AI data centre build-out hits grid bottlenecks and unprepared supply chains

AI data centre build surge: why 2025 is the tipping point for construction and supply chains

Developers are pivoting from traditional cloud facilities to high-density, liquid-cooled builds built for AI. The shift hits an inflexion point in 2025, bringing new technical demands, higher costs, and fresh pressure on utilities and suppliers.

Cost matters, but delivery risk is louder. Survey data shows 83% of sector leaders think local supply chains are not ready for advanced cooling, and 48% say grid access is the biggest schedule blocker. That's the bottleneck many projects will feel first.

What's different about AI-ready facilities

• Higher rack densities with direct-to-chip or immersion cooling replacing air as the primary medium.
• Larger electrical loads per hall, which drive upsized switchgear, UPS, generators, and heat-rejection systems.
• More complex commissioning and controls, with tight integration between IT loads, liquid loops, and facility MEP.
• Water strategy and heat reuse becoming site-selection factors, not afterthoughts.

Cost signals to watch

• 5.5% average global inflation in 2025 for traditional builds (US$/W basis across 52 markets).
• 7-10% premium in the US for AI facilities versus traditional projects of similar IT capacity.
• Key drivers: liquid cooling hardware (CDUs, manifolds, immersion tanks), larger electrical equipment, structural upgrades, and longer commissioning.

Supply constraints now driving schedule risk

• Advanced cooling readiness: 83% say local suppliers are not prepared to support liquid cooling at scale.
• Grid connection: 48% call it the top delivery obstacle, with utility interconnect queues and transformer lead times stretching start dates.
• Other pinch points: MV/LV switchgear, large-format generators, heat-rejection packages, CDUs, valves, and specialty piping.

Procurement and design moves to make now

• Engage utilities early; secure preliminary capacity letters and queue positions before land close.
• Co-develop substations where viable; evaluate interim on-site generation and storage for phased energization.
• Standardize reference designs at ≥25-40 kW/rack with native liquid-cooling options (direct-to-chip or immersion).
• Dual-track cooling: design for liquid-first with limited air assist to de-risk equipment availability.
• Frame agreements with OEMs for switchgear, UPS, generators, CDUs, and valves; lock factory slots.
• Qualify Tier-2 and Tier-3 suppliers; regionalize fabrication for manifolds, skids, and prefab MEP.
• Pre-buy long-lead items against baseline capacity, then allocate across campuses as permits land.
• Water risk: prioritize reclaimed sources, dry coolers, or hybrid solutions; plan heat reuse where district networks exist.

Where construction costs are highest in 2025 (US$/W)

• Tokyo: 15.2 (Rank 1; 2024: 1)
• Singapore: 14.5 (Rank 2; 2024: 2)
• Zurich: 14.2 (Rank 3; 2024: 3)
• Osaka: 14.1 (Rank 4; new entry)
• Silicon Valley: 13.3 (Rank 5; 2024: 4)
• New Jersey: 12.9 (Rank 6; 2024: 5)
• Oslo: 12.4 (Rank 7; 2024: 9)
• Auckland: 12.3 (Rank 8; 2024: 6)
• Stockholm: 12.3 (Rank 9; 2024: 11)
• Helsinki: 12.3 (Rank 10; 2024: 8)

Demand remains buoyant in these hubs, and baseline construction costs are high. Established US and European markets round out most of the top 15.

Implications for developers, owners, and contractors

• Site selection now hinges on grid capacity, interconnect timelines, latency routes, water, and heat-reuse partners.
• Permitting must include substation scopes, liquid-cooling infrastructure, and water discharge pathways.
• Delivery models: consider alliancing/IPD to secure long-lead gear early and align risk across the MEP stack.
• Workforce: train field teams for liquid-cooling install, flushing, chemical treatment, and leak detection.
• Leasing: build in TI allowances and density bands for liquid-first fitouts; document maintenance responsibilities for loops and manifolds.

90-day action plan

• Lock target rack densities and cooling approach; update basis-of-design and one-lines.
• Submit utility pre-apps; get indicative capacity and substation siting studies.
• Issue RFQs for switchgear, UPS, generators, CDUs, and valves; reserve production slots.
• Prequal vendors for direct-to-chip and immersion; pilot a single room to de-risk installation and O&M.
• Refresh cost plans with a 7-10% AI-premium and 5.5% inflation for traditional scopes; stress-test contingencies.
• Advance modular/e-houses and prefabricated MEP skids to compress field hours.
• Confirm water sources and discharge permits; evaluate dry/hybrid heat rejection.
• Explore off-grid interim solutions to bridge utility delays, subject to permits and noise constraints.

Helpful resources

• ASHRAE guidance on data centre thermal management
• IEA: Electricity report and grid capacity outlook

Upskill your team

If your construction or development team needs practical AI skills for planning, scheduling, and cost control, browse job-focused training here: Complete AI Training - Courses by Job.