AI Data Centers Shift to 800 VDC Power, Tightening Supply of SiC and GaN Components

NVIDIA’s move toward 800-volt direct current power architecture for next-generation AI data centers is creating new demand across the power semiconductor supply chain, particularly for SiC and GaN components, as infrastructure moves toward higher power density systems.

The architecture, designed to support megawatt-scale AI racks expected later this decade, is part of NVIDIA’s broader Kyber rack platform targeted for 2027 deployment, according to industry disclosures.

This shows the structural change in how AI data centers are powered, moving away from traditional low-voltage distribution toward high-voltage systems that reduce energy loss and enable higher compute density.

Power infrastructure becomes a key constraint in AI expansion

AI data centers are rapidly increasing in size and power demand, with some facilities requiring hundreds of megawatts of electrical capacity. This has put pressure not only on chip supply but also on power delivery systems, including transformers, switchgear, and conversion equipment.

Industry data shows long lead times for high-voltage electrical infrastructure, with some grid components taking multiple years to deliver. These delays are increasingly seen as a limiting factor in new data center construction.

The shift to 800 VDC is intended to reduce complexity in power distribution and improve efficiency, but it also introduces demand for a new set of semiconductor and electrical components.

The 800 VDC model requires multiple conversion stages, each dependent on different semiconductor technologies.

Silicon carbide devices are used in high-voltage stages that connect data centers to the electrical grid, while gallium nitride devices are used in high-frequency conversion closer to computing hardware.

Additional demand is emerging for power management integrated circuits (PMICs), protection devices, hot-swap controllers, and high-voltage control components used across server systems.

Several semiconductor suppliers, including Infineon, Texas Instruments, STMicroelectronics, onsemi, and Navitas Semiconductor, are participating in development of components for high-voltage AI data center power systems.

Supply constraints emerging in multiple component categories

Supply chain conditions for power semiconductors are tightening as demand from AI infrastructure grows alongside existing demand from electric vehicles and industrial applications.

Industry estimates indicate increasing adoption of SiC and GaN devices in data center power systems, with usage expected to rise steadily through the decade.

At the same time, lead times for power management ICs and related components have extended in some cases beyond 30 weeks, reflecting constrained capacity at mature-node semiconductor fabs.

These components, while less visible than AI processors, are required for nearly every server and rack system, meaning shortages can delay full system deployment even when chips are available.

Solid-state transformers and high-voltage systems gain traction

The transition to higher-voltage systems is also driving development of solid-state transformers, which replace traditional grid transformers with semiconductor-based conversion systems.

Companies including Microchip, Navitas, and Delta Electronics have demonstrated high-voltage power modules and transformer systems designed for AI data center applications.

These systems aim to simplify power conversion from the electrical grid to server racks while improving efficiency and reducing physical footprint.

However, these technologies remain in early deployment stages and will require years of qualification before widespread adoption.

Outlook

There is an expected demand for SiC and GaN components to grow steadily through the remainder of the decade as AI infrastructure expands. However, supply remains constrained in several areas, particularly for high-voltage devices, power management ICs, and specialized conversion components.

While semiconductor manufacturers are expanding capacity, lead times for key power components are expected to remain extended due to both structural demand growth and long manufacturing lead times for new fabrication facilities.

The transition to 800 VDC is therefore likely to add a new layer of supply chain pressure across the AI infrastructure ecosystem rather than replace existing constraints.

As a result, procurement attention is expanding beyond GPUs and memory toward power semiconductors, conversion systems, and grid-level infrastructure components that enable large-scale AI deployments.