Navitas Semiconductor unveils 24 kW, 800V-to-12V GaNFast power board built for NVIDIA’s MGX 800V AI push

Navitas Semiconductor said it has developed a new 24 kW, 800 V-to-12 V GaNFast DC-DC power delivery board that is designed to support NVIDIA’s MGX ecosystem for next-generation, 800 VDC AI data-center systems. The announcement, reported in early October 2026, frames the board as a practical step toward making high-voltage AI rack power more efficient and more robust at the component level, rather than relying only on system-scale changes.

In describing the platform, Navitas said the board is “NVIDIA-aligned” and was developed with NVIDIA. For investors and customers, the key point is that the product is positioned to match NVIDIA’s stated direction for the power layer around AI platforms, where 800 V architectures are intended to reduce current and improve efficiency through the distribution chain.

The board converts 800 V down to 12 V using Navitas’ GaNFast technology, referring to the use of gallium nitride (GaN) power semiconductors. GaN devices are widely used in high-efficiency power conversion because they can switch quickly and often enable smaller, lighter power designs compared with traditional silicon approaches. Navitas’ claim is that this kind of conversion, in a format intended for 800 V AI environments, can reduce engineering friction when building and validating AI racks.

Navitas also said the product supports the NVIDIA MGX ecosystem. MGX is NVIDIA’s platform for coordinating and accelerating development of AI infrastructure, spanning hardware, software, and system design guidance so that partners can build compatible components faster. While the company did not detail the depth of integration in the reported summary, the positioning suggests the board is intended to fit into reference power designs that downstream OEMs and system integrators may adopt.

The headline question in the report centers on durability and competitive advantage, essentially arguing that aligning early with NVIDIA-oriented power standards can help a supplier translate platform momentum into repeatable design wins. For Navitas, that means winning not just a single board iteration, but becoming the preferred GaN-based power conversion option within AI data-center builds that follow an 800 VDC approach.

What remains unclear from the reported description is how Navitas characterizes performance beyond the headline power rating. The summary does not provide metrics such as efficiency at specific loads, thermal constraints, switching frequency details, reliability data, or validation results against particular NVIDIA or data-center reference architectures.

The announcement also does not specify manufacturing scale, lead times, or whether the board is available as a general catalog product or primarily as an early partner reference design. Without those details, outside buyers would still need to evaluate engineering fit, qualification timelines, and supply reliability before tying the board to a procurement plan.

For the broader AI hardware supply chain, the message is that the power domain is becoming a competitive battleground. As data-center architects move toward 800 V architectures, the ability to deliver conversion that is compatible with ecosystem guidance can shorten design cycles for system vendors. The next things to watch are whether Navitas expands the product line around 800 VDC conversion, discloses qualification and performance test outcomes, and gains additional partner confirmations beyond the initial NVIDIA-aligned framing.