AI infrastructure planning is now energy portfolio design

The demand curve changed the planning unit

The IEA's Energy and AI analysis projects electricity generation serving data centres to exceed 1,000 TWh in 2030 in its base case. For companies, the strategic implication is not a single forecast number but the need to connect model growth, hardware density and location choices to physical energy constraints.

Plan one portfolio across four layers

The most resilient strategy treats power as a portfolio rather than a utility bill. It coordinates what the workload needs, what each site can access and how quickly infrastructure can actually be delivered.

• Demand: workload growth, utilization and latency requirements.
• Sites: grid headroom, connection lead time, water and network access.
• Supply: contracts, generation mix, storage and backup architecture.
• Flexibility: scheduling, geographic balancing and graceful degradation.

Turn energy into an operating signal

Energy data should influence workload placement, capacity commitments and product economics. When infrastructure, finance and product teams share the same demand model, the company can see constraints before they become launch delays.

The strategic advantage is not merely lower energy cost. It is the ability to add reliable compute capacity when competitors cannot.

Translate product demand into physical scenarios

The infrastructure plan should connect user demand, model mix, tokens or jobs, accelerator utilization, facility overhead and reserve margin. A base, constrained and accelerated scenario makes uncertainty visible and prevents one top-down megawatt forecast from becoming an unexamined construction target.

Each product team can then see the physical consequence of latency, model size, context length and service-level choices. Optimization becomes a joint decision: some demand needs new capacity, some can move in time or geography, and some can be reduced through model or software changes without harming the user outcome.

• Useful workload by service
• Hardware and utilization path
• Power and cooling requirement
• Flexibility without breaking service promises

Score supply by deliverability and correlation

A portfolio of sites is not diversified if every location depends on the same constrained equipment, fuel market or regulatory assumption. Score power options by commissioning date, firmness, price exposure, carbon characteristics, water dependency and correlation with other sites.

The review should include failure and delay: late grid connection, generation outage, drought constraint, network loss or a workload that cannot move. This reveals where storage, a second route, flexible scheduling or reduced service is the least-cost resilience measure.