Navitas Pivots to AI Power Infrastructure With 800V GaN Platform
The wide-bandgap semiconductor specialist is targeting data center power delivery as the hidden bottleneck of AI scaling.
Navitas Semiconductor is chasing one of AI's least obvious opportunities: power delivery. As artificial intelligence workloads explode, the real constraint isn't just chip performance—it's getting electricity to those chips efficiently enough to keep them running without melting.
The Power Bottleneck
Traditional silicon-based power systems are hitting physical limits. AI GPU clusters now demand rack densities exceeding 120kW, generating heat levels that conventional power delivery architectures struggle to manage. Energy lost as heat becomes both a cooling problem and an economic one.
Navitas is betting that wide-bandgap semiconductors—specifically Gallium Nitride (GaN) and Silicon Carbide (SiC)—can solve this. These materials handle higher voltages and temperatures with significantly lower energy loss than silicon, enabling what the industry calls "power density": more watts delivered in less physical space.
Reference Designs Hit the Market
The company has unveiled two key reference designs targeting hyperscale data centers. First, a 12 kW power supply unit compliant with Open Rack v3 and Open Compute Project guidelines, built for 120kW rack densities.
More significantly, Navitas introduced an 800V-to-6V GaN power platform that performs direct voltage conversion in a single stage. This eliminates the traditional 48V intermediate bus converters that have long been standard in data center architecture. The result: up to 96.5% efficiency and approximately 2,100 W/in³ power density.
Removing that intermediate conversion stage matters because every conversion loses energy. At hyperscale, even single-digit percentage improvements translate into megawatts of saved capacity and reduced cooling loads.
NVIDIA Collaboration
In May 2025, NVIDIA selected Navitas to collaborate on its next-generation 800V HVDC architecture designed for rack-scale systems like Rubin Ultra. Mass production of chips from this partnership is expected to begin in 2027.
The collaboration signals that major GPU manufacturers view power delivery as a critical path constraint for future AI compute scaling.
Financial Momentum
Navitas reported Q1 2026 revenue of $8.6 million, up 18% quarter-over-quarter, with high-power markets growing approximately 35% year-over-year. While still a small player compared to established power semiconductor vendors, the company is carving out a niche in the AI infrastructure buildout.
Why This Matters
Power delivery remains the hidden bottleneck of the AI revolution. If data centers cannot efficiently deliver power to GPUs, scaling AI compute becomes limited by thermal and electrical constraints rather than chip design. Navitas' ability to shrink power supply footprints while increasing output allows data centers to pack more compute into the same physical space—directly impacting how fast AI infrastructure can scale.
The bet is straightforward: as AI models grow more complex, the companies that solve the power problem will be as essential as those building the chips themselves.