AI Pushes Data Centers to 6% of UK and US Power
Key insights
- Data centers now account for 6% of total electricity consumption in both the UK and US, a significant upward revision from prior estimates.
- Global data center energy demand rose 15% in two years, with AI identified as the primary driver of that acceleration.
- Hyperscaler AI capital expenditure is running more than 80% above prior-year levels, signaling demand growth has not yet plateaued.
Why this matters
Grid capacity is now a hard constraint on AI deployment timelines: if national grids in the UK and US are already at 6% utilization by data centers, permitting bottlenecks and transmission buildout will directly gate where and when new inference and training capacity can come online. For founders and technical leaders, this shifts the competitive calculus toward energy-efficient model architectures and geographies with surplus renewable capacity, since raw compute access will increasingly depend on power procurement deals rather than hardware availability alone. Regulators revising capacity projections in real time creates near-term policy risk for hyperscalers, including potential mandatory efficiency standards or interconnection queue delays that could affect capex-to-capacity conversion rates through 2027.
Summary
This piece lays out the concrete scale at which AI infrastructure has reshaped electricity demand in two of the world's largest economies.
New research finds data centers now consume 6% of total electricity supply in both the UK and the US, with AI identified as the primary cause of a 15% rise in global data center energy demand over just two years. That 15% figure is a significant upward revision from prior estimates, arriving precisely when hyperscaler capex (think Microsoft, Google, Amazon, Meta) is running more than 80% above prior-year levels. Grid operators and regulators in both countries are now working from capacity projections that were built on stale baselines.
Essentially: (Microsoft, Google, Amazon, Meta) are building faster than the grid was designed to absorb.
- Data centers have crossed 6% of national electricity supply in both the UK and US, a threshold that puts them in the same tier as major industrial sectors.
- The 15% global surge happened in roughly 24 months, compressing what planners expected to be a decade-long transition.
- Hyperscaler AI capex running 80%+ above prior-year levels means the demand curve has not yet peaked.
The energy constraint isn't a future risk to model; it's already a present-tense ceiling on how fast AI infrastructure can physically scale.
Potential risks and opportunities
Risks
- UK and US grid operators (National Grid ESO, PJM, MISO) face reliability risk if interconnection queues are not revised upward quickly enough to match hyperscaler build schedules through 2027.
- Microsoft, Google, and Amazon could see data center project delays of 12-24 months in constrained regions if regulators impose mandatory efficiency thresholds or pause new large-load interconnection agreements.
- Utilities in high-density AI build zones (Northern Virginia, London metro, Dublin) may face credit pressure if capital expenditure required to serve new load outpaces rate-case approvals, creating a financing gap that slows grid expansion.
Opportunities
- Advanced nuclear and long-duration storage developers (Kairos Power, Form Energy) gain direct commercial leverage as hyperscalers seek firm 24/7 low-carbon power contracts that intermittent renewables cannot provide alone.
- Liquid cooling and data center efficiency vendors (Vertiv, Schneider Electric, Asetek) are positioned for accelerated procurement cycles as operators face regulatory and cost pressure to reduce per-rack energy draw.
- Grid infrastructure software and demand-response platforms (AutoGrid, Leap, Voltus) can capture new enterprise contracts as large data center operators are pushed by regulators to participate in demand flexibility programs to ease peak load.
What we don't know yet
- Which specific grid regions within the UK and US are closest to capacity limits, and whether any have already triggered interconnection queue freezes for new data center projects.
- Whether the 15% global demand figure accounts for the wave of hyperscaler projects announced but not yet operational in 2025 and 2026, which would suggest the real peak is still ahead.
- How the revised consumption figures change the regulatory posture of Ofgem (UK) and FERC (US) toward data center permitting and whether either body has a formal review underway.
Originally reported by theguardian.com
Read the original article →Original headline: Data Centers Now Consume 6% of Electricity Supply in UK and US, Research Finds — AI Drives 15% Global Surge in Two Years