- The IEA projects global data center electricity demand will more than double from 415 TWh in 2024 to around 945 TWh by 2030, driven almost entirely by AI workloads.
- Grid connection delays now exceed three years in major U.S. cluster zones, pushing hyperscalers toward on-site generation: Google’s $20 billion Intersect Power partnership and Microsoft’s Three Mile Island restart are the clearest examples.
- Climate change is cutting Columbia River hydropower output precisely when summer demand peaks, exposing a reliability risk for Pacific Northwest data center buildout that on-site renewables alone cannot resolve.
Grid connection delays, not capital or compute, are the binding constraint on AI infrastructure buildout. The International Energy Agency projects global data center electricity consumption will more than double from 415 TWh in 2024 to around 945 TWh by 2030, driven almost entirely by AI workloads. Hyperscalers are responding by bypassing the grid entirely and building their own power.
Power Costs Redraw the Map
Local grids near major U.S. data center clusters have struggled to keep pace, with wholesale electricity costs in some affected areas rising sharply in recent years, according to industry analysts. Where a hyperscaler can plug in, and at what cost, now determines where it builds.
The Nordic Advantage
Iceland runs on roughly 70% hydropower and around 30% geothermal. Wind power in Sweden fell to around $0.03/kWh in 2024, putting Nordic electricity costs well below Western European rates. That arithmetic is hard to ignore at data center scale.
OpenAI announced its first European data center in Narvik, Norway, skipping the traditional hubs of Frankfurt, Dublin and Amsterdam. Google pledged an additional €1 billion to expand its Finnish campus. For more on how hyperscalers are controlling the full stack from silicon to power sourcing, see our coverage of custom chips and liquid cooling.
Pacific Northwest: A Reliability Risk
Climate change is cutting into mountain snowpack, with projected summer flow reductions in the Columbia River reducing available hydropower during the months demand peaks. Washington and Oregon generated measurably less hydropower in 2023 than in 2021. That is a direct reliability risk for a region that has long sold itself as a clean-power data center destination.
Building Their Own Power
Three-year interconnection queues have pushed the industry toward a different model: own the generation. Rather than waiting for utility connections, major operators are co-locating data centers with dedicated power infrastructure on the same land.
Google’s relationship with Intersect Power, which began as a $20 billion co-location partnership in December 2024, has since deepened into a full acquisition: Alphabet bought the company outright for $4.75 billion in a deal that closed March 10, 2026, giving Google direct control over the generation pipeline built for its data centers. Microsoft contracted 389 MW from solar installations in Illinois and Texas in February 2025 and has invested $1.6 billion in restarting the Three Mile Island nuclear plant for stable baseload power. Amazon maintains 13.6 GW of solar projects in development across the United States. A joint initiative backed by Amazon, Google, Meta and Microsoft plans to invest $500,000 to $5 million per project in startups working on energy storage, advanced electrical systems and industrial cooling through the end of 2027. The scale of that commitment reflects how seriously the industry is treating the supply problem, and how far the grid alone is from solving it. The infrastructure demands of next-generation AI hardware are pushing that urgency further.
The Clean Energy Gap
The IEA projects global data center electricity consumption will reach around 945 TWh by 2030, roughly 3% of global electricity use. Renewables are expected to cover a portion of the additional demand, but natural gas and coal will supply the remainder, with the split depending heavily on how fast clean capacity can be built. Next-generation geothermal and nuclear are increasingly cited as the firm, 24/7 carbon-free sources that variable renewables cannot reliably provide alone.
The geographic concentration of AI compute compounds the grid challenge. Unlike electric vehicles, which spread demand across a wide area, a single large data center campus can stress a regional grid in ways that require targeted infrastructure upgrades. Utilities are beginning to apply AI-enabled monitoring and load-balancing tools to manage that pressure, but the scale of buildout needed over the next five years means the gap between AI power demand and clean supply will widen before it narrows.
