Overview

Infineon Technologies AG is expanding its collaboration with NVIDIA on 800 V direct-current (VDC) power architectures for artificial intelligence (AI) data centers. The latest phase focuses on high-voltage intermediate bus converters (IBCs), a critical stage in the power delivery chain that steps down the main distribution voltage to levels suitable for downstream conversion and processor power delivery. The work builds on earlier announcements in May 2025 and October 2025 that addressed serviceable, hot-swappable server racks operating with 800 VDC power distribution. The companies are targeting higher power density and efficiency as AI racks demand increasingly higher power levels.

Why 800 VDC and Why the IBC Matter

As AI workloads drive power consumption higher, data centers are exploring 800 VDC distribution alongside existing AC and lower-voltage DC architectures. Higher distribution voltages can reduce current and the associated resistive losses, which becomes increasingly important as rack power levels approach the megawatt range. The IBC does not generate the final core or point-of-load voltages used by AI processors. Instead, it converts the main bus voltage into intermediate rails that subsequent stages use to produce the precise, low-voltage supplies required by processors. That makes the IBC an architectural adapter between the high-voltage distribution backbone and the fine-grained conversion near the compute silicon.

GaN Technology at the IBC Stage

Gallium nitride (GaN) power devices offer faster switching and lower conduction and switching losses than conventional silicon technology. At the IBC stage, switching frequencies up to 1 MHz enable smaller magnetic components and passive devices, reducing converter size and board height. These characteristics are particularly important as rack power levels approach the megawatt range and space, thermal management and transient response become increasingly demanding. Higher switching frequency can shrink the energy-storage components that dominate converter volume, but it also places greater demands on layout, gate drive, thermal design and electromagnetic compatibility. The value of GaN in this application therefore depends on more than the device itself; it requires a co-designed power stage, controller and driver.

Integrated Controller and Driver Solutions

Infineon combines its GaN portfolio with PSoC Control C3 controllers and EiceDRIVER gate-driver products to provide integrated IBC solutions. The company supports conversion ratios including 800 V to 50 V, 800 V to 12 V, and 800 V to 6 V, allowing designers to select intermediate bus architectures based on system requirements. The availability of multiple conversion ratios is significant because different downstream topologies and processor power delivery schemes may favor different intermediate rails. Integrated control and gate-drive functions are especially relevant for high-frequency GaN conversion, where timing, protection and signal integrity strongly influence achievable efficiency and reliability.

A Multi-Technology Portfolio

Infineon offers intermediate bus conversion solutions based on silicon, GaN and silicon carbide (SiC), covering different performance and design requirements. Its positioning combines power semiconductor technology with in-house manufacturing, application engineering and system-level expertise. This portfolio approach is important in a market where 800 VDC adoption is still evolving alongside existing AC and lower-voltage DC architectures. Designers may need to balance conversion efficiency, power density, thermal constraints, cost and supply-chain considerations. GaN is therefore one option within a broader power semiconductor toolset rather than a universal replacement for every stage of the conversion chain.

Power-Path Protection and Serviceability

The collaboration with NVIDIA also extends beyond intermediate bus conversion. It focuses on power-path protection, including hot-swap controller functionality intended to support safe servicing and the deployment of future server boards in 800 VDC architectures. Serviceability is a central requirement for data center power design because server boards and power components must be maintained or replaced without compromising the safety and availability of the rack. At higher system voltages, protection and hot-swap control become more critical parts of the architecture, not secondary accessories. The fact that the collaboration addresses both conversion and protection suggests a system-level approach to 800 VDC rack design.

Implications for AI Data Centers

Together, these developments target power delivery across the data center, from grid-facing power supply units through rack-level distribution and intermediate conversion to the processor’s point of load. The objective is to meet the growing demands of AI infrastructure through improved conversion efficiency, increased power density, and serviceability at higher system voltages. For AI data centers, the significance is that power delivery is becoming a primary design constraint rather than a supporting subsystem. As racks approach megawatt-class power levels, every conversion stage contributes losses, heat and physical volume. Moving to 800 VDC distribution can reduce current, while GaN-based IBCs can help manage the intermediate conversion stage with smaller components and higher switching frequency.

Future Outlook

The Infineon-NVIDIA collaboration builds on prior work in May 2025 and October 2025 and now extends into high-voltage IBCs and power-path protection. The article does not specify product release dates or quantified efficiency targets, but the direction is clear: 800 VDC architectures will require an ecosystem of converters, controllers, drivers and protection devices, not just a single semiconductor component. Infineon’s combination of GaN, silicon and SiC options, together with in-house manufacturing and system-level expertise, positions it to address different stages of that chain. As AI workloads continue to raise rack power, the industry’s ability to deliver efficient, dense and serviceable power conversion will be as important as the compute performance itself.