SemiAnalysis: Data center power bottleneck intensifies, backup gas generators are shifting to all-weather power supply Odaily Planet Daily News: SemiAnalysis has published an article on the X platform stating that traditional backup power equipment in data centers - reciprocating engines - are shifting from their role as "emergency backup" to being the main power source 24/7 to meet the huge electricity demand brought about by the expansion of AI computing power. SemiAnalysis points out that in the past, data centers typically only activated on-site reciprocating engine backup power during grid failures, with an annual operating time of only a few hours. But as the capacity of the power grid is limited, more and more data centers are starting to reuse these devices for Prime Power to achieve continuous power supply. According to SemiAnalysis's energy model, reciprocating engine manufacturers including Caterpillar, INNIO, and Cummins are expected to have signed approximately 1GW of Behind the Meter (BTM) project orders this year; The expected order sizes for 2027 and 2028 are both expected to exceed 4GW. SemiAnalysis states that the current order sizes are still very limited compared to potential future demand. After modeling the constraints of the US power grid, it was found that the existing power supply surplus is expected to be depleted between 2027 and 2028, and the new power generation plans at the utility level by 2030 are insufficient to meet the growing demand for data center loads. Based on its data center model, SemiAnalysis estimates that there will still be approximately 140GW of identified data center projects in the future that have not yet signed power supply contracts and may adopt a post table power supply mode. SemiAnalysis expects reciprocating engines to account for the largest share of technologies available for power supply behind the meter, including reciprocating engines, aeroderived gas turbines, and fuel cells. Institutions believe that reciprocating engines have the best balance between cost and deployment speed, as well as modular advantages, and can support phased expansion of data centers; Compared to fuel cells, their commercial maturity and financing feasibility are higher. As manufacturers expand their production capacity in the coming years, reciprocating engines are expected to become an important infrastructure for AI data centers to solve power bottlenecks. SemiAnalysis states that the growth of AI computing power is driving data centers to shift from relying on traditional grid power to a hybrid model of "grid+self owned energy", and energy infrastructure is becoming a new core constraint for the expansion of the AI industry.