Power requirements for AI server clusters are rapidly heading toward 1MW per rack, triggering an intense race for silicon carbide and gallium nitride components. Hardware manufacturers Delta Electronics and Lite On are securing partnerships with semiconductor foundries to roll out 800VDC platforms and solid state transformers. But managing heat losses across multi stage power delivery will separate the market winners from the rest.
Modern AI computing hardware pulls massive amounts of electricity. Speaking at the SEMICON industry conference, Infineon Technologies operations chief Alexander Gorski noted that advanced server racks already draw hundreds of kilowatts. That number is set to reach 1MW soon. Power travels a long path from the electrical grid through local transformers, energy storage banks, and server shelves before reaching chip cores. Every single stage of voltage conversion bleeds power. Trimming conversion losses across that entire path is now essential to keep facilities running within utility limits.
Power semiconductor production moves slowly. Delta Electronics chairman Ping Cheng explained that building production lines for silicon carbide and gallium nitride requires long capital cycles, creating unpredictable supply bottlenecks. Delta manages this risk by locking in development partnerships with large component fabricators long before products enter mass assembly. This strategy allows engineering teams to build new power supplies around next generation materials before competitors can secure chip allocations.
Lite On is taking a similar route by working directly with Wolfspeed. The manufacturer is integrating silicon carbide MOSFET units into its upcoming 800VDC Sidecar power architecture. This platform is built for hyperscale cloud operators that need to reduce power distribution losses inside dense compute clusters. Tying semiconductor supply deals directly to equipment build schedules helps vendors guarantee delivery windows for large corporate buyers.
The next major step involves cutting out intermediate conversion equipment entirely. Traditional oil filled magnetic transformers often come with order delays stretching across multiple years. Infineon pointed to solid state transformers as a compact replacement that reduces equipment footprints and cuts power conversion losses. Delta chief operating officer Victor Chang highlighted that solid state systems provide active control to handle the sharp spikes in power draw typical of AI processing loads. Chang expects solid state power delivery to see widespread commercial adoption as early as 2027.
