The scale of capital flowing into data center infrastructure is historically unusual. Sovereign wealth funds, pension funds, private equity, and hyperscalers are simultaneously competing for the same assets — land with power, operational facilities, and development sites — at a pace and price that has no direct precedent in infrastructure investing.
The fundamental question any serious investor or advisor must answer is: is this a structural reallocation driven by genuine long-term demand, or a capital-cycle phenomenon driven by the AI narrative? The evidence points strongly toward the former — but the risks are real, and concentrating capital without understanding them is a mistake.
The Demand Driver: AI Is Not Optional Infrastructure
Every AI model — training and inference — runs in a data center. The compute requirements for frontier AI models have been doubling roughly every six months. GPT-4 required approximately 10x the compute of GPT-3. The next generation of models will require multiples more. This is not speculative demand: hyperscalers have published multi-year capital expenditure guidance totalling hundreds of billions of dollars, and that guidance has been consistently revised upward.
The demand is not limited to AI training. Inference — the running of trained models to respond to user queries — scales with adoption. As AI-enabled products reach hundreds of millions of users, the inference workload grows continuously, creating a persistent and expanding demand for compute capacity that must physically sit somewhere.
The Institutional Investment Case
Data centers offer a return profile that is structurally attractive to institutional investors: long-term contracted revenue (hyperscaler leases of 10–20 years), inflation-linked escalators, investment-grade or near-investment-grade counterparty credit, and a real asset underpinning that provides capital value. The combination of bond-like income and real asset backing is rare, and institutional portfolios have historically underallocated to it.
Yield on cost for stabilised hyperscale data centers — particularly those with long-term leases in place — typically ranges from 8–14% depending on location, power configuration, and tenant credit. Development margins (the difference between build cost and stabilised value) have historically run at 20–40% in supply-constrained markets, attracting value-add and opportunistic capital alongside core and core-plus.
"Data centers combine the income profile of infrastructure with the scarcity dynamics of prime real estate — in a sector where demand is growing faster than supply can respond."
Capital Sources Currently Active
| Investor Type | Strategy | Target Return | Typical Position |
|---|---|---|---|
| Hyperscalers (Google, Microsoft, Amazon) | Own & operate; co-location; cloud | Internal ROI metrics | Developer / operator |
| Infrastructure PE (Blackstone, KKR, Brookfield) | Development, stabilised acquisition | 15–25% IRR | Equity developer/owner |
| REITs (Equinix, Digital Realty, Iron Mountain) | Stabilised co-location, retail/wholesale | 8–12% total return | Long-term owner/operator |
| Pension & sovereign wealth | Core stabilised; long-lease assets | 7–10% IRR | Equity / preferred equity |
| Infrastructure debt funds | Construction and term debt | SOFR + 250–500bps | Senior secured lender |
| Investment banks | Construction finance, mezzanine | SOFR + 400–700bps | Construction / mezzanine |
Where the Risks Are Concentrated
Power availability is the single most consequential risk. A data center without a path to power is not an asset — it is a stranded development. Grid interconnection queues in Northern Virginia, the UK, the Netherlands, and parts of Singapore run 3–7 years. Development projects that have not secured power commitments are significantly higher risk than they appear on paper.
Construction execution is the second major risk. Supply chain bottlenecks for high-voltage transformers, cooling equipment, and specialist mechanical and electrical contractors have extended construction timelines and increased costs. A JPMorgan analysis in mid-2026 found over 60% of planned 2027 capacity not yet under construction.
Technology evolution creates long-term obsolescence risk — particularly for older, lower-power-density facilities that cannot accommodate next-generation GPU compute. New builds must be designed for power densities of 30–100kW per rack or higher to remain relevant through a standard 20-year asset life.
Concentration is increasingly a concern. Markets like Northern Virginia now host a disproportionate share of global hyperscale capacity — creating grid, permitting, and single-event concentration risk that diversified portfolios should manage deliberately.
What Separates Good Investments from Bad
The most fundable data center investments — at any position in the capital stack — share three characteristics: secured power (ideally behind-the-meter or with grid capacity confirmed), a creditworthy tenant or anchor customer, and a management team with a demonstrated track record of delivering complex M&E construction on schedule. Everything else is negotiable. These three are not.
Finance Your Data Center Project
OAKRG advises on data center project finance, construction debt, hyperscale equity raises, and energy-linked infrastructure capital across North America, Europe, and Asia-Pacific.
Speak with an Advisor