$725B
2026 hyperscaler capex
Up ~77% YoY from $410B in 2025 — the vast majority landing in datacenters and what fills them.
Source: Microsoft, Alphabet, Amazon, Meta combined guidance
Market
The datacenter market,
in numbers.
Independent third-party data on the buildings the AI economy runs on — demand, supply, the cost of a megawatt, and the capital now forming around the asset class. All figures attributed to source.
Demand & supply
1.4%
Primary-market vacancy
All-time record low. Demand has consistently outrun new supply.
Source: CBRE, North America Data Center Trends, H2 2025
9,432 MW
Under construction
+36% YoY in primary markets. 74.3% preleased before delivery — capacity is sold before it exists.
Source: CBRE, H1 2025
~100 GW
New capacity 2026–2030
Approximately $1.2 trillion of new real-asset value creation over five years.
Source: JLL, 2026 Global Data Center Outlook
+165%
Datacenter power demand
Reaching ~92 GW by 2027; demand growth structurally exceeds grid expansion.
Source: Goldman Sachs Research, 2030 vs. 2023
67%
Hyperscale share by 2031
Hyperscale operators projected to account for two-thirds of global datacenter capacity.
Source: Synergy Research Group
Unit economics
The cost of
a megawatt.
The megawatt is the unit of account of the datacenter market — capacity is leased by it, facilities are priced by it, and capital is raised against it. Brookfield's 2026 outlook puts the all-in cost of AI-ready capacity at $40 million or more per usable IT megawatt.
| Layer | Capital | Character |
|---|---|---|
| Powered shell & site infrastructure | $10M+ / MW | building, switchgear, cooling plant, distribution — 30–40 year asset life |
| Compute fit-out inside | $30M+ / MW | racks, networking, liquid cooling, GPUs — refreshed every 3–6 years |
| Fully equipped AI facility, all-in | $40M+ / IT MW | a single 100 MW facility represents roughly $4 billion of capital |
Source: Brookfield 2026 Investment Outlook. Facility figures are order-of-magnitude benchmarks; actual costs vary by market, density and design.
The split matters as much as the total: the majority of the capital sits in fast-refreshing compute, but it all depends on the powered shell beneath it — the layer that appreciates while everything above it depreciates. That asymmetry is the heart of our thesis.
Returns & valuation
Priced like infrastructure,
built like a growth market.
4.25–6.25%
Stabilized cap rates
Where stabilized datacenter assets currently trade.
12–19%+
Stabilized leveraged IRRs
Developer targets, depending on risk profile.
15%+
Development IRRs
Targeted by leading operators on ground-up projects.
Source: RCLCO.
Whyte Consolidated underwrites to these benchmarks — we do not project returns above them. The spread between stabilized cap rates and development returns is earned in the stages before the first rack arrives, as described in our strategy.
Capital formation
The institutions
have arrived.
The clearest market signal is not a forecast — it is the capital already announced. Private credit, infrastructure equity and insurance capital are building dedicated platforms to own and finance datacenter capacity, and the facility is the collateral in every one of them.
| Platform / program | Scale | Notes |
|---|---|---|
| NVIDIA financing initiative | $500B+ target | six platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR |
| Brookfield AI infrastructure program | $100B | anchored by a targeted $10B fund across datacenters, power and transmission |
| BlackRock / GIP AI Infrastructure Partnership | up to $100B | $30B equity sought, up to $100B including debt |
| Apollo — Broadcom/Anthropic transaction | $35B | initial asset-backed capital solution supporting more than 1 GW |
| KKR Helix digital infrastructure | $10B+ | committed capital across datacenters, power and connectivity |
| Blackstone — Google TPU cloud JV | $5B equity | targeting 500 MW of dedicated capacity |
| IREN investment-grade GPU financing | $3.65B | ~96% of GPU capex funded at investment grade; Fitch A / DBRS A(low) |
Source: company announcements and primary financing disclosures, as of August 2026. Targets and programs are ambitions, not committed capital; several may overlap.
We examined what this wave of capital does — and does not — mean in our analysis of NVIDIA's $500 billion financing initiative. The disciplined read: the headline numbers are frameworks, not order books — but the direction is unambiguous. The datacenter is becoming an institutional asset class, and power-secured capacity is its scarcest input.