800V DC power is fast becoming the new standard for feeding AI data centres, and a new board from Navitas Semiconductor shows where it is heading. On October 12, 2026, Navitas announced a 24 kW power delivery board, developed in collaboration with NVIDIA, that converts 800 volts DC straight down to 12 volts in a single step. Here’s what that means and why it matters, along with the caveats.
All details come from Navitas’s own announcement, so the performance figures are the company’s claims and targets, not independent test results.
What is 800V DC power, and why the shift?
AI racks now draw so much power that the old ways of distributing it are struggling. Using a higher voltage like 800V DC lets operators move the same power with less current, which can mean thinner cables and lower losses. We explained this shift in detail in 800 Volt DC: the next power shift in AI data centers.
The problem is that AI chips themselves run at very low voltages. So power has to be stepped down, usually in several stages, and each stage wastes some energy and takes up space.
What Navitas and NVIDIA’s board does
According to Navitas, the new power delivery board (PDB):
- Delivers 24 kW, enough to power an entire compute tray
- Converts 800V DC directly to 12V DC in one stage
- Removes the existing 50V-to-12V conversion stage (called the IBC stage)
- Targets 97% efficiency at full load
- Has a power density above 2,000 watts per cubic inch
- Fits NVIDIA’s latest 400 mm × 60 mm form factor, so it can sit on the motherboard close to the processors
The 12V output feeds the voltage regulator module that powers the processor. Navitas says cutting out a stage improves efficiency and reliability while saving board space.
The board is built for NVIDIA’s MGX ecosystem and is planned to be shown at the OCP Global Summit 2026. It follows an earlier 20 kW, 800V-to-6V board the two companies developed.
Why gallium nitride (GaN) matters here
The board uses 16 of Navitas’s NV6034 GaN transistors, rated at 650V. Gallium nitride is a material that can switch power faster and more efficiently than traditional silicon in many uses, which helps make power converters smaller. Navitas says its strategy is to replace “legacy silicon” with GaN and silicon carbide across AI infrastructure. That’s the company’s business goal, and silicon is still widely used.
Navitas CEO Chris Allexandre said the collaboration is “breaking the AI infrastructure power wall.”
The caveats
There are a few important points to keep in mind:
- 97% is a target. Navitas describes the efficiency as a target, not a measured result, and no independent test data was shared.
- A showcase, not a shipping product. The release talks about a display at a trade show. It doesn’t give availability dates, prices or named customers.
- Safety. 800V DC is a hazardous voltage. Bringing it closer to the motherboard raises the bar for protection, isolation and technician training. The announcement doesn’t go into these safety measures.
- Market context. Plenty of companies are racing to supply 800V power parts, as we noted in our look at the $95 billion power conversion market.
My view
As a writer, I feel power delivery is where a lot of the real AI engineering is happening now. Removing a whole conversion step is a neat idea, and if the 97% target holds up, the savings across thousands of racks could be large. But I’d want to see measured results and a clear safety story before getting too excited about 800V running this close to the chips.
FAQ
What is 800V DC power in data centers?
It’s a way of distributing electricity inside AI data centres at 800 volts direct current, which moves the same power with less current than lower-voltage systems.
What did Navitas and NVIDIA announce?
A 24 kW board that converts 800V DC to 12V DC in a single stage, with a target efficiency of 97% at full load, according to Navitas.
Is the Navitas board available now?
The announcement says it is planned to be showcased at OCP Global Summit 2026. It does not give a release date.
1 thought on “800V DC Power: Navitas and NVIDIA 24 kW GaN Board”