Navitas Semiconductor and Nvidia have struck a partnership to overhaul power delivery in AI data centers, targeting the energy losses that have become a growing headache as racks of GPUs demand ever more electricity. The two companies said the collaboration is aimed at boosting power efficiency while also trimming material costs tied to the hardware that moves current from the grid to the chip.
The deal pairs Nvidia's dominance in AI accelerators with Navitas' work on power semiconductors, including gallium nitride and silicon carbide technologies. For data center operators, the payoff could show up in lower electricity bills and the ability to pack more compute into the same footprint.
Why power delivery is suddenly a boardroom topic
AI models keep getting bigger, and the servers training them draw power at levels that old data center designs weren't built to handle. A single rack of high-end GPUs can pull as much electricity as a small neighborhood. Most of that power still has to pass through multiple conversion stages, and each stage wastes some of it as heat. Those losses stack up fast when you're running thousands of racks around the clock.
Navitas has been pitching wide-bandgap semiconductors as a fix. These chips switch faster and waste less energy than traditional silicon. Nvidia, meanwhile, has been under pressure from customers who want more compute per watt, not just raw speed. The partnership suggests Nvidia sees power delivery as a limiting factor for the next generation of AI systems, not just a peripheral concern.
Material costs and the scalability question
Beyond efficiency, the companies say the tie-up could reduce material costs associated with power delivery. That's a significant claim in an industry where copper, rare earths, and specialized components have all seen price swings. If Navitas' designs can replace bulkier power stages or cut the number of components needed per server, the savings could ripple through data center build-outs.
Scalability is the other piece. AI data centers are being planned at a scale that would have seemed absurd a few years ago — campuses drawing hundreds of megawatts, sometimes more. Power delivery systems that work in a lab often stumble when you try to deploy them across acres of server halls. The Navitas-Nvidia collaboration is explicitly aimed at making sure the technology can grow with demand rather than becoming a bottleneck.
Neither company disclosed financial terms or a timeline for when products from the partnership will hit the market. That leaves open questions about which specific Nvidia platforms will get the new power designs first, and whether existing data centers can retrofit them or will need new builds.
What this means for the AI supply chain
Power has become one of the hardest constraints on AI expansion. Utilities in some regions are already delaying connections for new data centers because the grid can't keep up. Anything that squeezes more computation out of each megawatt helps, but it doesn't solve the fundamental problem of generating and transmitting that power in the first place.
For Navitas, the Nvidia partnership is a credibility boost. The company has been trying to win design slots against larger power semiconductor rivals, and an endorsement from the biggest name in AI carries weight. For Nvidia, it's another move to control more of the stack around its GPUs, following earlier work on networking, software, and cooling.
The collaboration doesn't guarantee results. Power delivery is a conservative field where reliability matters more than raw performance, and data center operators are slow to swap out proven components. The real test will come when Navitas and Nvidia show actual hardware and independent testing backs up the efficiency claims.
Next step: hardware on the horizon
Neither company has said when the first products from the partnership will be available or which customers will get them first. That leaves the AI hardware industry watching for a concrete announcement — likely a specific Nvidia platform paired with Navitas power stages — before treating the collaboration as more than a promising press release.
