NVIDIA’s future electric infrastructure for ultra-high-performance AI data centers will feature GaNFast™ and GeneSiC™ technology to improve efficiency, scale power, and reduce operational costs.
In a strategic alliance marking a turning point in the evolution of data center infrastructure for artificial intelligence, Navitas Semiconductor has announced its collaboration with NVIDIA to implement next-generation 800 V HVDC power architecture. This architecture will power rack-scale systems like Kyber, designed to run GPUs such as the upcoming Rubin Ultra, and will be driven by GaNFast™ (gallium nitride) and GeneSiC™ (silicon carbide) technologies.
A Paradigm Shift: From 54 V to 800 V to Scale AI
The traditional data center design based on 54 V DC power is reaching its physical and operational limits. The increasing energy demand of AI requires massive scalability. While current systems operate at a few hundred kilowatts per rack, NVIDIA aims for racks of up to 1 megawatt, and eventually gigawatts at the data center scale.
NVIDIA’s solution involves directly converting 13.8 kV AC input into 800 V HVDC using solid-state transformers (SST), eliminating several intermediate AC/DC and DC/DC conversion stages. This not only enhances energy efficiency but significantly reduces infrastructure complexity and copper usage—a 1MW rack with a 54V system would require over 200 kg of copper.
GaN and SiC Technology: Efficiency, Power, and Reliability
Navitas brings two key technologies to the architecture:
- GaNSafe™: integrated power chips with advanced protection features, ultrafast short-circuit response (350ns), 2kV ESD protection, and programmable switching speed control.
- GeneSiC™ G3F SiC MOSFETs: proprietary “trench-assisted planar” technology that reduces operating temperature by up to 25°C and triples lifespan compared to equivalent solutions.
Both cover the initial conversion from the power grid to the final GPU power supply. The new infrastructure promises to improve energy efficiency by up to 5%, reduce maintenance costs by 70%, and decrease thermal load by connecting racks directly to the HVDC system.
Recent Innovations and Roadmap
Navitas has already demonstrated its leadership with products such as:
- 3.2 kW CRPS (2023) – 40% more compact than silicon solutions.
- 4.5 kW CRPS (2024) – Record density of 137 W/in³.
- 8.5 kW PSU (2024) – 98% efficiency, compliant with OCP and Open Rack v3.
- 12 kW PSU – Presented during Navitas’ AI Tech Night at COMPUTEX 2025.
These units are powered by IntelliWeave™, a patented digital control technique that, combined with GaN and SiC, allows for PFC efficiencies of 99.3%, reducing energy losses by 30%.
Statements
“We are proud to have been chosen by NVIDIA to collaborate on its 800V HVDC architecture. Our advanced power technology transforms efficiency and scalability in AI data centers,” said Gene Sheridan, CEO and co-founder of Navitas.
Conclusion: With this collaboration, NVIDIA solidifies its commitment to an energy-optimized architecture ready for the challenges of large-scale artificial intelligence. The alliance with Navitas aims not only to solve technical issues but also to enhance overall energy efficiency and reduce the environmental impact of future data centers.
Source: Press Release