Texas Instruments introduced a complete 800 V DC power architecture for next-generation AI data centers, developed in alignment with NVIDIA’s 800 V DC reference design. The solution was showcased at NVIDIA GTC 2026 and demonstrates how TI’s analog and embedded processing technologies support high-voltage AI infrastructure.

The architecture is designed to maximize efficiency and power density across the entire power delivery path, addressing the growing power demands driven by AI workloads. By simplifying the conversion chain, the solution enables more scalable and reliable data center operations.

A key feature of TI’s approach is a streamlined two-stage power conversion from 800 V directly to GPU core voltages. The first stage uses an 800 V to 6 V isolated bus converter, followed by a 6 V to sub-1 V multiphase buck converter for high-current GPU power delivery. This architecture reduces conversion losses, improves efficiency and supports higher power density compared to traditional multi-stage designs.

The full solution includes several reference designs. An 800 V hot-swap controller provides scalable input power protection for high-voltage rails. The 800 V to 6 V DC-DC converter integrates GaN power stages and achieves up to 97.6 percent peak efficiency with power density exceeding 2000 W/in³. The 6 V to sub-1 V multiphase buck converter delivers high current density for advanced GPU cores and offers improved performance compared to 12 V-based architectures.

TI also presented additional system-level components, including a 30 kW 800 V high-power-density AC-DC power supply for AI servers and 800 V capacitor bank units using supercapacitor technology for energy storage and transient support. An 800 V to 12 V DC-DC converter was also demonstrated for compute tray applications.

The solution targets the transition toward 800 V DC data center architectures, which are increasingly required to support megawatt-scale racks and high-density AI compute platforms. By leveraging high-voltage distribution and advanced semiconductor technologies, TI aims to enable more efficient, compact and scalable power systems for future AI infrastructure.

Original – Texas Instruments