Renesas tests 20 kW system converting 800 V directly to 6 V for AI servers

CONFIRMED: Renesas Electronics is developing a 20 kW power system for AI servers that converts 800 V DC directly to 6 V in a single stage. The information was published on October 23, 2026, by EE Times Japan, based on its own reporting, and the company has already supplied prototype platforms to major AI data center OEMs and ODMs.

How the system reaches 800 V with 650 V parts

The design combines four 5 kW full-bridge LLC converters. Two converters are paralleled into a 10 kW block, and two blocks are connected in series on the input side and in parallel on the output side. That way, each 5 kW converter sees about 400 V of the 800 V input — which is what makes it possible to use 650 V GaN (gallium nitride) devices, such as Renesas's own TP65H020G4PLSGBD, with 20 mΩ on-resistance, an 8 x 8 mm PQFN package and cooling on both sides of the package.

What is not yet proven

One important caveat: the roughly 96.3% efficiency at 20 kW cited for the system is an analytical projection, calculated from component losses, not a value measured on real hardware. Evaluation of the prototype is still under way, and Renesas itself says electromagnetic interference, protection, thermal management and long-term reliability still need verification. No mass-production timing or volume design win has been announced for this 20 kW platform either.

Why it matters for AI data centers

The logic is straightforward: as AI racks consume more power, distributing power at 800 V cuts cable losses, which grow with the square of the current. A conventional architecture would go from 800 V to 48 V and then to 6 V; Renesas's system removes that intermediate stage. Note that 6 V is not the GPU's final voltage — a downstream multiphase converter steps it down to about 1 V or less, the operating range of the compute circuits.

Image credit: image released by Renesas Electronics of the dual-side-cooled GaN device, reproduced by ENGtechnica.

Sources

Transparency: this article was produced with AI assistance, based on cross-checking public posts and the internet sources listed above. Check the original sources to confirm technical details and availability before making any decision.

Muse