No matter how many GPUs you have, you might still lack power? JPMorgan: $19.2 Billion Power Chips Are the Real Key - odaily.news

J.P. Morgan applies first-principles reasoning to analyze the complete power supply chain for AI data centers. The core finding is that the AI power semiconductor market will grow from $2.7 billion in... Read original →

The shift to 800V HVDC architecture is triggering a silent revolution in AI data centers—power semiconductors, not GPUs, are now the true bottleneck. Technically, SiC and GaN adoption forces redesigns across PDUs, VRMs, and UPS systems, cascading into advanced packaging and thermal demands beyond EUV. Regulatory pressure from EU energy efficiency mandates and U.S. carbon-adjusted tariffs will penalize legacy AC infrastructure, accelerating architectural shifts. In the competitive arena, Infineon and Wolfspeed leverage substrate scale, while MPS and Navitas target edge AI with integrated GaN—but NVIDIA’s potential bundling of power specs via Grace Hopper could redefine supply chain control. Over the next 18 months, grid interconnection delays and fragmented HVDC standards will create regional chip shortages, making power semiconductor lead times—not GPU counts—the real constraint on AI deployment.