Overview: Infineon and NVIDIA Extend Their 800VDC Partnership

Infineon Technologies AG of Munich, Germany has announced that it is advancing its collaboration with NVIDIA on 800V direct current (VDC) power architectures for AI data centers. The announcement was timed to coincide with NVIDIA's GTC Berlin developer conference, held 20–22 October, and focuses squarely on the high-voltage intermediate bus converter (IBC) — described by Infineon as one of the most critical and demanding conversion stages in the power delivery chain of an AI data center.

The move builds on joint work between the two companies first announced in May 2025 and detailed further in October 2025, which centered on enabling serviceable, hot-swappable 800VDC server racks. Where that earlier phase addressed rack-level serviceability and safety, the current phase drills into the semiconductor technology that makes the voltage conversion itself viable at scale.

The IBC: A Critical Stage in the Power Delivery Chain

Driven by the extreme power draw of AI workloads, data centers have adopted intermediate-bus architectures. In this scheme, an IBC steps down the main bus voltage that feeds accelerator cards into a lower-voltage rail, allowing downstream stages to efficiently generate the tight core rails and point-of-load rails that AI processors require.

This multi-stage approach reduces distribution losses while meeting strict constraints on board area, thermal limits and uptime. It is precisely this combination of requirements — efficiency, density and reliability simultaneously — that makes the IBC one of the hardest stages to optimize in modern AI infrastructure.

Why Gallium Nitride: Technical Details

Infineon reckons that its gallium nitride (GaN) technology is well suited to the IBC stage. Compared with silicon, GaN allows power devices to switch significantly faster with lower losses. That capability enables higher switching frequencies of up to 1MHz at the IBC stage.

The consequences ripple through the physical design of the converter. Higher switching frequencies allow the size of magnetic components and passives to shrink considerably, which translates directly into higher power density and lower board heights. This is a critical advantage as rack space becomes an increasingly scarce resource in AI data centers.

At the same time, GaN's low conduction and switching losses support best-in-class conversion efficiency, helping to limit heat generation even as power levels per rack continue to climb toward the megawatt range. Taken together, these characteristics make GaN the preferred semiconductor technology today for meeting the combined demands of power density, efficiency and fast transient response at the IBC stage — a position Infineon has built on many years of GaN development and manufacturing experience.

Portfolio Flexibility and System-Level Integration

Infineon's GaN portfolio supports the full range of conversion ratios required at the HV IBC stage, including 800V to 50V, 800V to 12V and 800V to 6V. This gives data-center operators the flexibility to optimize rack architectures without compromising power density or efficiency. At this combination of high switching frequency, power density and efficiency, Infineon argues that GaN power switches are used to their full benefit.

The company pairs its GaN devices with the PSoC control C3 and EiceDRIVER product families to deliver full system solutions for IBCs rather than discrete components alone. Infineon says its power semiconductor technology relies on extensive in-house manufacturing capabilities as well as application and system expertise built up over many years. Its portfolio spans IBC solutions across silicon, gallium nitride and silicon carbide, supporting customers along the entire range of their design requirements.

Market Context: 800VDC Alongside Existing Architectures

As AI infrastructure scales, 800VDC is being introduced alongside existing AC and lower-voltage DC architectures as an additional distribution option for higher-density compute. It is not positioned as a wholesale replacement but as an incremental option that operators can adopt where density and efficiency pressures are greatest.

That framing matters commercially. Data-center operators rarely rip out working power distribution; they layer new architectures into new builds or specific high-density halls. A portfolio that spans silicon, GaN and silicon carbide, and that covers multiple conversion ratios, allows Infineon to serve customers across that transitional landscape rather than betting on a single voltage standard.

Industry Impact and the Grid-to-Core Vision

Infineon and NVIDIA collaborate along the entire data-center power chain. In addition to IBCs, this includes power path protection solutions announced in October 2025, targeting safety and service aspects such as hot-swap controller functionality and enabling future server boards to operate in 800VDC power architectures.

Infineon says the collaboration underscores its position as a grid-to-core power semiconductor partner for AI infrastructure, supplying solutions across the full power delivery chain, from grid-facing power supply units to the point of load at the processor. That breadth is the strategic point: as AI racks move toward megawatt-class draws, the value increasingly sits in system-level co-design rather than in any single device.

Implications for AI Data Centers

For operators, the practical implications are flexibility and thermal headroom. Higher power density and lower board heights free up rack volume; better conversion efficiency limits the heat that cooling systems must remove. NVIDIA highlighted the 800V HVDC system's ability to deliver high reliability and energy efficiency across data centers — a framing that aligns with the industry's broader concern that power delivery, not compute silicon alone, may become the binding constraint on AI capacity expansion.

Future Outlook

Infineon's CEO Jochen Hanebeck framed the effort broadly: “AI is fundamentally reshaping the demands we place on power infrastructure, and semiconductors are at the heart of solving that challenge. Our collaboration with NVIDIA on 800V power architectures shows what is possible when deep semiconductor expertise meets the pace of innovation required by AI. Gallium nitride-based system solutions are a key enabler of this next generation of data centers, and Infineon is proud to help power the AI era from grid to core.”

The next milestones will likely be judged on how quickly 800VDC rack designs move from announced architectures to deployed fleets, and on whether GaN-based IBCs can hold their efficiency and transient-response advantages as per-rack power continues its climb toward the megawatt range.