AI data centers could need 30 times more power by 2035, Deloitte warns

The electricity demands of AI data centers are set to explode over the next decade, surging from 4 gigawatts today to 123 gigawatts by 2035, new Deloitte research shows.

That thirtyfold increase won't happen without major hurdles. The report identifies seven critical gaps in AI infrastructure development that the firm says could constrain the technology's growth if not addressed.

The biggest challenge is timing. While data centers can be built in a year or two, gas power plant projects that haven't already contracted equipment aren't expected to become available until the 2030s, according to the report. Some regions are experiencing seven-year waits for new grid connections, while utilities ordering necessary grid technology today, such as combustion turbines, won't receive them until 2029.

While there is plenty of focus on what tech giants are spending on AI data centers this year, companies also need to think much further ahead, said Kelly Marchese, a Deloitte infrastructure leader and a co-author of the report. "When our clients are looking to make an investment, it's not a short-term horizon," she said. "These are major investments that require planning well beyond just a few years."

The long timeline becomes even more critical in light of other constraints. The industry also faces a severe labor crunch, with 63% of data center executives citing skilled worker shortages as a top challenge. Competition with other industries for workers is intensifying as companies race to build infrastructure, creating what the report calls a "top challenge" for the sector.

Perhaps most concerning is the lack of cooperation between data center operators and power companies. Only 15% of data center executives and 8% of power company executives describe their partnerships as "highly effective," despite 72% of both groups viewing power and grid capacity constraints as very or extremely challenging.

Supply chain bottlenecks are contributing to delays, with construction material costs having increased by 40% over the past five years. Critical components for power infrastructure are still subject to tariffs and import dependencies, potentially slowing the buildout further.

To overcome these challenges, massive investments are already planned. Electric and gas utilities are expected to spend more than $1 trillion over the next five years, while hyperscalers could hit that same trillion-dollar threshold in just three years. The tech industry is also planning to invest more than $1 trillion in U.S. manufacturing of AI supercomputers, chips, and servers over the next four years alone.

Meanwhile, data centers are already straining the power grid, with some facilities consuming as much electricity as hundreds of thousands of homes, and utilities charging millions just to study whether they can handle new connections. With such massive buildouts planned, the industry is scrambling to find efficiencies wherever possible.

The challenges don't stop there. Deloitte identified several other infrastructure bottlenecks complicating the AI buildout. Cybersecurity concerns are mounting as AI data centers become more vulnerable to supply chain attacks, with hackers potentially targeting everything from servers to cooling equipment.

The pace of development is outpacing regulators' ability to keep up, according to Marchese, with the report noting that regulatory delays are also slowing progress. Environmental impact statements, for instance, take more than two years to complete, while state-level restrictions on renewable projects have more than doubled over the past year.

Even natural gas infrastructure is strained, with many top data center markets facing constrained pipeline capacity despite plans for more than 99 gigawatts of new gas-fired power plants across 38 states.

It's not all bad news, though. The report suggests AI itself could help solve some of these infrastructure challenges. Eighty-three percent of respondents expect grid-enhancing technologies to play an increasing role in meeting data center energy demands through 2035. The technology could also enable more flexible data center operations — 68% of industry executives believe data center demand flexibility will become an acceptable tradeoff to secure faster grid connections.

Out of the crisis, though, something new may emerge. Industries that traditionally operated in silos are being forced to work together in unprecedented ways that could reshape how major infrastructure gets built in the future. "Utilities and hyperscalers are having to solve some of these problems together," Marchese said. "It could create a different type of innovation."