AI Data Centers Will Need a Lot More Power Going Forward

As AI data centers rapidly multiply, their electricity needs are set to soar, posing major challenges for power grids, industry partnerships, and infrastructure investment, according to a new Deloitte report cited by Quartz. The firm warns that the electricity demand from AI data centers is expected to jump dramatically—from 4 gigawatts last year to 123 gigawatts by 2035, a thirtyfold increase.

• Roadblocks: The report outlines several obstacles that could slow this growth, with timing being one of the most critical issues. While data centers can be constructed in a year or two, the process for bringing new power plants online can take much longer, with some regions already facing lengthy waits for new grid connections. Skilled labor shortages pose another significant constraint, with nearly two-thirds of data center execs reporting trouble finding qualified workers. Supply chain bottlenecks and rising construction material costs are also creating delays, while tariffs and import dependencies further complicate matters.

• Roadblocks II: A lack of effective cooperation between data center operators and power companies is highlighted in the report as well, with only a small percentage of executives from both sides (15% of data center execs, 8% of power company execs) describing their partnerships as highly effective, despite widespread agreement (72% of both groups) that grid capacity is a serious challenge. Meanwhile, cybersecurity threats and regulatory holdups are adding to the complexity, as are environmental reviews and restrictions on renewable energy projects.
• Investment: Despite these hurdles, companies are planning major money outlays. Utilities are expected to spend more than $1 trillion over the next five years, with "hyperscalers" (i.e., large-scale tech companies) likely to match that figure in just three years. The tech industry is also slated to invest more than $1 trillion in US manufacturing for AI hardware over the next four years.
• Key take: "Collaboration will be critical, as will an understanding of the promise of additive infrastructure [that] builds upon the existing resources to bring efficiency, capacity, and flexibility to power AI," says Martin Stansbury, principal for Deloitte & Touche, in a release.