Overview of the Navitas and Nvidia Collaboration

As artificial intelligence models continue to dominate technological headlines, the underlying electrical infrastructure required to power them is undergoing a critical transformation. Navitas Semiconductor (Nasdaq: NVTS) is addressing this foundational challenge by collaborating with Nvidia to develop advanced power semiconductor solutions. Announced initially on May 21, 2025, this partnership focuses on next-generation 800 VDC architectures specifically designed for AI data centers and GPU racks. By leveraging gallium nitride (GaN) and silicon carbide (SiC) technologies, the collaboration aims to significantly enhance power efficiency and density, effectively rewiring how massive computational facilities consume and distribute electricity.

Technical Specifications of the Power Delivery Board

The core of this collaboration is a direct 800 V-to-6 V DC-DC conversion board, which represents a significant departure from conventional 48 V and 50 V systems that require inefficient intermediate conversion stages. The Navitas board bypasses these layovers entirely, utilizing two primary product lines: GaNFast for gallium nitride applications and GeneSiC for silicon carbide solutions.

The technical specifications of the board are highly optimized for next-generation server racks:

  • Peak efficiency targets ranging from 96.5% to 97.5% at 1 MHz switching frequencies.
  • Power densities reaching up to 2,100 W/in³.
  • An ultra-low-profile design that is roughly 20% thinner than a standard smartphone.
  • Under-the-hood integration of 16 GaNFast FETs rated at 650 V, featuring 11–17 mΩ specifications.

Material Savings and Architectural Efficiencies

Beyond raw electrical efficiency, the 800 V architecture promises substantial material and structural savings for data center operators. The design facilitates up to 45% thinner wiring, which directly translates to a 45% reduction in overall copper requirements. This reduction not only lowers material costs but also minimizes conversion losses inherent in traditional setups. Furthermore, the new architecture is engineered to target 1 MW+ power delivery, a massive leap forward compared to traditional systems that typically max out at hundreds of kilowatts.

Integration with Nvidia's Next-Generation Ecosystem

The Navitas power solutions are deeply integrated into Nvidia’s broader hardware ecosystem, specifically the MGX ecosystem and its 800 V HVDC initiative. This initiative encompasses advanced platforms, including the upcoming Kyber racks and Rubin Ultra GPUs. The partnership has rapidly progressed from initial press announcements to tangible hardware demonstrations. Navitas successfully exhibited the 800 V-to-6 V power delivery board at COMPUTEX, which took place from June 2 to June 5, 2026, showcasing the physical realization of their collaborative engineering efforts.

Market Context and Industry Impact

The shift toward 800 V HVDC architectures is catalyzing broader industry participation, extending beyond a single vendor. On October 5, 2026, Navitas joined forces with Microchip to unveil a joint reference design. These reference designs serve as critical blueprints, allowing hardware manufacturers to adopt proven solutions rather than engineering complex power delivery systems from scratch. The market has reacted strongly to these developments, with NVTS shares reportedly posting intraday gains of up to 20% following the partnership announcements, reflecting investor confidence in the company's strategic positioning within the AI infrastructure boom.

Implications for AI Data Centers and Future Outlook

While the technical targets are ambitious, the ultimate success of this initiative will depend on real-world deployment and scalability. Industry observers will be closely monitoring several key metrics in the coming quarters. First, the broader adoption of the 800 V HVDC standard among major data center builders will dictate the pace of infrastructure upgrades. Second, Navitas’s ability to convert its current design wins into sustained revenue growth will be a critical measure of commercial success. Finally, the practical rollout and performance of platforms like the Kyber racks and Rubin Ultra GPUs will validate whether the projected 96.5% to 97.5% peak efficiency figures hold up under the extreme, continuous loads characteristic of modern AI training workloads.