By 2035, data centers will require twice the amount of electricity currently used by California

The rapid expansion of artificial intelligence data centers is intensifying, with new research projecting that the technology sector will need 106 gigawatts of electricity by 2035. This surge in demand raises questions about whether power utilities and regional grid operators can supply sufficient energy to meet the growing needs.

A recent report from BloombergNEF reveals a 36% increase in electricity requirements for both existing and planned data centers compared to their April analysis. The company added 150 new facilities to its tracker, noting that over a quarter of them are massive structures requiring more than 500 megawatts to operate.

Meeting this demand is challenging because producing more electricity necessitates substantial investment and time to construct power plants, solar arrays, and battery storage systems. Already, supply and demand dynamics are pushing electricity prices higher in regions experiencing the fastest data center growth.

To put it in perspective, a watt measures the instantaneous electricity a device or facility consumes, while megawatts and gigawatts represent thousands and millions of watts, respectively. Currently, data centers use roughly 40 gigawatts, meaning an additional 60 gigawatts will be needed over the next decade to keep pace with AI growth.

The peak demand for California's grid hit 52 gigawatts in 2022. If the BloombergNEF forecast proves accurate, AI data centers alone will require nearly double that amount by 2035, equating to the output of several full-scale power plants just to sustain these facilities.

BloombergNEF highlights that this surge could mark a critical moment for U.S. power grids, as they struggle to support AI-driven electricity loads without compromising reliability or driving up costs. PJM, a major regional grid covering areas from Virginia to Chicago, recently warned federal regulators that the influx of data centers could threaten grid stability and increase blackout risks.

According to Monitoring Analytics, an independent grid monitor, PJM's current proposals could allow new data centers to connect in ways that cannot be reliably supported, potentially resulting in curtailments or outages for these facilities or other customers. They urged the Federal Energy Regulatory Commission to ensure that any new data centers can be reliably served.

BloombergNEF's projections indicate that new data centers on the PJM grid could outpace electricity generated from new power plants or renewable sources forecasted by the U.S. Energy Information Administration. In Texas, the Electric Reliability Council (ERCOT) could face challenges as early as 2028, when its reserve power margin may drop into a risky range, signaling that while short-term growth is manageable, long-term supply may lag behind demand.

Analysis: AI Data Centers and the Strain on U.S. Power Grids

The rapid rise of artificial intelligence data centers is creating a significant challenge for U.S. power infrastructure. Current projections indicate that electricity demand from these facilities could reach 106 gigawatts by 2035, more than doubling the capacity currently used by data centers. This surge highlights the urgency for utilities and regional grid operators to plan and invest in additional generation and storage capacity.

BloombergNEF's recent report shows a 36% increase in electricity requirements compared to earlier estimates, reflecting both the addition of 150 new facilities and the scale of some operations exceeding 500 megawatts. Meeting this demand will require substantial new energy projects, from power plants to solar and battery systems, which take years to develop.

The implications are immediate for regions like California and PJM, where grid operators warn of potential instability and blackout risks if growth outpaces infrastructure. PJM has stressed that without careful oversight, new connections could lead to curtailments or outages for both data centers and other customers. ERCOT in Texas may face similar constraints as early as 2028 if reserve margins decline.

These developments underline a critical intersection between AI expansion and energy policy. Ensuring reliable power supply while accommodating rapidly growing AI loads will require coordinated planning, investment, and regulatory oversight to prevent disruptions and manage costs effectively.

Sources:

• Data centers projected to need double California's current power by 2035