Building up the US power grid won't be wasted, even if the AI bubble bursts

US power companies must decide how much capacity to build for soaring datacenter demand, knowing that some planned facilities may never materialize and the AI bubble could deflate. Consulting firm McKinsey & Company says the greater near-term risk is building too little. Driven by the AI craze, a boom in datacenter building is expected to see the energy consumed by these facilities grow by 26 percent this year, as The Register reported not too long ago. McKinsey's own figures indicate that datacenters will account for about 75 percent of projected US power demand growth over the next decade. The current building rate would require the equivalent of almost 30 GW of additional power each year, including roughly 20 GW for IT equipment and the rest for cooling, distribution and resilience. Whether that growth materializes depends on demand for AI compute holding up. McKinsey notes that although corporate interest in AI remains high, implementation is uneven: 71 percent of organizations report "negative implementation outcomes," while parts of the ecosystem exhibit bubble-like characteristics. Even if AI compute growth slows, McKinsey argues, the generation and transmission infrastructure built to support it is unlikely to become stranded. The assets could serve other demand while strengthening grid reliability and resilience. A similar report from Bain & Company almost two years ago warned that the US energy sector needed to ramp up spending on generation and grid infrastructure, or demand could outstrip supply within a few years. Datacenters are not the only source of new demand. Electric vehicles and the electrification of industrial processes will also place pressure on supplies. The report estimates that the US has roughly 40 GW of spare dispatchable capacity to accommodate near-term demand growth, plus about 100 GW of committed new capacity. However, 50 to 75 GW of coal and gas-fired steam capacity is expected to retire, while projected demand growth stands at about 120 GW. That leaves a nationwide capacity gap of approximately 30 to 55 GW by 2030. To help bridge this looming gap and maintain grid reliability, utilities have been bringing retired plants back online and extending the lives of coal and gas facilities. These measures provide important temporary system relief but are likely insufficient, McKinsey says. However, in June, the Trump administration authorized up to $500 million in funding to keep an extra 13 coal-fired power plants going to boost the resilience of critical energy infrastructure. At the same time, the report notes that many of the developers and operators planning datacenter campuses are turning to on-site power generation via gas turbines, as well as fuel cells and battery storage, due to delays in getting a grid connection. In McKinsey's survey of power sector leaders, nearly 60 percent of respondents expected datacenters to retain permanent on-site generation by 2030, even after grid connections became available. Of those planning to have on-site power, 64 percent will rely on natural gas. Beyond 2030, the firm expects to see solar paired with battery storage in addition to the growing role of gas. This is because of the cost competitiveness and speed of deployment of solar, though it warns this may be dependent on policy support and continued declining costs of storage, and that it will be a part of the broader energy system rather than the sole on-site source of large-scale power. As for nuclear, McKinsey expects any material growth over the next decade to come from extensions and upgrades to existing plants rather than new builds. Emerging technologies such as small modular nuclear reactors (SMRs) and next-generation geothermal energy are attracting increasing capital and buyer interest, but they are unlikely to have much impact on system capacity before the middle of the 2030s. McKinsey's conclusion is that the US power sector faces a likely near-term shortfall, making underbuilding a greater risk than overbuilding. ®