AI Infrastructure
How Much Electricity Will AI Require?
Don't just read what happened. See what could happen next.
Prediction
AI becomes a material driver of incremental electricity demand in major grids over the next five years, with local hotspots already binding.
- Confidence
- 72%
- Horizon
- 24–60 months
- Impact
- High
- Direction
- Increasing
Prediction changed +6 points in 21 days
Short answer
Past-year utility filings and hyperscaler PPAs show AI load as a planning input, not a footnote. Global totals remain a slice of all electricity, but regional peaks and interconnection queues are the real story.
Why this question matters
Power availability may throttle model growth as much as chip supply — and will reshape energy markets and climate politics.
What's happening now?
Utility load forecast revisions
Grid planners raise peak demand expectations near data-center hubs.
Signal · strong
Corporate power purchase deals
AI firms lock long-term clean and firm power.
Signal · strong
Interconnection backlogs
Projects wait years for grid hookups.
Signal · strong
Efficiency and cooling innovation
Liquid cooling and better chips partially offset growth.
Signal · moderate
What could happen next?
Scenario A
Power-constrained AI
Energy becomes the binding limit on cluster growth in key regions.
Scenario B
Buildout matches demand
Generation and transmission expand fast enough to keep pace.
Scenario C
Efficiency surprise
Algorithmic and hardware gains flatten AI power curves.
Key companies / entities
- Microsoft
- Amazon
- NextEra
- PJM
Evidence
- research
IEA / national AI electricity analyses
- policy
Utility integrated resource plans
- company
Hyperscaler energy deal announcements
Prediction history
- Sep 22, 202672%
- Sep 15, 202670%
- Sep 8, 202668%
- Sep 1, 202666%