AI Data Centers Shift to Prime Power as Grid Constraints Tighten, SemiAnalysis Says

SemiAnalysis said reciprocating engines, long used as backup power for data centers, are shifting from emergency reserve use to around-the-clock primary power to meet rising electricity demand from AI computing growth. According to Odaily, the firm said data centers have typically used on-site reciprocating engine backup power only for a few hours a year when the grid failed, but more facilities are now repurposing the equipment for Prime Power because of limited grid capacity.

Using its energy model, SemiAnalysis said manufacturers including Caterpillar, INNIO and Cummins have signed about 1 GW of behind-the-meter project orders this year. It expects related orders in 2027 and 2028 to each exceed 4 GW.

SemiAnalysis said those orders remain small compared with potential future demand. After modeling U.S. grid constraints, it found that existing spare power supply is expected to run out between 2027 and 2028, while planned utility-scale generation additions through 2030 are insufficient to meet growing data center load demand.

Based on its data center model, SemiAnalysis estimated that about 140 GW of identified data center projects have not yet signed power supply contracts and may use behind-the-meter power.

Among technologies available for behind-the-meter supply, including reciprocating engines, aeroderivative gas turbines and fuel cells, SemiAnalysis expects reciprocating engines to take the largest share. It said the equipment offers a balance between cost and deployment speed, along with modularity that supports phased data center expansion. It also said reciprocating engines have higher commercial maturity and financing feasibility than fuel cells.

SemiAnalysis said AI computing growth is pushing data centers from reliance on traditional grid power toward a hybrid model of grid power and self-supplied energy.