Why Nvidia And Corning’s Fiber Deal Could Change The Game For The AI Boom
Nvidia and Corning have announced a major partnership in which Corning will build three new factories in North Carolina and Texas to supply fiber optic components for Nvidia's AI systems. The deal likely involves co-packaged optics technology, which could replace copper cables inside Nvidia's rack-scale AI systems with glass fiber, dramatically reducing energy consumption and latency. This signals a broader industry shift from copper to fiber as AI data centers scale to hundreds of thousands of GPUs.
Summary
Nvidia and Corning have announced a significant partnership that could reshape the infrastructure of AI data centers. Corning will construct three new factories in North Carolina and Texas dedicated entirely to producing products for Nvidia, with the deal potentially worth tens of billions of dollars. Analysts believe this may mark the commercial debut of co-packaged optics, a technology that would replace the approximately two miles of copper cables found inside Nvidia's rack-scale systems, such as Vera Rubin, with ultra-thin glass fiber — something never before accomplished at scale.
The shift from copper to fiber carries major implications for AI infrastructure. Fiber optic cables transmit data as pulses of light (photons) rather than electrical signals (electrons), making them 5 to 20 times more power-efficient than copper. They also offer lower latency and higher data transmission speeds, which is critical for enabling the hundreds of thousands of GPUs in an AI data center to communicate rapidly and function as a unified system. As power constraints become an increasingly pressing issue for AI data centers, the economic and efficiency case for fiber optics grows stronger.
Corning, which is celebrating its 175th anniversary, has a long history of reinventing itself around pivotal technologies — from glass for Edison's light bulbs to Pyrex, TV screens, COVID vaccine vials, and iPhone glass. Its largest and fastest-growing business segment is now optical communications. The company's CEO, Wendell Weeks, helped invent a new, denser, thinner fiber specifically optimized for AI workloads. Corning is already supplying millions of miles of fiber to major AI players including Meta, OpenAI, AWS, Microsoft, and Google. Meta has committed up to $6 billion for Corning's fiber and is helping fund a major expansion of the world's largest fiber factory.
Industry experts note that as AI nodes grow from sub-100 GPUs to hundreds of GPUs, the internal distances within servers will increase, making fiber optics increasingly economical and power-efficient compared to copper. The Nvidia deal is seen as a strong signal that the transition from copper to glass within AI server infrastructure is accelerating.
Key Insights
- The Nvidia-Corning deal is expected to introduce co-packaged optics at scale, potentially replacing the approximately two miles of copper cables inside Nvidia's rack-scale AI systems like Vera Rubin with glass fiber — something never before accomplished at commercial scale.
- Moving data as photons through fiber optics is between 5 and 20 times more power-efficient than moving electrons through copper, making fiber increasingly attractive as energy constraints become a dominant challenge for AI data centers.
- Corning CEO Wendell Weeks argues that as AI nodes scale from sub-100 GPUs toward hundreds of GPUs, the internal distances within servers will grow, at which point fiber optics become significantly more economical and power-efficient than copper.
- Corning is already supplying millions of miles of fiber to Meta, OpenAI, AWS, Microsoft, and Google, and Meta has committed up to $6 billion for Corning's fiber, funding a major expansion of the world's largest fiber factory.
- Corning's largest and fastest-growing business segment is optical communications, and Wendell Weeks helped invent a new, denser, thinner fiber specifically optimized for AI workloads, signaling deep technical investment in AI infrastructure.
Topics
Full transcript available for MurmurCast members
Sign Up to Access