en.Wedoany.com Reported - The next bottleneck for AI development may not be chip computing power, but electricity supply. For cloud service providers, colocation operators, and enterprise AI buyers, the challenge has shifted from acquiring accelerators or cloud capacity to whether utility companies can provide stable power for planned AI campuses.

In 2023, U.S. data centers consumed approximately 4.4% of the nation's electricity, and by 2028, this share could rise to between 6.7% and 12%. A report supported by the U.S. Department of Energy (DOE) projects that data center electricity consumption could climb from about 176 terawatt-hours in 2023 to between 325 and 580 terawatt-hours in 2028, with the high-end scenario nearly tripling the sector's electricity share. Against this backdrop, recent AI computing power deals have demonstrated that power, cooling, networking, and data center space must be secured simultaneously.

National figures may obscure more pressing local issues. The Secretary of Energy Advisory Board (SEAB) notes that hyperscale connection requests, typically ranging from 300 to 1,000 megawatts or more with lead times of one to three years, are straining local grids, while the permitting, financing, and construction of transmission infrastructure often take much longer. Grid access risks are particularly pronounced in transmission-constrained areas. SEAB cites an example where a utility company can supply power to data centers for about 350 days per year, relying on backup generation, energy storage, or other reliability solutions for the remaining 15 days, failing to fully meet the timelines of AI deployment roadmaps.

Load flexibility is a near-term option. SEAB states that when latency is not critical, some AI inference workloads can be routed between different regions based on local grid load and renewable energy availability, thereby shifting some pressure during peak demand periods. However, flexibility cannot eliminate the need for stable power. SEAB notes that stakeholders generally view new natural gas capacity, along with solar, wind, and battery storage, as the primary options for maintaining reliability.

Nuclear power is emerging as part of a long-term hedging strategy. The Associated Press reports that Constellation Energy plans to bring the 835-megawatt reactor at Three Mile Island back online by 2027, a plan linked to a 20-year power supply agreement with Microsoft. Google, Kairos Power, and the Tennessee Valley Authority have indicated that Hermes 2 in Oak Ridge is scheduled to begin operations in 2030, supplying up to 50 megawatts of power to the TVA grid. However, neither project addresses the near-term grid gaps facing current AI campuses. Regulators are also weighing who should bear the costs of grid upgrades associated with new campuses. SEAB recommends tariffs that require data centers to share or cover these costs, warning that if domestic power timelines remain unfeasible, some companies are considering sites outside the United States.

For enterprise AI teams, power availability has now become a critical factor in infrastructure planning, alongside model performance, accelerator supply, and workload placement.

This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com