On May 6, 2026, Nvidia and Corning announced a multi-year strategic collaboration.
Nvidia will invest up to $3.2 billion in Corning. Corning will build three new factories in the U.S., increasing its optical communications manufacturing capacity tenfold and boosting fiber production by more than 50%. (Source: 21st Century Business Herald, May 7, 2026)
The standard media interpretation goes like this: Nvidia is investing in optical communications, upgrading data center infrastructure, the computing giant is securing its supply chain, bullish on opportunities in the optical module industry chain...
But in my opinion, what's most noteworthy this time isn't any of the above.
It's a more fundamental question: Why optics, not copper? Behind this switch from copper cables to optical fibers lies a technological evolution logic that has been ongoing for decades, but most people have never paid attention to it.
To truly understand why Nvidia is willing to spend $3.2 billion, you must grasp this logic—it's not because the optical communications concept is hot, but because it has no other choice.
The common understanding might be: Optical fibers are faster, so switch to fibers.
Actually, it's not that simple.
As always, I'll try to explain it clearly in one article.
What is the Fundamental Difference Between Copper and Optical Fiber?
Let's start with the basics.
Inside data centers, GPUs need to transmit data to each other—this is called "interconnect." Traditionally, this task was handled by copper cables, much like the network cables in our homes.
The problems with copper cables are worsening exponentially as the scale of AI computing power expands:
Problem One: Signal Attenuation. When transmitting data over copper, electrical signals attenuate with increasing distance. This is fine over short distances (tens of centimeters). But as GPU clusters in AI data centers grow larger, with increasing rack and server spacing, copper's attenuation becomes a bottleneck.
Problem Two: Energy Consumption. To combat signal attenuation, copper cables require higher drive currents, meaning more electricity. AI data centers are already among the world's largest energy consumers—Nvidia's data centers alone consume electricity equivalent to a medium-sized city annually. The interconnect network contributes a significant portion of that.
Problem Three: Heat. Current generates heat, which requires cooling systems consuming even more power. It's a vicious cycle.
Optical fibers don't transmit electrical signals; they transmit light. Light doesn't generate heat, isn't susceptible to electromagnetic interference, has minimal attenuation, and travels near the speed of light.
So, the question isn't "Is light better than copper?" but rather "The scale of AI computing power has reached a tipping point where copper can no longer sustain it."
What is CPO? Why is it the Core of Next-Generation Interconnects?
Understanding the problems with copper helps clarify the three development stages of optical communication:
Stage One: Traditional Copper Cable Interconnect (Mainstream in the 2010s)
GPUs are directly connected via copper cables. The advantages are simplicity and low cost; the disadvantages are high energy consumption and severe attenuation. This solution is being phased out as AI computing scale expands.
Stage Two: Pluggable Optical Modules (Current Mainstream)
Replace copper cables with optical fibers, but the optical modules remain "external" next to the GPU chip, requiring optical-electrical (O-E) conversion. Electrical signals travel inside the chip, are converted to light upon exiting, transmitted via fiber, and converted back to electrical signals at the other end. This conversion process itself consumes energy and introduces latency.
Stage Three: CPO—Co-Packaged Optics (Next-Generation Technology)
The optical engine is packaged directly with the GPU chip. The distance electrical signals travel is compressed from "tens of centimeters" to "a few millimeters." O-E conversion happens right next to the chip. Energy consumption is drastically reduced, latency is minimized, and data transmission density is greatly increased.
As Jensen Huang said at Nvidia's GTC conference in 2025: Co-packaged optics technology is a core, essential technology for AI computing infrastructure. (Source: Sina Finance, May 8, 2026)
This isn't just promotional talk for a supplier; it's an engineering-logic-driven assessment.
I refer to these three stages as the "three-step compression" of optical interconnect—from the copper era, to the external optical module era, to the CPO packaging era. The core logic is singular: continuously compress the transmission distance of electrical signals until it becomes negligible.
Why is Nvidia Investing in an Upstream Supplier? What Does This Move Signify?
Nvidia's investment strategy is undergoing a systematic shift—from being a "buyer of components" to "controlling the supply chain."
In January 2026, Meta committed up to $6 billion to help Corning expand its fiber optic cable factory. In May 2026, Nvidia invested up to $3.2 billion to secure Corning's production capacity. (Source: 21st Century Business Herald, May 7, 2026)
This pattern illustrates that optical communications capacity is transitioning from "something readily available in the procurement market" to a "strategic resource."
When the supply growth of a component cannot keep pace with demand growth, those who secure capacity first gain a competitive advantage.
This is precisely what's happening in the fiber optic market today. The price of specialty fiber G.657.A2 has skyrocketed from 32 yuan/core-kilometer to 240 yuan/core-kilometer, a gain of approximately 650%. (Source: East Money Information, May 7, 2026) The reason is simple: demand is far outpacing capacity expansion, and the lead time to build new factories is 18 to 24 months, leaving a supply gap that cannot be filled in the short term.
Nvidia's $3.2 billion investment isn't for products; it's to secure the manufacturing capability that will be the most constrained in the market for the next two to three years.
What Does This Mean for Chinese Optical Communication Companies?
Some worry: After Corning's expansion, the market share of Chinese optical communication companies in North America will be squeezed.
This concern is valid but incomplete.
Here's a counterpoint: It will take 2 to 3 years for Corning's three new factories to reach full production capacity. During this time, the global construction of AI data centers won't wait. A supply gap will persist; demand will still exceed supply.
Institutional estimates indicate a global fiber optic supply gap of 5% to 10% in 2026, potentially widening to 15% in 2027. (Source: Sina Finance, May 8, 2026)
A gap represents a window of opportunity for Chinese companies.
Especially in segments like fiber preforms, optical chips, and optical modules, the mass production capabilities and cost-control prowess of Chinese companies are competitive advantages on a global scale.
In this optical communications arms race, no one will refuse "fast and affordable Chinese suppliers."
This isn't a China-U.S. rivalry; it's a re-division of labor within the global computing power industry chain during a technological upgrade cycle.
Nvidia chose to invest $3.2 billion in securing fiber optics not because optical communications suddenly became trendy, but because copper has reached its physical limits. In every technological revolution, the step taken out of necessity is often the most certain one.
This article is for information sharing and industry analysis only and does not constitute any investment advice, investment analysis opinions, or trading invitations. The market carries risks; investment requires caution. Anyone making investment decisions based on the content of this article shall bear all risks and profits/losses themselves. The author and the publishing platform assume no legal responsibility.
This article is from the WeChat public account "BT Finance" (ID: btcjv1), author: Shi Yan.
