Component Market Insight · AI Data Center Power
NVIDIA 800 VDC Architecture Supply Chain: Power Components Move to the Center of AI Infrastructure
AI infrastructure is forcing a major change in data center power delivery. As rack power moves from tens of kilowatts toward hundreds of kilowatts and eventually megawatt-class systems, traditional 48 V or 54 V distribution becomes harder to scale. NVIDIA’s 800 VDC architecture is designed to reduce current, lower copper usage, simplify conversion stages, and support future AI factories with higher compute density.
This shift is not only important for GPU makers and system integrators. It also creates a new sourcing theme for power semiconductors. Devices such as high-voltage hot-swap controllers, isolated gate drivers, current and voltage sensors, SiC MOSFETs, GaN HEMTs, digital power controllers, smart fuses, and high-current multiphase power stages are becoming more strategic in the AI server supply chain.
Based on public ecosystem announcements, companies such as Texas Instruments, onsemi, and Infineon are important names to watch. The following analysis does not describe a confirmed NVIDIA bill of materials. Instead, it explains the likely technical roles, representative new products or product families, and the sourcing logic behind the 800 VDC transition.
1. Texas Instruments: Power Management, Sensing and System-Level Control
TI’s role is closely tied to analog control, isolation, protection, sensing, and power conversion from the high-voltage bus down toward the GPU power rail. In an 800 VDC architecture, the system must monitor voltage and current in real time, detect abnormal conditions quickly, isolate faults, and manage power conversion with very high efficiency.
TI has publicly shown an 800 VDC power architecture for future-generation AI data centers with NVIDIA. The architecture includes an 800 V hot-swap controller for input protection, an 800 V to 6 V isolated DC/DC bus converter using integrated GaN power stages, and a 6 V to below 1 V multiphase buck solution for advanced GPU core power. This is important because it shortens the power path and reduces the number of conversion stages between the 800 V bus and the processor rail.
Representative TI parts and product areas worth watching include TPSI3050-Q1 and TPSI2140-Q1 for isolated solid-state relay and high-voltage isolation functions, UCC5880-Q1 as a high-performance isolated gate driver example for SiC and IGBT systems, and TI’s 800 V hot-swap, GaN power stage, current sensing, voltage sensing, and multiphase controller portfolios for AI data center power designs.
From a supply chain perspective, TI-related demand may not be limited to one single part number. The bigger opportunity is the full analog power path: protection, monitoring, bias supply, isolation, control loop, hot-swap, and point-of-load regulation. In high-power AI racks, these devices are not optional support chips. They help determine safety, uptime, conversion efficiency, and serviceability.
2. onsemi: SiC, Smart Power Conversion and High-Reliability Devices
onsemi’s position is more centered on intelligent power conversion across the grid-to-GPU power chain. Its public AI data center positioning emphasizes EliteSiC MOSFETs and JFETs, PowerTrench MOSFETs, GaN HEMTs, PoL regulators, smart fuses, and Vcore power solutions for high-density server architectures.
For the front-end AC/DC rectification stage and high-voltage power conversion, SiC MOSFETs are especially relevant. SiC devices can reduce switching and conduction losses, improve thermal performance, and support higher power density compared with traditional silicon devices. In 800 VDC data centers, this matters because every conversion loss becomes expensive at rack and facility scale.
Representative onsemi products and families to watch include 650 V EliteSiC M3S MOSFETs such as NTH4L023N065M3S, 1200 V EliteSiC MOSFETs for higher-voltage conversion stages, and newer 1200 V SiC MOSFET-based intelligent power modules for compact high-efficiency power conversion. onsemi also highlights the combination of EliteSiC and PowerTrench T10 MOSFETs for AI data center power supplies.
In sourcing terms, this means onsemi demand may appear in several layers: front-end rectification, server power supply, rack power conversion, protection, and processor-level power delivery. The market should not only watch large SiC discrete devices. It should also monitor smart fuses, PoL regulators, low-voltage MOSFETs, and power modules that sit closer to the server and accelerator board.
3. Infineon: SiC, GaN, IGBT, Gate Drivers and Digital Power Control
Infineon has a broad power semiconductor portfolio, which makes it relevant across multiple stages of the 800 VDC power chain. Its product coverage includes CoolSiC devices, CoolGaN switches, silicon power MOSFETs, IGBTs, gate drivers, digital isolators, sensors, auxiliary power solutions, MCUs, and digital power controllers.
Infineon has publicly supported NVIDIA’s 800 VDC architecture and has introduced CoolGaN-based high-voltage intermediate bus converter reference designs. These include 800 VDC to 50 V and 800 VDC to 12 V conversion paths for next-generation AI data center systems. For custom implementations, Infineon also offers the XDPP1188-200C digital controller, supporting flexible output voltages such as 48 V, 24 V, and 12 V.
Representative Infineon devices and product families to watch include IGT65R035D2, a 650 V CoolGaN G5 power transistor, CoolGaN bidirectional switch 650 V G5 devices for high-frequency high-density server power supplies, CoolSiC JFET technology for serviceable 800 VDC rack designs, and 650 V / 750 V CoolSiC MOSFET families for high-efficiency AI server PSUs.
Infineon’s role is especially important where power density and conversion efficiency must improve without sacrificing safety and reliability. In an 800 VDC architecture, the design challenge is not simply “higher voltage.” The system must manage fast transient loads, hot-swap behavior, EMI, thermal stress, isolation, and long-term reliability at data center scale.
Supply Chain Implications: The New Hot List Is Not Only GPUs
The 800 VDC transition changes how sourcing teams should think about AI infrastructure. The market usually focuses on GPUs, HBM, high-speed networking, retimers, and optical modules. But as rack power increases, power semiconductors become just as strategic. A shortage in hot-swap controllers, isolated drivers, SiC MOSFETs, GaN switches, smart fuses, current sensors, or digital power controllers can also delay production.
For buyers, the first step is to separate three categories: confirmed BOM parts, platform-related ecosystem parts, and adjacent watchlist parts. Confirmed BOM parts need exact suffix control and traceability. Ecosystem parts need technical monitoring because they may become future design-in candidates. Adjacent watchlist parts help identify broader market pressure before spot pricing rises.
For distributors, the opportunity is not to randomly stock every “AI power” part number. The better approach is to follow the actual power chain: grid-side AC/DC, 800 VDC distribution, rack hot-swap and protection, intermediate bus conversion, 6 V or 12 V board-level conversion, and sub-1 V GPU core regulation. Each layer has different suppliers, package types, qualification standards, and shortage behavior.
Representative Part Numbers and Product Families to Track
A practical watchlist may include TI’s TPSI3050-Q1, TPSI2140-Q1, UCC5880-Q1 and 800 V hot-swap / GaN power-stage reference designs; onsemi’s NTH4L023N065M3S, 650 V EliteSiC M3S, 1200 V EliteSiC MOSFET and SiC IPM families; and Infineon’s XDPP1188-200C, IGT65R035D2, CoolGaN 650 V G5, CoolGaN bidirectional switch and CoolSiC JFET / MOSFET families.
These models are not listed as direct substitutes for one another. They represent different locations in the AI data center power path. The sourcing value comes from understanding where each device sits and how close it is to a real customer platform.
Conclusion
NVIDIA’s 800 VDC architecture is turning power delivery into a core AI infrastructure layer. For TI, onsemi and Infineon, the opportunity spans high-voltage protection, sensing, conversion, isolation, SiC, GaN, IGBT, digital control and GPU-near power regulation. For the semiconductor market, this means the next wave of AI demand will not be limited to compute chips. It will also move through the entire power chain.
For sourcing teams, the message is clear: track the power devices before they become obvious shortage parts. In the AI server cycle, the most valuable components are often the ones that make the whole rack work reliably.