In 2024, private equity firms deployed over $108 billion into data centre and adjacent digital infrastructure deals, more than triple the prior year. Separately, the hyperscale technology companies committed capital of a different kind: Microsoft allocated $80 billion in corporate capex to AI data centres in its fiscal year 2025; Amazon earmarked $86 billion. These are distinct categories of capital. The PE figure reflects fund-based investment with finite holding periods. The hyperscaler figures reflect balance sheet spending with no embedded exit clock. The structural tension examined here applies to the former.
A typical closed-end infrastructure fund targets exits within five to seven years of initial investment, within a total fund life of ten to twelve years. An AI data centre is a 20 to 30-year asset. Its building shell lasts 50 years. Its power and cooling systems operate on eight to 15 year replacement cycles. Its GPU hardware turns over every three to five years, sometimes faster. The mismatch is not subtle.
The closed-end fund was designed for corporate buyouts: acquire, improve, sell. Transplanted into infrastructure, the model fits poorly. Research from Stanford Graduate School of Business, analysing the cash flows of 633 infrastructure funds, found that returns were below market rates and just as volatile as traditional private equity. The risk-return profiles of closed infrastructure funds were, the study concluded, indistinguishable from those of standard buyout vehicles. The study examined historical fund vintages, but the structural observation holds: fixed exit windows impose constraints that sit uneasily with long-duration assets.
This is not to dismiss private equity's role in building the sector. Blackstone's acquisition of QTS through its perpetual capital vehicles, now representing over $80 billion in data centre assets, demonstrates what happens when fund structure matches asset duration. DigitalBridge, Brookfield, and Macquarie have each deployed meaningful capital into digital infrastructure. Several of these managers have launched open-ended or perpetual vehicles specifically to hold infrastructure indefinitely. The sector's growth would have been slower without them.
The question is narrower than it appears. It is not whether PE capital belongs in data centres. It does. It is whether the standard closed-end structure produces optimal outcomes when applied to AI native facilities whose technical demands differ materially from the colocation and enterprise buildings that dominated the previous cycle. When a fund must exit within a fixed window, every decision bends toward saleability at year six, not performance at year twenty. Site selection, tenant terms, capex phasing, and technical specifications all reflect that deadline. Power headroom, cooling flexibility, and deep grid interconnection, the features that matter most across decades, are precisely the ones most likely to be sacrificed for near term lease optics.
Open-ended and perpetual vehicles mitigate this by removing the exit deadline. But the majority of PE capital deployed into data centres still operates within closed-end structures.
NVIDIA's annual release cadence now drives the entire data centre supply chain. Its roadmap, Blackwell to GB300 to Vera Rubin, dictates when hyperscalers expand, how facilities are designed, and which cooling systems are adopted. Each generation brings higher power density. Current generation Blackwell GPUs draw between 700 and 1,200 watts per chip, depending on configuration, with the most commonly deployed B200 variants operating at 1,000 watts. The GB200 NVL72, NVIDIA's flagship rack-scale system, requires approximately 132 kilowatts per rack. By 2027, NVIDIA's Kyber architecture is projected to push rack densities to 600 kilowatts, packing 576 GPUs into a single enclosure. Industry roadmaps already target one megawatt per rack by the end of the decade.
Servers that once lasted seven to ten years are now replaced every three to five years, sometimes faster in AI-intensive environments. A facility that begins operation in 2026 will see four to six full hardware refresh cycles before its building reaches the end of life. Each transition demands capital for refit, cooling adaptation, and power upgrades.
A fund structured to exit within five to seven years encounters, at most, one of these transitions. Its incentive is to underwrite the initial deployment and sell before the first major refresh. The next owner inherits the deferred capital requirements. The pattern repeats with each successive sale, progressively eroding the asset's technical resilience.
Research from the University of Illinois makes this dynamic explicit. Facility capex amortises over 15 to 30 years; GPU capex amortises over three to five. An investor holding for six years captures a fragment of the facility's economic utility while absorbing the full weight of at least one hardware obsolescence cycle.
The PE industry is currently demonstrating what happens when closed-end structures meet an unreceptive exit environment. As of mid 2025, the industry held an estimated $1 trillion of unrealised assets. The immediate cause was the rapid rate increases of 2022 and 2023, which collapsed bid-ask spreads on exits. But it is the closed-end structure that converted a cyclical valuation problem into an acute liquidity problem. Managers cannot simply wait for markets to recover. Their fund terms impose disposition deadlines.
Buyout managers were holding twice as many portfolio companies as in 2019. Exit volumes across all PE and VC strategies fell from 1,210 in 2021 to 323 in 2023, recovered partially to 516 in 2024, and declined again to 321 in 2025. The backlog has produced a range of structural responses: continuation funds, dividend recapitalisations, NAV-based lending, and secondary sales at a discount. Some represent genuine innovation. Others add complexity and cost that serve the manager's liquidity needs rather than the asset's long-term performance. Bain's Hugh MacArthur has described the situation as a five-year problem comparable to the aftermath of 2008.
For data centres, the implications are direct. An operator backed by a fund approaching its terminal date faces pressure to sell into a weak market or restructure the vehicle at additional cost. Neither outcome serves the asset or its tenants. Hyperscale offtakers understand this well. A prospective anchor tenant evaluating two identical facilities will prefer the one whose capital carries no embedded disposition deadline.
The Gulf Cooperation Council states present a structural opportunity precisely because their dominant capital pools operate on different time horizons. GCC sovereign wealth funds collectively manage approximately $5 trillion in core assets. In 2025, the seven largest Gulf funds deployed $119 billion, up 43 per cent from the prior year. Artificial intelligence and digitalisation attracted $66 billion of sovereign wealth fund investment globally, with Abu Dhabi's Mubadala the single largest sovereign deployer at $12.9 billion.
These funds serve different purposes and should not be conflated. The Abu Dhabi Investment Authority and Kuwait's Future Generations Fund are intergenerational savings vehicles: their mandates prioritise wealth preservation and long-duration, inflation-protected returns. Saudi Arabia's Public Investment Fund is a development fund, mandated under Vision 2030 to drive domestic economic transformation. Both models align with long-duration infrastructure, but for different reasons. The savings mandates seek duration matching. The development mandates seek infrastructure that directly serves national objectives: digital sovereignty, compute localisation, and economic diversification.
This distinction matters for anyone assessing GCC capital flows into AI infrastructure. The investment is not purely financial. It reflects sovereign priorities that include national security, technological self-sufficiency, and the positioning of Gulf states as globally significant nodes in the AI landscape. Capital partners who understand this operate on a different footing from those who see the region solely as a source of patient funding.
Saudi Arabia's national data centre strategy targets 1.5 gigawatts of capacity by 2030. HUMAIN, the PIF-backed sovereign AI entity launched in early 2025, has signalled ambitions of up to $100 billion in AI investment, with the phasing between committed and aspirational capital still taking shape. Projects such as the HUMAIN AirTrunk campus in Riyadh indicate sustained domestic demand. The Saudi market is projected to grow from $2.1 billion in 2025 to $4.35 billion by 2031. Cloud-first directives from the Digital Government Authority require government agencies to migrate workloads into sovereign infrastructure, creating predictable, contracted demand.
The UAE has moved in parallel. Khazna, the country's largest data centre operator and a G42 subsidiary, manages over 30 facilities with nearly 650 MW of capacity. It plans to add more than 1 GW of AI-ready capacity across the UAE, Saudi Arabia, and Europe. Its 100 MW AI-optimised facility in Ajman and its role as infrastructure provider for the planned 5 GW UAE US AI Campus in Abu Dhabi reflect the scale of the UAE's ambition. Qatar's Investment Authority has formed a partnership with Blue Owl Capital to launch a digital infrastructure platform with more than $3 billion in initial assets, part of a broader national commitment to digital infrastructure under Qatar's Digital Agenda 2030.
Across the GCC, the pattern is consistent: sovereign capital deploying at scale into AI infrastructure as a core element of national strategy, not as a cyclical investment thesis.
The GCC's position in AI infrastructure cannot be assessed without reference to US chip export controls. The Bureau of Industry and Security has imposed, revised, and rescinded successive rounds of controls on advanced GPU exports since October 2022. The Biden administration's AI Diffusion Rule, which would have imposed country tier allocation caps, was rescinded in May 2025. A replacement framework has been signalled but not yet issued.
For Gulf-based developers and their capital partners, this introduces a procurement variable that must be managed alongside conventional infrastructure risks. Saudi Arabia and the UAE fall under Country Group D:4, and the updated controls issued in October 2023 tightened licence requirements for advanced chip exports to these jurisdictions. The Trump administration's stated policy of promoting American AI technology exports to allied nations, reinforced by major chip deals announced during the President's May 2025 visit to the region, has eased the near term outlook. But the regulatory framework remains subject to change. Any long-duration investment in the region must price in the possibility of future tightening.
This reinforces the case for patient capital rather than undermining it. Navigating the export control environment requires sustained government-to-government engagement, compliance infrastructure, and the kind of institutional credibility that takes years to build. Platforms embedded in the region's sovereign ecosystem, with the time horizon to manage regulatory complexity across multiple chip generations, hold a structural advantage over those operating on shorter cycles.
The counterargument to patient capital is speed: short-cycle funds deploy faster and capture market windows ahead of slower institutional capital. The argument contains a kernel of truth but misidentifies the binding constraint. In AI data centre development, the constraint is not capital deployment. It is a demand for contractualisation.
Building without a contracted offtake is the primary risk. It is also the risk that short-cycle capital is most likely to encourage, because waiting for binding agreements often exceeds the fund's deployment window. The result is speculative capacity: facilities built on the assumption that demand will arrive before the exit clock expires.
When that assumption holds, returns can be attractive. When it does not, the consequences are severe. Premature deployment against speculative capacity strands power assets and erodes operator credibility. As the IBM chief executive has observed, servicing the capital cost of multi-gigawatt campuses requires hundreds of billions of dollars in annual profit to break even, a requirement grounded in current hardware economics, not speculative efficiency gains.
The alternative is demand-led sequencing: securing contractual offtake before committing capital and phasing construction against contracted demand. The approach is slower at the outset but structurally more resilient across the asset's full life.
It requires a particular kind of capital partner. The investor must hold through an origination phase that, in Claymont Equivator's experience, takes 60 to 120 days for initial offtake frameworks and considerably longer for binding contracts. They must accept phased deployment, with capital drawn down against contracted milestones. And they must accept returns designed to compound over 15 to 25 years rather than crystallise in a single exit.
This profile fits sovereign and quasi-sovereign capital, large pension funds, and select open-ended infrastructure mandates. It does not fit the standard closed-end PE fund, though a growing number of managers have developed perpetual vehicles that come closer.
The GCC context makes duration alignment essential for a further reason: execution timelines. Grid access, permitting, power procurement, and sovereign partnership structures all operate on timeframes that exceed a typical fund cycle. In Saudi Arabia, a data centre project must navigate the power allocation process involving the Saudi Electricity Company, the Electricity and Cogeneration Regulatory Authority, and potentially the Saudi Power Procurement Company, alongside approvals from multiple regulatory bodies and, in many cases, formal partnerships with sovereign entities.
These processes are the mechanisms through which durable assets are built. They demand capital that can participate without the distortion of an approaching exit. An investor who must demonstrate realisable value within 60 months will inevitably seek to compress the regulatory and partnership timelines that give GCC assets their resilience.
In Europe, similar dynamics apply in a different form. Grid interconnection in the Nordics, environmental permitting in France, and energy procurement in the United Kingdom each carry distinct regulatory cadences. Facilities designed to operate on low-carbon power, anchored by long-term renewable energy purchase agreements, derive their economic advantage from precisely the kind of commitment that short-cycle capital cannot accommodate.
Different capital structures suit different asset profiles. The specific demands of AI native data centres, multi-decade facility life, accelerating hardware refresh cycles, complex regulatory environments, and sovereign partnership requirements are best served by capital whose time horizon matches the asset's duration. The closed-end model has contributed meaningfully to scaling the sector, but its limitations become more pronounced as data centres transition from enterprise colocation to AI native infrastructure, where the technical complexity and capital intensity of each generation increase materially.
The institutions whose capital structures most naturally align with these assets are sovereign wealth funds, long-dated pension mandates, and development capital vehicles with multi-decade horizons. Their interests track the asset across hardware cycles, regulatory evolutions, and technological transitions. Claymont Equivator's model, securing offtake before capital is committed and phasing deployment against contracted demand, reflects this alignment.
The AI data centre buildout is the defining infrastructure programme of this decade. The capital structures financing it will determine whether the result is a stock of durable, productive assets or a cycle of speculative construction and premature disposition.
The GCC, with its sovereign capital, its national digital transformation agendas, and its strategic commitment to domestic AI capability, offers a structural home for duration-aligned investment. Select European jurisdictions, where renewable energy access, regulatory stability, and institutional depth run deep, offer another.
The question is no longer whether demand exists. It does. The question is whether the capital deployed against it is fit for purpose across the full life of the assets it finances.
1. A Bridge Too Far: The Pitfalls of Private Infrastructure Funds. Stanford Business. Link
2. Unlocking data center ecosystems: The power of private equity. Schneider Electric. Link
3. When Private Equity Meets the Data Center Gold Rush: Exit Strategies for Private Equity. Georgetown University. Link
4. Private Equity: In the Doldrums and Out of Favor with Some Institutional Investors. CEPR. Link
5. Global Private Markets Report 2026 Private equity: Clearer view, tougher terrain. Mckinsey & Company. Link
6. 2025 Private Markets Year-End Review. AlterDomus. Link
7. U.S. Private Equity Market Recap - September 2025. Ropes & Gray. Link
8. Data Centers in an AI-Driven Era: Key Trends Reshaping Valuation. SitusAMC. Link
9. Data center infrastructure market: AI-driven CapEx pushing IT and facility equipment spending toward $1 trillion by 2030. IOT Analytics. Link
10. Ticking Time Bomb: AI, Data Centers, and the Looming E-Waste Crisis. Data Center Knowledge. Link
11. Rearchitecting Datacenter Lifecycle for AI: A TCO-Driven Framework. Arxiv. Link
12. Gulf cooperation council sovereign wealth funds at the forefront of a strategic global expansion. Deloitte. Link
13. Region’s wealth funds racked up $119bn of spending in 2025. AGBI. Link
14. Unlocking the data centre opportunity in the Middle East. PWC. Link
15. Saudi Arabia’s Data Centre Expansion: Regulatory Framework and Strategic Considerations. GreenbergTraurig. Link
16. Saudi Arabia ICT New Data Center Strategy to Accelerate AI and Cloud Expansion. International Trade Administration. Link
17. Saudi Arabia Data Center MArket Size & Share analysis - Growth Trends And Forecast (2026 - 2031). Link
18. "There is 'no way'" - IBM CEO says current AI data center trends are unsustainable, and he would know. Techradar. Link
19. In the Know: Private Equity in Data Centers – Growth, Risks & Opportunities. Morgan Lewis. Link
20. Meeting the Demands of the Digital Age: What Today’s Data Centers Require for Long-Term Success. PDGgroup. Link
21. Keeping data centres up-to-date to meet ever-evolving challenges. Haskoning. Link
22. QTS Realty Trust to Be Acquired by Blackstone Funds in $10 Billion Transaction. Blackstone. Link
23. Blackstone still positive on data centres as infrastructure business grows 34%. IPE Realassets. Link
24. Navigating GPU Export Controls and AI Use Restrictions in Data Center Operations. GreenbergTraurig. Link
25. AI Chip Export Controls: A New Challenge for Data Center Operators. Data Center Knowledge. Link
26. Department of Commerce Announces Rescission of Biden-Era Artificial Intelligence Diffusion Rule, Strengthens Chip-Related Export Controls. Bureau of Industry and Security, U.S. Department of Commerce. Link
27. US Commerce Department Announces New Export Compliance Expectations Related to Artificial Intelligence. Mayer Brown. Link
28. Khazna Data Centers to Add Over 1GW Hyperscale Capacity Across Multiple Countries By 2030. Khazna Data Centers. Link
29. Qatar Investment Authority and Blue Owl Capital Enter Agreement to Establish Digital Infrastructure Partnership. Blue Owl. Link
30. 1,000 homes of power in a filing cabinet - rising power density disrupts AI infrastructure. Goldman Sachs. 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.