Grid Constraints Lift Focus on Behind-the-Meter Power for Data Centers

Bloom Energy has emerged as a market beneficiary of the data-center industry’s search for faster access to power amid constrained grid interconnections and long equipment lead times. Its shares rose from $157 on July 29 to a recent high of $292.72 on September 25, a gain described by the source as 86%.

The underlying issue is the widening mismatch between data-center electricity demand and the pace of grid expansion. Goldman Sachs forecasts that global data-center electricity demand will increase by 170% by 2030 relative to 2025, with the United States accounting for more than 60% of incremental demand growth. In the U.S., annual additions of high-voltage transmission lines fell from 1,700 miles in 2010–2014 to 350 miles in 2020–2023. The median interval from an interconnection application to commercial operation for the projects cited in the report is close to five years.

Against that backdrop, Goldman Sachs increased its forecast for global data-center behind-the-meter generation capacity in 2030 from 40 GW to 67 GW. The bank estimates that such generation could supply about 25% of global data-center power demand by that point. The forecast concerns potential future capacity, rather than commissioned or contracted projects.

The economics remain a constraint. In a Goldman Sachs model for a 500 MW data center, fuel-cell lifecycle LCOE is estimated at about $117/MWh, compared with $80/MWh for reciprocating engines and $81/MWh for combined-cycle gas turbines. Even assuming a 30% investment tax credit, the model puts fuel-cell LCOE at about $90/MWh.

This suggests that data-center operators may increasingly accept a higher cost of power where local generation can reduce exposure to grid-connection delays. If those constraints persist, demand could move toward power solutions with shorter delivery and commissioning timelines, expanding the addressable market for suppliers able to deploy onsite or nearby generation quickly. That remains an inference from forecast demand and modeled economics, not evidence of installed capacity or specific customer deployments.