The Global Opportunity to Build & Lease AI Data Centers

AI isn’t just a software story—it’s a power, real-estate, and specialized-infrastructure story. Hyperscalers and model labs need massive, reliable capacity fast, and they increasingly rent it. In the U.S. alone, data-center construction spending hit a record ~$40B annualized in June 2025, with growth still accelerating as generative AI scales out. (Reuters)

Demand that outpaces traditional build cycles

Analysts now project trillions in capex this decade to meet compute needs. McKinsey estimates ~$6.7T in global data-center investment by 2030, with ~$5.2T tied to AI-capable facilities. (McKinsey & Company) Meanwhile, the IEA expects total data-center electricity consumption to roughly double by 2030, approaching ~945 TWh in its base case (just under 3% of global power). (IEA) Other recent analyses point even higher, up to ~1,400 TWh (~4% of global power), underscoring the “demand shock” scenario driving this asset class. (McKinsey & Company)

Why lease (wholesale colo, powered shell, BTS)?

Speed-to-AI beats owning the land and shell for many operators. That’s why wholesale colocation (ready-to-use MW blocks), powered shells (developer delivers power, shell, and cooling; tenant finishes), and build-to-suit (BTS) are booming. Scarcity is lifting lease pricing: CBRE reported a 6.5% jump in H1-2024 average asking rates across primary wholesale markets to ~$174/kW/month for 250–500 kW blocks, with AI demand pushing further in H2. (CBRE)

Technical shift: density, liquid cooling, and grid constraints

AI training clusters are dense. Racks above 100 kW are no longer theoretical; they’re becoming the reference design for state-of-the-art GPU clusters—even though just ~1% of operators reported >100 kW/rack as of 2024. (Uptime Institute Blog) Systems like NVIDIA’s GB200 NVL72 are architected for liquid cooling and extremely high intra-rack bandwidth; reference materials and integrators peg ~120 kW/rack as a realistic envelope for this class. (NVIDIA) Result: sites must be engineered for liquid cooling (rear-door heat exchangers or direct-to-chip), robust water or heat-rejection strategies, and higher floor loading—plus short, redundant fiber paths for low latency.

On the supply side, grids are the new bottleneck. Developers are co-locating with firm power (gas, hydro, nuclear) and embracing behind-the-fence generation. Texas, with abundant gas, rights-of-way, and transmission, has emerged as a fast-track region aligning upstream energy investment with AI load growth. (Midland Reporter-Telegram)

Site selection: the five essential filters

1. Power: MW/GW-scale, firm, and expandable (PPA or on-site gen).
2. Cooling & water: liquid-cooling readiness; water-use minimization and heat-reuse options.
3. Fiber: diverse long-haul and metro rings; proximity to IXPs and cloud on-ramps.
4. Regulatory & incentives: land-use certainty, tax abatements, and accelerated interconnects.
5. Resilience: dual feeds, top-tier uptime design, and verifiable O&M capability.

Sustainability is also commercially material. The Uptime Institute notes industry PUE benchmarks and the operational challenge of densifying for AI while maintaining efficiency; liquid cooling can materially reduce footprint and energy per unit of compute when designed well. (Uptime Institute) Vendors now claim step-function performance per watt with next-gen platforms, strengthening the green-premium thesis for modern builds. (NVIDIA)

Business models for investors

• Wholesale colocation: 10–50 MW blocks leased to hyperscalers/AI labs; long-dated, investment-grade covenants.
• Powered shell: Developer delivers site, shell, and megawatts; tenant completes white space—faster lease-up and capital recycling.
• Build-to-suit/JV: Structured with take-or-pay offtake, often backed by creditworthy counterparties; aligns returns with tenant expansion roadmaps.
• Campus roll-ups: Assemble regional platforms for REIT exit or private-market consolidation.

Returns, risks, and how to underwrite

• Revenue: Anchored by $/kW/month with escalators; cross-connects and managed services add uplift. (Recent U.S. wholesale guidance: ~$174/kW/month average—with AI-led markets higher.) (CBRE)
• Capex: AI-grade fit-outs cost more per MW (liquid cooling, switchgear, high-density busways). Underwrite phased delivery to de-risk.
• Interconnect risk: Queue times and substation upgrades can blow timelines—secure PPAs and LOIs before breaking ground.
• Tech obsolescence: Design for modularity; ensure aisles and manifolds can accept next-gen racks without core rebuilds.
• Policy & power prices: Model power-price volatility and carbon policy; consider behind-the-meter gas, hydro, SMR or heat-reuse credits where viable. (Analysts expect AI to be a material driver of national power demand growth this decade.) (IEA)

Where to build (and why)

• North America: Record construction spend; deep hyperscaler pipelines; Texas and Virginia remain power-fiber hubs, with secondary markets rising. (Reuters)
• Nordics: Renewable baseload, cool climates, and attractive policy for heat reuse.
• Middle East: Low-cost energy and ambitious national AI programs spur BTS/JV deals.
• Africa (select markets): Proximity to load growth, improving grids, expanding subsea landings—often best pursued via PPPs and sovereign-backed power solutions, with strong developmental impact.

How offshore capital can plug in—now

1. Pre-development equity for land, permits, and interconnects—secured with tenant ROFRs.
2. Construction finance tied to signed offtake; tranche by energization milestones.
3. Sale-leaseback or REIT exit on stabilization.
4. Energy-linked strategies: Pair sites with firm generation (gas, hydro) and monetize heat reuse or capacity products.

Bottom line: AI’s compute curve is outpacing grid and construction cycles. Leasing models that deliver MWs fast, dense, and efficient are primed for outsized, durable cash flows—particularly where developers can secure power and interconnects ahead of the wave.