McKinsey Warns of US Power Shortfall Amid AI Boom

US power companies are weighing how much capacity to build to meet soaring datacenter demand, facing the possibility that some facilities may never be built or that the AI bubble could deflate, as first reported by The Register.

Consulting firm McKinsey & Company asserts that the primary near-term risk is building too little. According to McKinsey's figures, datacenters are expected to account for about 75 percent of projected US power demand growth over the coming decade. Current construction rates would require nearly 30 GW of additional power annually, with roughly 20 GW dedicated to IT equipment and the remainder for resilience, distribution, and cooling.

While McKinsey notes that 71 percent of organizations report "negative implementation outcomes" regarding AI, the firm argues that infrastructure built for AI is unlikely to become stranded. These assets could strengthen grid resilience and serve other demand sources, such as electric vehicles and industrial electrification.

According to McKinsey, the US currently has about 40 GW of spare dispatchable capacity and 100 GW of committed new capacity. However, projected demand growth is approximately 120 GW, while 50 to 75 GW of gas and coal-fired steam capacity is expected to retire. This creates a projected nationwide capacity gap of 30 to 55 GW by 2030. A similar report from Bain & Company nearly two years ago also warned that failure to ramp up spending could lead to demand outstripping supply.

To address the gap, utilities are extending the life of coal and gas plants. Additionally, the Trump administration authorized up to $500 million in June to keep 13 coal-fired plants operational. Many developers are also opting for on-site generation; a McKinsey survey found nearly 60 percent of power sector leaders expect datacenters to maintain permanent on-site generation by 2030, with 64 percent of those relying on natural gas.