Tec Nikan
فارسی
Talk to us
All news

An 800 Volt DC Architecture Arrives for AI Data Centre Racks

A complete chipset built to NVIDIA's reference design includes a 30 kW power supply and a bus converter rated at 97.6% peak efficiency, with megawatt racks targeted from 2027.

data centrespower electronics800 VDCAI infrastructuremonitoring

Texas Instruments announced a complete 800 VDC power portfolio on 16 March 2026, developed in collaboration with NVIDIA and aligned to NVIDIA's 800 VDC reference design, and showed it at GTC 2026.

The components span the whole chain: an 800 V hot-swap controller, an 800 V to 6 V DC/DC bus converter rated at 97.6% peak efficiency and more than 2000 W per cubic inch, a 6 V to sub-1 V multiphase buck converter for GPU cores, a 30 kW 800 V AC/DC power supply for AI servers, 800 V capacitor bank units at 40 W per cubic inch, and an 800 V to 12 V bus converter. The architecture uses two-stage conversion from 800 V down to processor voltage rather than the traditional multi-stage chain.

NVIDIA's 800 VDC programme targets 1 MW IT racks and beyond starting in 2027, with Schneider Electric and Vertiv among the named data centre power partners. NVIDIA, Google and Microsoft published a joint 800 VDC white paper through the Open Compute Project in March 2026, followed by a low-voltage DC solid-state transformer specification in July.

The reason this belongs in an industrial IoT publication is what it does to monitoring. An 800 V DC busway in the white space breaks the electrical assumptions that rack instrumentation has been designed around for two decades. DC arc detection is a different product from AC arc fault detection and a harder one, because there is no zero crossing to help extinguish an arc. Insulation monitoring becomes relevant in a way it is not on a grounded AC system. Branch circuit telemetry on a DC bus needs different sensing and different isolation ratings.

The shift from 54 V rack distribution to 800 V also rewrites what the hardware has to survive and what qualifies the people working on it. None of that is a reason to avoid the architecture — the efficiency and copper savings at megawatt scale are substantial — but it does mean the monitoring layer is a new design rather than a rated-up version of the existing one.

Source: Texas Instruments

Want to work with us?

Tell us what you're building and we'll help you scope the first deployment.