AI boom could see US data centres consume one-fifth of nation's electricity by 2035, report warns

The race to build artificial intelligence infrastructure is expected to transform the United States' electricity landscape over the next decade, with data centres projected to account for around 20 per cent of the country's total power generation by 2035, according to a new report by BloombergNEF.

The forecast reflects the extraordinary pace at which AI computing capacity is expanding, as technology companies continue investing billions of dollars in larger facilities to train and operate increasingly sophisticated models. BloombergNEF estimates that total data centre capacity in the US will approach 200 gigawatts within the next 10 years, with AI workloads becoming the dominant driver of growth.

Nearly half of that capacity is expected to be dedicated to AI training and inference, underlining how generative AI has shifted from a niche computing task to one of the world's most energy-intensive digital industries. The consultancy also predicts that the US will remain the global centre of AI infrastructure, hosting around 64 per cent of AI chips by power demand by 2033.

More from Tech

The latest projections represent a dramatic upward revision from previous estimates. BloombergNEF said its outlook for electricity demand in 2035 is now 83 per cent higher than the forecast it published in December, illustrating how quickly expectations have changed as companies accelerate data centre construction.

Other research organisations have also substantially revised their estimates. The Electric Power Research Institute (EPRI), a nonprofit focused on the electricity sector, has more than doubled its projections made in 2024, while S&P has increased its forecast by over one-third between October and April. Together, the revisions point to an industry expanding much faster than analysts anticipated only months ago.

The surge in electricity demand is expected to place considerable strain on power networks that are already operating under pressure. BloombergNEF says most new facilities planned during the coming decade will connect to grids facing limited spare capacity, creating fresh challenges for utilities and grid operators.

One of the regions expected to face the greatest impact is the PJM Interconnection, which manages the electricity grid across parts of the eastern and midwestern United States, stretching from Virginia to Illinois. By 2035, BloombergNEF estimates that data centres could account for 34 per cent of electricity consumption across the network.

Texas is also expected to experience significant growth. Within ERCOT, the grid operator covering most of the state, data centres are projected to require around 22 per cent of generating capacity over the same period.

PJM has already struggled to accommodate the rapid rise in demand. The grid operator previously suspended applications for new electricity generation projects for four years while it worked through a growing backlog of connection requests. Although that queue reopened in April, the pressure has continued to mount as both electricity producers and large industrial consumers seek access to the network.

The imbalance between electricity supply and demand has already affected prices. BloombergNEF noted that wholesale electricity costs within the PJM market have climbed by 76 per cent over the past year. At the same time, demand from data centres remains robust, accounting for 38 per cent of bids in the grid operator's latest capacity auction.

While the United States is expected to remain the world's largest hub for AI computing, BloombergNEF believes the rapid expansion of data centres will extend well beyond North America. Under an aggressive AI adoption scenario, global data centres could generate an additional 1,935 terawatt-hours of electricity demand by 2033—almost equivalent to India's current annual electricity consumption—highlighting the scale of the energy challenge accompanying the AI revolution.