Navitas targets AI data center power needs with new Nvidia board

Solving the power wall

AI data centers are facing a significant physical challenge: getting enough electricity to high-performance processors without losing too much energy as heat. Navitas Semiconductor recently unveiled a new 24 kW power board designed to work with Nvidia hardware to address this specific bottleneck.

The board is designed to convert 800VDC directly to 12VDC in a single stage. By doing this in one step, the system eliminates the need for a 50V to 12V intermediate converter. This streamlined process allows the board to feed power straight to the voltage regulator located next to the xPU, which is the primary processing unit in these systems.

According to TIKR.com, the goal of this design is to achieve 97% efficiency at full load. This high level of efficiency is critical because every percentage of power lost is wasted as heat, which increases cooling costs and reduces the amount of computing power available to the user.

Engineering and scale

The new hardware utilizes Nvidia’s 400mm by 60mm MGX form factor. It is built to be incredibly compact, reaching power densities of over 2,000 W per cubic inch. This density is a key metric for data center operators who want to pack as much computing power as possible into a limited physical footprint.

This project builds on previous work between the two companies, including a 20 kW board that converts 800V to 6V. It also complements a separate reference design created with Microchip Technology. CEO Chris Allexandre stated that the collaboration is "breaking the AI infrastructure power wall."

Navitas is attempting to capture the entire power path within the data center. Their strategy involves using different materials for different parts of the journey: Silicon Carbide (SiC) for the grid, and Gallium Nitride (GaN) and SiC for the 800V conversion. They recently completed an acquisition of Claros to handle the final 'point of load' delivery, which is the very last inch of power delivery to the processor.

The path to revenue

While the engineering wins are notable, Navitas is still in a transition phase regarding its financial performance. The company reported Q2 revenue of $10.5 million, which was a 27% decrease compared to the previous year. However, management has guided Q3 revenue to approximately $13.5 million, which would represent a 28% sequential increase.

The company currently faces a difficult financial profile, with a last twelve months (LTM) EBIT margin of negative 252.3%. This means that operating costs are currently much higher than the revenue being generated. To offset these costs, Navitas holds about $557 million in cash as of the end of Q2.

Investors are watching the upcoming Q3 report on November 3 closely. The market is looking for evidence that the new AI data center demand is beginning to lift sales figures. Because the stock currently trades at 29.9x forward EV/sales, the valuation relies heavily on the assumption that these new design wins will eventually turn into high-volume revenue.

Competitive landscape

Navitas is not alone in this space. Other major semiconductor players like STMicroelectronics and Texas Instruments have also launched 800VDC products around Nvidia’s reference designs. Navitas must differentiate itself through its integrated approach to the entire power path.

Beyond the data center, the company is diversifying its technology applications. It was recently selected by the U.S. Army to develop 10 kV SiC IGBT technology. These types of high-voltage components are essential for heavy-duty industrial and military applications, providing a secondary path for growth if the AI data center market takes longer to materialize.

The company is also actively protecting its intellectual property. In August, Navitas filed a lawsuit against Renesas Electronics in Texas regarding GaN patents. This legal move highlights the importance of Gallium Nitride—a material used for high-frequency, high-efficiency power conversion—as a core component of their competitive advantage.