Nvidia used to be a company that designed chips and let others worry about where they were made. This week it leaned harder into the physical end of the business. On Tuesday it confirmed plans to upgrade a chip plant in Texas alongside its manufacturing partner Coherent, a facility the two say will lean on silicon photonics, the technology of moving data with light rather than electrons. The promised payoff is a roughly 50 percent cut in the plant's power consumption, and, by Coherent's estimate, about 1,000 new jobs.
The efficiency claim is the part worth lingering on, because it lands in the middle of a louder argument about whether AI's appetite for electricity can be tamed. Photonics matters because shuttling data between chips is one of the quieter energy sinks in a modern data centre; replace copper interconnects with light and you cut both the power and the heat that power becomes. "With Coherent, Nvidia is pioneering next-generation silicon photonics to enable AI infrastructure at unprecedented scale, speed and energy efficiency," chief executive Jensen Huang said, framing the factory itself as part of the product.
Huang has taken to calling these sites "AI factories," and described them as "the infrastructure of the new industrial revolution." It is salesmanship, but it is also a fair description of where the money is going. By one Federal Reserve Bank of St. Louis figure, sectors tied to AI accounted for 37 percent of US GDP growth across the first three quarters of 2025. When a single industry is doing that much of the lifting, where its supply chain physically sits stops being a corporate detail and becomes a question of national strategy.
Which is why the announcement came wrapped in politics. Huang appeared with President Trump, who used the moment to cast chip manufacturing as a race the United States is winning. "Whoever leads is going to really lead the world to a large extent," he said. "We are leading China by a lot." The backdrop is real: China is reported to be pouring some 295 billion dollars into building data centres over the next five years, and Washington has spent the past year trying to keep advanced chipmaking on American soil rather than ceding it abroad.
It is worth keeping the scale honest. A single retooled plant and a thousand jobs are modest against an industry measured in hundreds of billions, and a 50 percent efficiency claim is a projection from the companies that benefit from it, not yet a measured result. But the direction is the interesting signal. The competitive edge in AI is shifting from who has the cleverest model to who can build and power the hardware behind it, and increasingly to who can do that at home. Nvidia, long the most asset-light winner of the boom, is now betting that the factory floor is where the next advantage gets won.