The AI Power Bottleneck

Executive Summary

AI infrastructure is increasingly constrained by power rather than compute. Interconnection queues, transformer lead times, and substation capacity are becoming gating items for new data-centre builds. Investors should treat this theme as an infrastructure value chain rather than a single-stock story.

• Demand: hyperscalers, cloud, and colo operators are planning multi‑year capacity expansions.
• Bottlenecks: grid interconnection, transmission, transformers, switchgear, and skilled labour.
• Winners: electrical equipment, EPC/utilities services, thermal management, and select generators.

Why AI is turning into a power problem

Training and inference workloads push sustained utilisation, driving higher load factors than many traditional enterprise data rooms. The practical constraint is not just megawatts on paper, but deliverable power: interconnection approvals, redundancy, and time-to-energise.

Key idea: the “AI buildout” behaves like a capex cycle across utilities and industrials, with multi‑year lead times.

Where the bottlenecks are

• Interconnection: queue backlogs and upgrade costs can delay projects by quarters or years.
• Transformers & switchgear: long lead times, constrained manufacturing capacity.
• Transmission & substations: permitting, land, and skilled construction crews.
• Cooling & power density: thermal constraints increase total facility power draw.

Value chain map

Think in buckets:

Publicly traded exposure

Representative U.S.-listed names by segment:

• Electrical equipment: ETN, HUBB
• Grid build & services: PWR
• Power & generation: CEG, NRG
• Machinery / engines: CAT
• Infrastructure beneficiaries: GEV (generation/grid exposure), VRT (power + thermal)

These are examples of exposure, not recommendations. Always verify current business mix and filings.

What can go wrong

• Timing risk: capex cycles slip; interconnection reforms or policy changes can shift winners.
• Efficiency surprises: compute efficiency gains reduce power per unit of output.
• Execution risk: project delays, cost overruns, and supply chain issues.
• Macro risk: rates and credit conditions affect capex and valuation.

Investment Framework

A practical underwriting approach:

1. Pick the bottleneck (transformers, switchgear, construction labour, cooling).
2. Prefer pricing power and backlog visibility.
3. Watch lead times and order growth for early signals.
4. Size for cyclicality; use risk controls as this behaves like an industrial cycle.

Practical Guidance