The race for artificial intelligence is no longer solely fought on GPUs, HBM memory, or low-latency networks. It is also shifting to a less visible but increasingly critical layer: power supply within data centers. Infineon Technologies has joined the NVIDIA MGX AI Factory ecosystem to support the transition to 800 VDC power architectures, a key component in future high-density AI deployments.
This move carries both technical and strategic significance. NVIDIA is expanding its influence beyond just accelerators and compute racks to push for a more integrated AI infrastructure design. Infineon, in turn, enters the MGX ecosystem as a provider of power management solutions capable of handling power flows from the grid through intermediate voltages and ultimately to the processor core.
The pressure is clear. AI data centers need to pack more computing into the same physical space, constrained by electrical, thermal, and distribution limits that aren’t growing at the same rate. In this context, every energy conversion, cable, intermediate stage, and distribution loss matters. The 800 VDC architecture aims precisely to reduce these losses, simplify infrastructure, and enable higher rack densities.
800 VDC: Less Conversion, Higher Density
Traditional electrical architectures of data centers were designed for loads quite different from today’s. For years, enterprise racks consumed moderate power and could be supplied via conventional AC schemes, with multiple conversions up to servers. AI has transformed that logic. Next-generation accelerators handle hundreds of kilowatts, and industry is looking toward megawatt-scale configurations.
NVIDIA claims that the 800 VDC architecture reduces conversion stages, distribution losses, cabling volume, and copper usage compared to rack-level 54 VDC systems or 480 VAC at the facility level. It is presented as a gradual evolution from current architectures toward data centers optimized for much greater densities.
The involvement of Infineon reinforces this roadmap. The German company will provide solutions for power conversion and management compatible with NVIDIA MGX, including silicon, silicon carbide (SiC), and gallium nitride (GaN) technologies. According to Infineon, its solutions will enable conversions from 800 V to 50 V, 12 V, or even 6 V, covering much of the electrical pathway needed for computing systems.
The technical details are significant. Infineon emphasizes the use of GaN switching at frequencies near 1 MHz to create ultra-compact bus converters, along with its SiC JFET technologies combined with control circuits for protection and hot-swap functions on native 800 V server boards. Simply put: it’s not enough to supply higher voltage to the rack; it must be converted, protected, and managed without compromising availability or security.
NVIDIA Aims to Control More of the AI Factory Design
MGX was conceived as a modular architecture to accelerate the design of AI servers and racks. But its evolution shows that NVIDIA does not want to limit itself to selling GPUs. The company is building a comprehensive ecosystem encompassing servers, racks, networking, liquid cooling, power supplies, software, and third-party components.
The partnership with Infineon aligns with this strategy. If AI requires a redesign of the data center’s energy chain, NVIDIA needs power semiconductor manufacturers, electrical suppliers, and infrastructure partners to adopt its architecture. NVIDIA has already positioned Infineon among the silicon providers for its 800 VDC initiative, alongside players like Analog Devices, Texas Instruments, Renesas, STMicroelectronics, ROHM, Navitas, onsemi, MPS, and Innoscience.
This points to a clearer idea: AI data centers are being designed as complete systems, not just a collection of individual components. Chip efficiency depends on memory, networking, cooling, orchestration software, and how the energy is delivered. Bottlenecks can occur anywhere along this chain.
For operators, this approach offers advantages. A more modular and standardized architecture can accelerate deployments, improve efficiency, and reduce integration risks. It also protects current investments by allowing a hybrid transition to 800 VDC without waiting for entirely new data centers. NVIDIA claims its compatible 800 VDC MGX power racks offer an upgrade path for existing infrastructure, increasing density and performance without immediately requiring a fully native 800 V AI Factory.
However, there is also a competitive angle. The deeper the MGX ecosystem, the more influence NVIDIA can wield over the architectural decisions of major AI buyers. It’s not just about choosing GPUs, but about adopting a specific way to build the AI factory.
Electric Power Becomes Part of the AI Value Chain
Infineon’s entrance into the MGX ecosystem demonstrates that the AI supply chain is expanding. Until now, much of the public debate centered on NVIDIA, TSMC, HBM memory, servers from Supermicro, Dell, HPE, and high-speed networks. Moving forward, power electronics suppliers will have greater influence.
Infineon is well-positioned in this area. The company describes itself as a global leader in power systems and IoT, with about 57,000 employees as of September 2025 and revenues around €14.7 billion in fiscal year 2025. Its experience with silicon, SiC, and GaN allows it to cover various layers of electrical conversion—crucial during a transition that cannot be solved with a single technology.
The opportunity is large, but so are the demands. AI data centers seek more power per rack, higher efficiency, less copper, reduced complexity, and increased availability. A failure in the power supply of a high-density AI rack isn’t minor: it can impact systems with million-euro hardware, critical training loads, or production inference services.
This drives industry toward a co-design model. NVIDIA cannot push megawatt racks without electrical suppliers, power semiconductor manufacturers, integrators, and data center operators moving in the same direction. Infineon also cannot just sell isolated components if the market demands integrated solutions from the grid to the processor.
| Element | Contribution to the 800 VDC architecture |
|---|---|
| NVIDIA MGX | Modular reference for AI servers and racks |
| 800 VDC | Reduced losses, less copper, higher power density |
| Infineon SiC | Protection, hot-swap, high-voltage conversion |
| Infineon GaN | Compact high-frequency converters |
| Conversion to 50 V, 12 V, and 6 V | Adapts power for servers and processors |
| MGX Ecosystem | Coordination among chips, racks, power, and suppliers |
Electrical Efficiency Will Drive AI’s Pace
This announcement comes at a time when AI infrastructure is reaching physical limits. Power availability, connection to the electric grid, cooling, and rack density already influence where data centers are built and what workloads they can host. In some markets, securing reliable power has become as challenging as acquiring GPUs.
While the 800 VDC architecture won’t solve all these issues by itself, it can reduce internal inefficiencies. Eliminating conversions, reducing cabling, and delivering direct current closer to the rack can provide headroom for scaling. In facilities where each percentage point of efficiency impacts economic, thermal, and operational metrics, this margin can be decisive.
It may also change procurement strategies. Large AI customers won’t just evaluate accelerator performance but will consider entire system efficiency: input power, losses, cooling, density, maintenance, expandability, and total operational cost. In this context, partnerships like NVIDIA and Infineon become especially relevant.
For Europe, this move has further significance. Infineon is one of the continent’s leading power semiconductor manufacturers, and its involvement in the MGX ecosystem shows that the European industry can play an important role in the energy layer of AI, even if it does not lead the GPU market. In an era of technological sovereignty, power chips, electrical conversion, and data center efficiency are also part of digital competitiveness.
AI is becoming a heavy infrastructure industry. It’s no longer enough to increase model size or parameters. The key question is whether the electrical, thermal, and physical systems supporting these models can grow at the same pace. NVIDIA and Infineon are directly targeting that bottleneck.
Frequently Asked Questions
What have NVIDIA and Infineon announced?
Infineon has joined the NVIDIA MGX AI Factory ecosystem to provide power management and conversion solutions compatible with the 800 VDC architecture for AI data centers.
Why is 800 VDC discussed in data centers?
Because DC distribution at 800 V can reduce losses, decrease copper use, simplify conversion stages, and enable higher power density racks.
What technologies does Infineon contribute?
Infineon combines silicon, silicon carbide (SiC), and gallium nitride (GaN) solutions to convert from 800 V to intermediate voltages like 50 V, 12 V, or 6 V, along with protection and hot-swap functions.
Why is this important for AI?
Because the growth of models and accelerated racks is constrained by power, cooling, and density. Improving electrical distribution can help scale AI infrastructure more efficiently and with less complexity.
via: Infineon