Megawatt AI Racks: Infrastructure Evolution & Wide-Bandgap Power Semiconductor Opportunities by 2030 - digitimes

The rapid advancement of artificial intelligence, particularly generative AI and large language models (LLMs), has fundamentally altered the landscape of data center power consumption. Traditional dat... Read original →

The surge in AI compute density is triggering a foundational overhaul of data center power delivery. With NVIDIA racks approaching 1 MW, legacy 12V/48V architectures collapse under thermal and efficiency constraints, accelerating adoption of SiC and GaN in AC/DC and point-of-load converters—silicon simply can’t sustain >98% efficiency at these power densities. This cascades upstream: 3nm EUV GPUs generate extreme localized heat, demanding power stages be co-packaged with compute dies, forging a new 'compute-power-thermal' co-design paradigm. Regulatory pressure from EU carbon tariffs and U.S. PUE mandates will penalize inefficient facilities, boosting wide-bandgap semiconductor deployment—but supply remains concentrated in Taiwan, China, and Japan, exposing geopolitical fragility. NVIDIA’s ecosystem (TSMC, onsemi) is securing SiC capacity, while AMD and Intel may leverage GaN expertise from consumer fast-charging into mid-power AI racks. Within 18 months, power architecture will become the decisive battleground: control over efficient energy delivery equals control over AI economics.