Microchip's 3.3 kV SiC modules target AI data centre power crunch
Microchip Technology has launched a family of 3.3 kilovolt silicon carbide (SiC) power modules purpose-built for solid-state transformers (SSTs) in AI hyperscale data centres — a product category that sits at the precise junction where semiconductor physics, grid infrastructure, and the economics of large-scale AI compute collide.
The HV-D3 mSiC modules integrate SiC MOSFETs and Schottky barrier diodes in a standard 62 mm package, allowing power engineers to draw directly from medium-voltage grids — typically 13.8 kV or 34.5 kV distribution lines — and deliver regulated DC to the server rack with fewer intermediate conversion stages than conventional low-frequency transformer architectures. The company says the 3.3 kV rating allows designers to roughly halve the number of series-connected devices compared with lower-voltage SiC alternatives when interfacing with those grid voltages.
The power wall behind AI scaling
The timing reflects a structural tension that is increasingly visible on hyperscaler earnings calls: GPU clusters can only generate tokens as fast as the grid can feed them. Traditional iron-core transformers are bulky, thermally inefficient, and ill-suited to the high-frequency switching that modern rack-density targets demand. Solid-state transformers — power-electronic equivalents that replace magnetic cores with high-speed switching circuits — promise higher efficiency, smaller footprints, and the ability to deliver regulated high-voltage DC directly to next-generation racks without multiple lossy conversion hops.
Microchip's modules use a silicon nitride substrate, which the company says delivers better thermal conductivity and power-cycling endurance than conventional aluminium-oxide alternatives, supporting higher power density without proportionally more aggressive cooling. The packaging is rated for 6 kV isolation with CTI 600-rated materials and extended creepage distances, designed to allow safe series stacking. The product line covers the 100–300 A range — a segment the company describes as a gap between discrete SiC devices and the much larger power modules already available for utility-scale applications.
"As AI data centres continue to push limits in supplying power from the grid to the GPU, the need for solid-state transformers becomes increasingly important," said Clayton Pillion, vice president of Microchip's high-power solutions business unit. "Our 3.3 kV HV-D3 mSiC power modules enable designers to reduce the number of series connected devices by roughly half versus lower-voltage SiC alternatives when interfacing to 13.8 kV or 34.5 kV grids."
Convergence: chips, grid, and the AI capex supercycle
For cross-sector strategists, the more consequential read-across is what this product category signals about where semiconductor value is migrating. The AI compute build-out — dominated in public discussion by GPU procurement and hyperscaler capex — is quietly generating a parallel investment wave in power semiconductors, grid interconnection hardware, and thermal management. Analysts tracking the space note that SiC-based power components are one of the faster-growing segments of the broader semiconductor market, driven simultaneously by EV charging infrastructure, industrial motor drives, and now AI data centre power delivery. Microchip is not alone: Wolfspeed, onsemi, and STMicroelectronics are all competing in the high-voltage SiC module space, meaning the real contest is moving to yield, cost per watt, and the depth of reference designs that reduce customers' time-to-deployment.
For investors already allocated to AI infrastructure, the signal here is that the power electronics stack — long treated as a commodity procurement decision — is becoming a strategic bottleneck in its own right. Data centre operators facing grid-connection queues in Northern Virginia, London, and the GCC are under pressure to extract maximum efficiency from every megawatt of contracted capacity. SST adoption, if it scales, directly improves that equation. Microchip's move to productise 3.3 kV modules for production quantities — rather than sampling only — suggests commercial deployment timelines are compressing. The secondary beneficiaries include grid operators, cooling-system vendors, and the sovereign and institutional investors now underwriting gigawatt-scale data centre campuses who will need to stress-test their power assumptions against a rapidly changing hardware landscape.