Texas Instruments (TI), together with Nvidia, has unveiled an 800VDC power architecture for next-generation AI data centers. With the proliferation of generative AI driving increasing power demand at the server and rack levels, this architecture is seen as an attempt to improve efficiency and power density by reducing power conversion steps.

TI unveiled this architecture at the NVIDIA Developer Conference GTC 2026, held in San Jose, California, from March 16 to 19. It presented a new high-voltage direct current-based power supply method at a time when the limitations of data center power distribution structures are being discussed due to the expansion of AI infrastructure.

The core of the technology lies in reducing 800V power to the GPU core voltage in just two conversions. It is a structure that first passes through an isolated bus converter that converts 800V to 6V, and then applies a multiphase buck converter that lowers 6V to less than 1V. The plan is to simplify the power path to reduce losses and support the operation of AI accelerators that require high current.

Along with this, TI presented a hot-swap controller for 800V input power protection, an 800V-6V DC/DC bus converter based on an integrated GaN power stage, and a 6V-less-1V multiphase buck converter for GPUs. It also unveiled a 30kW 800V AC/DC power supply for AI servers, an 800V capacitor bank unit, and an 800V-12V DC/DC bus converter.

The application targets the entire high-density computing infrastructure, including AI servers, compute trays, server racks, and power distribution devices. In an environment where power consumption per rack continues to rise, high-voltage DC distribution is being discussed as an alternative that reduces losses by lowering current and alleviates the burden on wiring and power devices.

The industry views this announcement as an extension of the trend in which AI data center power structures are shifting from existing 12V-centric designs to higher-voltage-based systems. Given that semiconductor companies and AI platform firms have begun collaborating at the power structure stage, this is interpreted as an example demonstrating that future data center competitiveness may be determined not only by computational performance but also by power design methods.