
16. Bank for International Settlements, "Financing the AI Boom," 2026 Link
17. US Bureau of Industry and Security, "Rescission of AI Diffusion Rule," May 2025 Link
18. AGBI / Global SWF, "Wealth Funds Racked Up $119bn," January 2026 Link
19. Khaleej Times / Global SWF, "Mubadala invests Dh120 billion in 40 transactions," January 2026 Link
20. Deloitte, "GCC Sovereign Wealth Funds at the Forefront," November 2025 Link
21. PwC, "Unlocking the Data Centre Opportunity in the Middle East," 2025 Link
22. Built In, "Where AI Data Centers Are Headed After 2025's Boom" Link
This article is published for informational purposes only and does not constitute investment advice, a financial promotion, or an offer of securities. The views expressed reflect analysis of publicly available information and should not be relied upon as the basis for any investment decision.