Navitas and NVIDIA collaborate on 800 V-to-12 V AI data center power board

By Splitfeed EditorialLast updated Oct 12, 2026, 21:10 UTC

On October 12, 2026, Navitas Semiconductor and NVIDIA announced a jointly developed 800 V-to-12 V power delivery board aimed at next-generation AI data center infrastructure. The board's defining characteristic is that it eliminates an intermediate conversion stage, moving directly from an 800-volt bus down to the 12-volt level that server hardware typically requires. That single-step conversion is the central technical claim of the announcement, and it positions the board as a potential efficiency improvement over architectures that pass through an intermediate voltage, such as 48 V, before reaching the final output.

The broader context matters because data center power delivery has become a genuine bottleneck as AI workloads scale. Racks designed around large GPU clusters draw substantially more power than traditional compute racks, and losses at each conversion stage compound across thousands of servers. Skipping a conversion tier, if it can be done without sacrificing regulation quality or thermal stability, would reduce those losses and potentially allow higher rack densities. Separately, NVIDIA is participating in wider industry discussions around standardizing data center infrastructure, a process that involves aligning on voltage levels, connector types and thermal specifications across suppliers. What remains in dispute is whether 800 V direct-to-12 V architectures can achieve the tight voltage tolerances that sensitive AI accelerators demand, and whether the approach will gain enough industry backing to become a practical standard rather than a one-off demonstration.

None of the three camps, Pro-AI, Anti-AI or Middle Ground, had published specific reactions to this announcement by the time of writing. Pro-AI voices would typically frame a development like this as evidence that the hardware ecosystem is maturing to meet AI's infrastructure demands, pointing to efficiency gains as a reason to accelerate deployment. Anti-AI voices would typically raise questions about the energy appetite driving such engineering efforts in the first place, arguing that optimizing power delivery normalizes ever-larger consumption rather than questioning it. Middle Ground commentators would likely focus on the engineering trade-offs and whether standardization efforts produce genuine interoperability or simply lock buyers into a particular supplier's ecosystem.

The argument will sharpen as technical validation data becomes available and as industry bodies either adopt or bypass 800 V direct-to-12 V as a reference architecture. Any formal specification release from groups working on data center power standards, or independent efficiency measurements from third-party testing, would give observers a concrete basis for judging whether this collaboration delivered on its claims.