SemiAnalysis: AI Data Center Power Shortage Widens, Reciprocating Engine May Become the Mainstream After Formalization Power Supply Technology

BlockBeats News, July 21st – Semiconductor and AI independent research firm SemiAnalysis published an article stating that as the demand for AI computing power is rapidly increasing, the data center power supply mode is undergoing a transformation. The reciprocating engine, which was previously mainly used as a backup power supply during grid outages, is now being repositioned as a reliable primary power source for continuous operation. Reciprocating engine manufacturers have already signed contracts this year to supply around 1GW of behind-the-meter power projects, with the annual additional supply expected to exceed 4GW by 2027 and 2028.

After modeling the U.S. grid's carrying capacity, SemiAnalysis pointed out that the existing power surplus is expected to be depleted by 2027 to 2028, and the planned additional utility-scale generating capacity by 2030 is still insufficient to cover the new load demands from data centers. Combining this with their data center model, SemiAnalysis estimates that there are approximately 140GW of potential data center projects that have not yet secured power supply contracts. These projects may heavily adopt behind-the-meter power supply models in the future to bypass the grid expansion bottleneck.

Among various behind-the-meter power technologies, including reciprocating engines, aviation-derived gas turbines, and fuel cells, SemiAnalysis projects that reciprocating engines will occupy the largest share. This is due to their lower costs, rapid deployment, modular scalability, and stronger financing capabilities. Equipment manufacturers such as Caterpillar, INNIO, and Cummins are expanding their capacities to support large-scale deployments in the coming years. As AI data centers continue to expand, on-site self-generation is transitioning from traditional backup facilities to critical energy infrastructure, with reciprocating engines poised to become a key solution to fill the power gap in the era of computing power.