SemiAnalysis: Data Center Power Grid Bottlenecks Intensify, Backup Gas Generators Shifting to Round-the-Clock Power Supply

Odaily reports that SemiAnalysis posted on X platform, stating that traditional backup power equipment for data centers – Reciprocating Engines – are transitioning from an "emergency standby" role to a primary 24/7 power source, aiming to meet the massive electricity demands driven by the expansion of AI computing power.

SemiAnalysis pointed out that in the past, data centers typically only activated on-site reciprocating engine backup power during grid failures, operating for just a few hours each year. However, as grid capacity becomes constrained, an increasing number of data centers are repurposing these devices for Prime Power, enabling continuous electricity supply.

According to SemiAnalysis' energy model calculations, manufacturers of reciprocating engines, including Caterpillar, INNIO, and Cummins, are estimated to have signed around 1GW of Behind-the-Meter (BTM) project orders this year; the scale of related orders is expected to exceed 4GW in both 2027 and 2028.

SemiAnalysis indicates that the current scale of these orders remains very limited compared to potential future demand. After modeling U.S. grid constraints, they found that existing power supply margins are expected to be exhausted between 2027 and 2028, and utility-scale new generation plans through 2030 are insufficient to meet the growing load demands of data centers.

Based on its data center model, SemiAnalysis estimates that approximately 140GW of identified data center projects are yet to sign power supply contracts and may adopt a behind-the-meter power supply model.

Among the technologies available for behind-the-meter power supply, including reciprocating engines, aeroderivative gas turbines, and fuel cells, SemiAnalysis expects reciprocating engines to capture the largest share.

The institution believes that reciprocating engines offer the best balance between cost and deployment speed, along with modular advantages that support phased expansion of data centers; they also boast higher commercial maturity and financing feasibility compared to fuel cells. As manufacturers ramp up production capacity in the coming years, reciprocating engines are poised to become critical infrastructure for AI data centers to overcome power bottlenecks.

SemiAnalysis states that the growth of AI computing power is driving data centers to shift from relying solely on the traditional grid towards a hybrid model of "grid + self-generated energy," making energy infrastructure a new core constraint for the expansion of the AI industry.