SemiAnalysis: Động cơ piston có thể trở thành giải pháp nguồn điện sau đồng hồ quan trọng cho các trung tâm dữ liệu AI

The Huoxing Finance report states that on July 21, the semiconductor and AI independent research firm SemiAnalysis published an article indicating that as demand for AI computing power grows rapidly, the power supply model for data centers is undergoing a transformation. Reciprocating engines, previously used primarily as backup power during grid outages, are gradually being repositioned as base-load power sources operating around the clock. Reciprocating engine manufacturers have already signed contracts to supply approximately 1GW of behind-the-meter power projects this year, with projected annual新增 supply capacity exceeding 4GW in both 2027 and 2028. After modeling the U.S. grid’s capacity, SemiAnalysis noted that existing power reserves are expected to be fully consumed between 2027 and 2028, and the planned新增 utility-scale generation capacity through 2030 remains insufficient to meet the additional load demands from data centers. Combining its data center model, SemiAnalysis estimates that approximately 140GW of potential data center projects have not yet signed power supply contracts; these projects may increasingly adopt behind-the-meter power solutions to circumvent grid expansion bottlenecks. Among various behind-the-meter power technologies—including reciprocating engines, aviation-derived gas turbines, and fuel cells—SemiAnalysis expects reciprocating engines to capture the largest market share due to their advantages in lower cost, rapid deployment, modular scalability, and stronger financing capabilities. Meanwhile, equipment manufacturers such as Caterpillar, INNIO, and Cummins are expanding production capacity to support large-scale deployment in the coming years. As AI data centers continue to expand, on-site self-supplied power is evolving from a traditional backup facility into a critical energy infrastructure, with reciprocating engines poised to become a key solution for bridging the power gap in the age of computing power.