Navitas Pushes 'Grid-to-GPU' 800V Power Shift for AI Data Centers

Navitas Semiconductor is positioning itself at the center of a fundamental architectural shift in AI data center power delivery. The company is promoting a "Grid-to-GPU" transition designed to replace legacy power systems with high-efficiency Gallium Nitride (GaN) and Silicon Carbide (SiC) power ICs.

To accelerate this transition, Navitas is collaborating with the NVIDIA MGX ecosystem. The partnership focuses on the deployment of 800V DC (HVDC) AI infrastructure, a significant leap from traditional power delivery. By utilizing GaN technology, Navitas enables high-frequency, high-efficiency power conversion that moves electricity from the electrical grid directly down to the processor and GPU level with minimal loss.

The Shift to 800V Architecture

For years, data centers have relied on legacy 48V or 54V systems to distribute power. However, the massive energy demands of modern AI GPU clusters have rendered these lower-voltage systems inefficient. The "Grid-to-GPU" concept represents a transformational shift toward 800V DC architecture. This higher voltage allows for significantly increased power density, reducing the amount of cabling and cooling infrastructure required while increasing the overall efficiency of the power chain.

Why Efficiency Matters for AI

As AI models grow in complexity, the power requirements for GPU clusters have scaled exponentially. Traditional silicon-based power components often struggle with heat dissipation and energy loss at these scales. Navitas' use of GaN and SiC materials allows for faster switching speeds and lower resistance than standard silicon. For the industry, this means a reduction in the total cost of ownership for data centers and a decrease in the carbon footprint associated with powering massive AI workloads.

The Path Forward

The success of this transition depends on the widespread adoption of the 800V standard across the AI hardware ecosystem. While the collaboration with NVIDIA MGX provides a critical entry point, the industry must still move away from entrenched legacy power standards. Observers will be watching to see how quickly these HVDC systems move from specialized AI clusters into broader data center deployments. This shift is not merely a component upgrade but a reimagining of how energy reaches the silicon, ensuring that the physical power layer can keep pace with the rapid evolution of AI compute capabilities.