Scalable data centres will decide who wins AI infrastructure returns
AI demand is surging, but value won't flow to whoever builds the most facilities. It will flow to operators who can scale faster than the market. That takes power, permits, cooling, and a balance sheet built to absorb surprises.
Business analyst Nnadozie Odinaka puts it plainly: "Investors fixate on occupancy and megawatts. The real question is whether your operator has the financial and operational capacity to scale at the speed AI demands."
Why scale now defines the asset
AI workloads draw four to six times more electricity per server than traditional compute. Advanced GPU racks can hit 500 kW. A single training cluster can require 100+ MW - the size of a small city's grid link.
The headline spend is massive - over $320 billion in 2025, up from $241 billion in 2024 - but capital intensity and execution risk are where investors get burned. As Odinaka warns: "If your operator isn't securing substations, cooling, and permits early, they'll miss the market window. By the time they're scrambling for megawatts, you're already losing."
What this means for real estate and construction teams
Site control, interconnection position, and utility relationships are now the bottleneck. Lead times for transformers, switchgear, and substations stretch well beyond a year. Permitting paths and cooling choices determine whether a site can hit timelines.
Legacy designs won't carry future density. AI-ready builds need liquid cooling (direct-to-chip and rear-door), proven heat rejection strategies, and water risk mitigation. Retrofitting legacy halls can blow past 50% of original budgets - and cost overruns of 50-100% are common if scope shifts late. "Your operator needs financial breathing room," Odinaka says.
Red flags to spot early
- No locked-in power pipeline (interconnection queue position, substation plans, PPAs, or utility MOUs).
- Late negotiations for power or land options in constrained nodes.
- Air-only cooling for AI halls with no liquid roadmap or budget.
- No long-lead procurement plan for transformers, switchgear, and chillers.
- Thin contingency, weak EPC terms, and no escalation coverage.
- Single-source financing with tight covenants and little headroom.
Due diligence questions that actually matter
- Can they double or triple installed capacity within 18 months without re-permitting the entire site?
- How much reserved power do they control for the next five years, and what's the firm interconnect date?
- Do they have executed PPAs or utility agreements, not just intent letters?
- Where are they in the interconnection queue and substation design schedule?
- Are purchase orders placed for transformers, switchgear, pumps, and heat exchangers?
- What is the liquid cooling plan (direct-to-chip, rear-door), with CFD studies and vendor commitments?
- Water strategy: source, contingency (air-assisted dry coolers), WUE targets, and heat reuse options.
- Permitting: critical path, milestones, and a realistic schedule with float.
- Contracting: GMP or target price, change-order controls, retention, and performance guarantees.
- Capital structure: diversified funding, liquidity runway, and cushion for a 50% budget swing.
What winning operators look like
- Deep utility partnerships: early queue positions, joint substation planning, and credible delivery dates.
- Modular build programs: repeatable blocks, parallelized trades, and pre-fab MEP to compress schedules.
- Diverse funding: equity, structured debt, and forward commitments that keep projects moving through shocks.
As Odinaka puts it, the best teams treat scalability as a financial engineering problem as much as a construction problem. Balance sheet, procurement, and utility access are design inputs now.
The economic backdrop
Research suggests AI could add trillions to global GDP by 2030 - but only if infrastructure scales on time. For prepared investors, this is a once-in-a-generation build cycle. For the underprepared, it's a recipe for stranded assets.
Want a quick sense of the load profile shift? See how data centre electricity needs are trending via the IEA's overview of the sector here.
Practical moves for real estate and construction leaders
- Prioritize sites with upgradable substations, dual feeds, and short distances to high-capacity lines.
- Lock water alternatives early (recycled, dry coolers) and design for heat reuse where possible.
- Pre-buy long-lead gear; hold spares for critical items to de-risk single points of failure.
- Standardize on liquid-ready white space with clear density tiers (80-300+ kW/rack).
- Negotiate GMPs with escalation allowances and clear change-order thresholds.
- Stage capital in tranches tied to utility milestones, not just construction progress.
The bottom line
"In five years, we'll see which operators treated data centres as the foundation of the AI economy," Odinaka says. Foundations dictate how high you can build - and how much value you capture.
This isn't about the operator with the most facilities. It's about the one with foresight, disciplined financing, and the execution engine to scale profitably.
If your team needs to get up to speed on AI's impact on infrastructure and project economics, explore role-based learning options here.
Your membership also unlocks:
