Navitas develops 24 kW power board with NVIDIA to remove a conversion step
Its 97% full-load efficiency remains a design target.
Loading page…
Its 97% full-load efficiency remains a design target.
Listen to this story
Announced October 12, 2026, Navitas and NVIDIA’s 24 kW board gives AI compute-tray designers an 800 V-to-12 V path before the processor’s regulator, adding to their earlier 20 kW, 800 V-to-6 V option. The two outputs give customers a choice of intermediate voltage and regulator architecture; Navitas says the single conversion stage and close-to-processor placement can reduce losses and simplify power delivery. The efficiency case is not yet demonstrated in the announcement: 97% is a full-load target, not a measured result.
The board integrates 16 650 V, 18 milliohm NV6034 GaNFast FETs in a stacked half-bridge topology, using dual-cooled DFN 8 × 8 packages.
Its 400 mm × 60 mm form factor is designed for motherboard integration near the processors; Navitas also specifies power density above 2,000 watts per cubic inch.
Navitas plans to show the design in NVIDIA’s AI Factory MGX Wall partner display area at OCP Global Summit 2026.
Navitas Semiconductor has developed a power board with NVIDIA that removes a conversion step between an AI data center’s electrical supply and its processors. Announced October 12, 2026, the 24 kW board converts 800-volt direct current to 12 volts in one stage, supporting NVIDIA’s MGX ecosystem and supplying power for an entire compute tray.
The board’s 12-volt output then goes directly to a voltage regulator module, the component that powers the computing processor.
Navitas says eliminating the extra conversion stage improves system efficiency and reliability while reducing power losses, board space and power-delivery complexity. Those are the company’s stated benefits of the architecture. Its argument combines fewer conversion stages with a layout that brings the power-delivery board closer to the processors it serves.
The board uses NVIDIA’s 400 mm × 60 mm form factor. It is designed for integration onto the compute motherboard, as close as possible to the processors. Navitas says this placement minimizes power-delivery losses and supports high-power demands. The physical layout is therefore part of the company’s efficiency case, alongside the electrical conversion design.
Navitas specifies a target efficiency of 97% at full load and says the board’s power density exceeds 2,000 watts per cubic inch. The first figure concerns conversion efficiency at full output; the second describes power capacity relative to volume. The 97% figure is a design target, not a measured result presented in the announcement.
The new design follows a previously announced 20 kW board, also developed with NVIDIA, that converts 800 volts to 6 volts. This addition changes both the stated power capacity and the output voltage. Navitas describes the portfolio as giving customers flexibility to select an intermediate voltage and regulator architecture suited to their processor-side power requirements.
Navitas plans to show the 800-to-12-volt board in NVIDIA’s AI Factory MGX Wall partner display area at OCP Global Summit 2026.
Loading discussion...
Join the conversation
Explain when an clearly labeled target is useful—or premature.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.