Meta has found a new and unlikely partner as it races to build out the vast data center footprint needed to compete in artificial intelligence: Corning, a 175-year-old American glassmaker whose technology now sits at the heart of the AI infrastructure boom.
Meta has committed to paying Corning as much as $6 billion through 2030 for fiber-optic cable to wire its next generation of AI data centers, Corning CEO Wendell Weeks told CNBC. The agreement underscores how the AI arms race is no longer just about advanced chips and software models, but about the physical plumbing that connects, powers, and cools them.
The deal comes as Meta accelerates one of the most aggressive infrastructure expansions in corporate history. After unsettling investors in 2025 with soaring AI spending and limited clarity on near-term returns, the company pledged to invest as much as $600 billion in the United States by 2028 on data centers and related infrastructure. Of the roughly 30 facilities Meta plans to build, 26 will be located in the U.S., reflecting a broader push by Big Tech to secure domestic supply chains for strategically sensitive technologies.
“We want to have a domestic supply chain that’s available to support that,” said Joel Kaplan, Meta’s chief global affairs officer, pointing to rising geopolitical concerns and fears that China could outpace the U.S. in AI if investment and policy decisions fall short.
Two of Meta’s largest projects—the one-gigawatt Prometheus data center in New Albany, Ohio, and the massive five-gigawatt Hyperion facility in Richland Parish, Louisiana—will both rely on Corning’s fiber under the new agreement. The Louisiana site alone is expected to require around 8 million miles of optical fiber, highlighting the staggering scale of modern AI infrastructure.
The Meta deal is believed to be emblematic of Corning’s transformation years in the making. Once synonymous with boom-and-bust cycles during the dot-com era, the company has emerged as one of the quiet beneficiaries of the AI buildout. Its optical communications unit is now its largest and fastest-growing business, with revenue in the segment jumping 33% in the third quarter to $1.65 billion. Enterprise optical sales surged 58%, driven by demand tied directly to generative AI deployments.
Shares of Corning have risen more than 75% over the past year, a sharp contrast to its painful history two decades ago. During the late-1990s internet boom, demand for fiber propelled Corning’s stock nearly eightfold, only for it to lose more than 90% of its value after the bubble burst. Weeks says that experience continues to shape the company’s approach.
“What we learned then was that it wasn’t enough to do great innovations,” he said, emphasizing the need for diversified, cash-generating businesses that can absorb cyclical shocks. Corning still sells glass for smartphones, automobiles, televisions, and pharmaceutical vials, providing a buffer if AI-driven demand eventually cools.
Skepticism about the current AI spending wave is growing. Industry announcements in 2025 alone tallied more than $1 trillion in planned compute investments, prompting warnings from some analysts that a new bubble could be forming. Weeks, however, argues that fiber demand has historically grown at about 7% annually and that excess capacity, if it emerges, will ultimately find productive use.
He also voiced confidence in Meta’s long-term prospects, saying that technical execution and a willingness to commit capital remain decisive advantages in AI.
“Compute matters,” he said.
The technological case for fiber is central to Corning’s optimism. Unlike copper cables, which transmit data as electrical signals, fiber-optic cables send information as pulses of light through strands of glass, using far less energy and achieving much higher speeds. As power constraints become one of the defining bottlenecks for AI data centers, that efficiency gap is becoming harder to ignore.
“Moving photons is between five and 20 times lower power usage than moving electrons,” Weeks said.
As a result, fiber is being pushed closer and closer to the compute layer inside data centers.
AI workloads are also structurally different from traditional cloud computing. Training and running large language models requires dense, high-bandwidth connections between thousands of chips, creating what Weeks described as a “whole separate network” that mirrors neural connections in the human brain.
To meet that need, Corning developed a new fiber product called Contour, designed specifically for AI. The cable doubles the number of fiber strands that can fit in a standard conduit and dramatically reduces connector complexity.
Development of these AI-specific products began more than five years ago, well before ChatGPT’s public debut, after conversations with early leaders in generative AI who warned that compute requirements would scale far beyond existing assumptions. Today, Corning says it has manufactured more than 1.3 billion miles of optical fiber, yet demand continues to outpace supply.
The next frontier may be inside the servers themselves. Copper still dominates connections within server racks, including those housing Nvidia’s graphics processors. But as racks begin to hold hundreds of GPUs, Weeks says the shift to fiber is unavoidable on cost and energy grounds.
At that scale, he said, “fiber optics become much more economical and much more power efficient.”
However, the Corning partnership highlights how Meta’s AI ambitions extend deep into the industrial supply chain, tying Silicon Valley’s future to factories in places like North Carolina, Ohio, and Louisiana. For Corning, it marks another reinvention—one that positions a 19th-century glassmaker as a foundational supplier to the AI economy of the 21st century.
