Nvidia and Corning just announced a massive, multiyear deal to expand U.S. manufacturing for new optical fiber technology, a move that could dramatically improve the energy efficiency of AI data centers. As part of the partnership, Corning will build three new factories in North Carolina and Texas entirely devoted to products for Nvidia, creating 3,000 new jobs. While the companies did not share many specifics, the deal is likely worth tens of billions of dollars, according to a CNBC report by Katie Tarasov.

The centerpiece of the agreement is a technology called co-packaged optics. It has been discussed for years but never deployed at scale. Co-packaged optics replaces traditional copper cables with tiny glass fibers that carry data using light rather than electricity. Inside Nvidia's rack-scale systems, like its Vera Rubin architecture, the shift could be dramatic. Currently, those systems use about 5,000 copper cables totaling roughly two miles in length. Glass fibers would take up far less physical space and, more importantly, consume significantly less power.

Copper cables are a major source of energy loss in large-scale computing. Every signal sent through copper generates heat, requiring additional cooling and more electricity. Glass fiber, by contrast, transmits data with minimal resistance and almost no heat loss. In an AI data center where thousands of GPUs are linked together and running near-continuously, the energy savings from switching to optics could be immense. Scaling AI models has already created an enormous demand for power, and data center operators are looking for any advantage that reduces electricity consumption and cooling costs.

The deal also has implications for U.S. manufacturing. The three new Corning factories will be located in North Carolina and Texas, states that already have strong semiconductor and telecom supply chains. The 3,000 jobs will include roles in advanced manufacturing, optical engineering, and logistics. This is part of a broader trend where major tech companies are bringing production closer to their customers and reducing reliance on overseas supply chains. Nvidia has been investing heavily in domestic manufacturing for AI hardware, and this fiber deal aligns with that strategy.

Corning has long been a leader in glass and optical technologies. The company produces the glass for smartphone screens, Gorilla Glass, and the optical fiber that underpins much of the world's internet backbone. But this deal is different because it is a dedicated, multiyear commitment to produce a specific new type of fiber optic component designed explicitly for Nvidia's AI infrastructure. The partnership suggests that Nvidia expects co-packaged optics to become a standard part of its future data center designs.

What exactly are co-packaged optics? The term refers to integrating optical transceivers directly into the same package as the silicon chips that process data. In a traditional data center, data travels from a GPU to a separate optical module, which converts the electrical signal to light, sends it over a fiber, and then converts it back to electricity on the other end. That conversion step consumes power and creates latency. In a co-packaged system, the optical engine sits right next to the GPU, eliminating the electrical-to-optical conversion step. The data leaves the chip already in the form of light, traveling over fiber with far less energy loss.

The challenge has been manufacturing these integrated modules at scale. Optical components require precise alignment and clean-room conditions. Corning's new factories will be designed to churn out these components in volumes that match Nvidia's massive GPU production. If co-packaged optics succeed, they could become the standard for AI clusters, not just for Nvidia but for other chipmakers and cloud providers.

Replacing two miles of copper cable with a few ounces of glass sounds like a futuristic vision, but the deal between Nvidia and Corning suggests the future has arrived. For years, analysts have predicted that optical interconnects would eventually replace copper in high-performance computing. The limiting factor was always cost and manufacturability. Now, with Nvidia willing to commit to a multiyear, multibillion-dollar purchase, Corning has the incentive to build the factories and refine the production processes.

The timing matters. AI models are growing larger, and training runs now take weeks even on tens of thousands of GPUs. Data movement between GPUs is a bottleneck. Copper cables struggle to keep up with the bandwidth demands of modern AI workloads, and pushing more data through copper only increases energy consumption. Optical links solve both problems: higher bandwidth at lower power.

There are still open questions. The companies have not disclosed specifics about the optical fiber's performance characteristics, such as data rates or exact power savings. They also have not provided a timeline for when the factories will be operational or when co-packaged optics will start appearing in Nvidia's systems. The deal was announced as a multiyear commitment, which suggests the transition will be gradual.

For now, the announcement signals a significant shift in how AI hardware will be built. Nvidia has already dominated the GPU market for AI training and inference. By investing in co-packaged optics, the company is trying to solve one of the biggest infrastructure problems facing data center operators: power. Every watt saved in the interconnect means a watt available for computation, and in a world where AI datacenters can consume as much electricity as a small city, efficiency gains at this scale matter.

Corning, for its part, is positioning itself as an essential supplier for the AI boom, not just a passive component vendor. The three new factories represent a major bet that optical technology will be at the center of the next wave of computing hardware. If co-packaged optics deliver on their promise, the deal may be remembered as the moment when data centers finally went all-optical.