AI is often discussed as software. But in 2026, one of the biggest constraints on AI growth is physical: electricity. Data centres, GPUs, and cooling systems are pushing power demand upward — and governments and grid operators are reacting.

What credible sources are saying

Reuters has reported on U.S. power consumption projections, citing the U.S. Energy Information Administration (EIA) and linking rising demand to AI data centres and related workloads: Reuters (Mar 2026). The International Energy Agency has also highlighted growth in data centre electricity demand and the drivers behind it: IEA: Electricity 2024.

Why this changes AI strategy

Power constraints affect AI in three direct ways:

• Cost: electricity becomes a meaningful part of unit economics for inference-heavy products.
• Availability: grid connection limits slow down new data centre builds.
• Reputation: sustainability commitments collide with AI growth targets.

What organisations should do now

If you are deploying AI at scale, treat energy as a first-class metric:

• Measure: track inference cost and energy proxies per product workflow.
• Optimise: use smaller models where appropriate; cache; reduce context where possible.
• Procure responsibly: ask cloud and vendors about energy mix, location, and efficiency.
• Plan for regulation: expect more disclosure and sustainability scrutiny.

AI advantage is now partly an infrastructure advantage — and infrastructure is increasingly an energy question.