Navitas Semiconductor introduced a 24-kilowatt power-delivery board developed with Nvidia for the Nvidia MGX ecosystem. The design converts 800 volts directly to 12 volts and targets 97% efficiency at full load, addressing a growing constraint in AI data centers: moving large amounts of electricity to processors without wasting space or power.
What the board changes
Traditional architectures may use an intermediate conversion stage before delivering low-voltage power to a compute tray. Navitas says its board removes that step, reducing conversion losses and simplifying the power path.
The company targets power density above 2,000 watts per cubic inch. The board uses 16 of Navitas' 650-volt gallium nitride, or GaN, power transistors and is designed to supply an entire compute tray. GaN devices can switch power efficiently at high frequency, which can reduce the size of supporting components.
Those are engineering targets, not measured customer economics.
The 97% figure is a target stated by Navitas. It does not by itself establish production yield, system reliability, unit cost or total electricity savings in a deployed data center.
The phrase "developed in collaboration with Nvidia" is also narrower than a purchase order. It confirms technical alignment with Nvidia's architecture, but Navitas did not disclose an order volume, price, customer commitment or revenue forecast in the announcement.
Why the technical milestone still matters
AI racks are becoming more power dense. A one-stage conversion design can reduce heat and free motherboard space, both of which become more valuable as compute power rises. Navitas previously announced a 20 kW board that converted 800 volts to 6 volts and targeted up to 97.5% peak efficiency, so the 24 kW product shows a broader set of voltage and tray options rather than a first experiment.
The opportunity can be material for Navitas because the company is much smaller than Nvidia. A successful qualification and production ramp could change Navitas' revenue mix more than it changes Nvidia's. The reverse is also true: delays or a competing power architecture would matter disproportionately to Navitas.
What investors should watch next
The decisive evidence will be customer qualification, shipment timing, production volume, gross margin and named design wins. Investors should also compare claimed efficiency at full load with efficiency across typical operating ranges.
BTI's conclusion is that the board strengthens Navitas' technical position in AI power delivery, but the investment case should not count collaboration as revenue before commercial orders are disclosed.