On the afternoon of May 6th, NVIDIA announced an investment. The amount isn’t particularly large, $500 million. But the contract clearly states that it can be increased to $3.2 billion in the future. Corning’s stock price rose 14% on that day.

What’s more intriguing is the structure of this deal. Among the 18 million shares of equity certificates given to NVIDIA, 3 million shares have a conversion price of $0.0001. This means that these 3 million shares are almost given to Corning for free. On the same afternoon, Corning raised its revenue growth target for 2030 to $40 billion at its investor conference in New York.

But none of this is the most abnormal part of Corning in recent months. In its first-quarter financial report, the “iPhone screen glass supplier” stated that in recent months, two unnamed companies had signed multi-year contracts worth $6 billion with Corning. The reason for saying “again” is because Corning recently signed a similar-scale contract with Meta.

If you count, you’ll find that in the past four months, at least four multi-billion-dollar AI deals have been concentrated on this 174-year-old glass company. Over the past six months, Corning’s stock price has increased by 140%, and compared to two years ago, it has already quintupled.

From selling phone glass to becoming a hot commodity in AI factories

If you’re reading this on your phone, chances are the screen you’re looking at is made by Corning. Since Apple’s first iPhone in 2007, Gorilla Glass has almost become the default option for high-end smartphone screens worldwide. But “smartphone glass supplier” is just one aspect of Corning, and not the most profitable one.

Corning Gorilla Glass production line, source: Apple

This company was founded in 1851. It made the first incandescent bulb glass shell for Edison, and in the 1970s, it invented low-loss optical fiber from scratch, pioneering the entire modern fiber optic industry. The glass for that 2007 iPhone was its third major business transformation. Today, Corning is undergoing its fourth transformation, with optical communications becoming the real business engine.

Corning’s optical communications business has a history of over 50 years, but the customer structure of this business has undergone a thorough reversal in the past two years.

For a long time, Corning’s fiber optics were mainly sold to telecom operators like AT&T and Verizon. They used it to lay fiber to homes and build 4G and 5G base stations. In 2009, Corning launched a data center cabling solution called EDGE, officially bringing data center operators into its customer list. Over the past decade or so, with the explosive growth of mobile internet, cloud services, and remote work during the pandemic, Corning’s optical communications business steadily grew, but it was never the main source of revenue.

In November 2022, OpenAI launched ChatGPT to the public. From that moment, data centers worldwide began redesigning their physical infrastructure for new AI training workloads. The fiber density required for AI training is unprecedented in any previous era.

The earliest sign appeared in August 2024. A US telecom operator called Lumen booked 10% of Corning’s global fiber capacity for two consecutive years. This was the earliest public signal of Corning’s transition into the AI field.

By early 2026, the four previously mentioned $6 billion contracts exploded simultaneously. Corning had worked with data center operators for 15 years, but the “secondary customers” became “main players” only in the past 24 months.

The direct effect of this customer shift is reflected in Corning’s financial reports. In 2023, Corning’s full-year revenue declined 11% year-over-year, marking a low point for the industry, but by 2025, annual revenue surged to $15.6 billion, a 19% increase. In the first quarter of this year, revenue grew another 18% year-over-year. The most impressive growth was in optical communications, which increased 35% for the full year. The proportion of optical communications in total revenue rose from 30% in 2020 to 37% in 2025. The absolute value changed more visibly, from $2 billion five years ago to $6.3 billion in 2025, more than tripling.

This rise from a “secondary business” to a “main driver” is no accident, and behind it is a growth plan led by CEO Wendell Weeks. This plan has an internal code name, called Springboard, meaning “springboard.”

Two years ago, Corning was still regarded by Wall Street analysts as a “boring glass manufacturer,” classified as a mature, low-growth dividend stock. But after three years of implementing the Springboard plan, Corning’s stock price rose from just over $30 in early 2024 to $162, a fivefold increase in two years, with a 140% jump in the past six months alone. The glass factory has transformed into the “neural system of the AI revolution.”

Springboard was first announced in September 2024. Its starting point was the annualized revenue level in Q4 2023, about $13 billion. The initial goal was to increase annualized revenue by over $3 billion by the end of 2026, with an overall operating profit margin reaching 20%.

But over the next year and a half, this target was raised three times, reaching over $6.5 billion, pushing the annualized revenue for 2026 to $20 billion. On May 6th, after NVIDIA invested in Corning, the company directly raised its internal revenue target for 2030 to $40 billion. Meanwhile, Corning had already achieved a 20% profit margin one year ahead of schedule in Q4 2025.

The key to the Springboard plan is “premium.” The company’s sales grew 18%, but earnings per share increased 46%, with profit growth 2.5 times that of sales. On the business level, Corning mainly did three specific things:

First, raising prices on old businesses. Corning’s display glass has been a mature business with no growth for years. But by the end of 2024, Corning increased prices on this line by over 10%, while locking in the yen exchange rate until 2030. As a result, even in a yen depreciation environment, this line consistently contributed net profits of $900 million to $950 million annually, with a net profit margin of 25%.

Second, upgrading optical communication products. In 2025, optical communication sales increased by 35%, but net profit surged 71%. This means not only more sales, but also higher profit per fiber.

Third, activating idle capacity. Corning did not build large new factories but restarted idle capacity from previous downturn cycles, raising overall gross margin from 33% in 2024 to 36% in 2025.

Of course, the ability to raise prices depends on willingness to pay. The profitability of product upgrades depends on customers willing to pay more for upgraded products. The reason Springboard allows Corning’s profit growth to outpace revenue is fundamentally because its customer structure now includes a group willing to pay premiums.

Everyone is racing for fiber optics

The AI race and order demands make every data center operator extremely anxious about time.

Cloud giants’ core business has always been “leasing IT to enterprises.” New companies like Netflix, Airbnb, Uber, which grew with mobile internet, mostly generate “north-south” traffic. A user opens an app, requests data from cloud servers, and the server returns data. Servers occasionally communicate with each other, but the volume and frequency are low. This network structure does not require demanding physical infrastructure: Ethernet is enough, copper cables are enough, ordinary fiber is enough. Cloud giants have used this architecture for over a decade, stable, reliable, and profitable.

Until ChatGPT appeared, and the game changed.

In the following years, almost all cloud giants began training their own models. Microsoft is the largest compute provider for OpenAI, AWS is deeply tied to Anthropic, and Alibaba trains Tongyi. The core business of cloud giants has shifted from “leasing IT to enterprises” to “training AI for the world.”

But this physical infrastructure shift triggers chain reactions beyond all previous knowledge accumulated over the past 20 years.

AI training traffic is characterized by “east-west” flow. Training a large model may require tens of thousands of GPUs communicating simultaneously, synchronizing gradients. Any slow link causes the entire training to wait, turning tens of thousands of GPUs into “stalled cars.” Therefore, east-west traffic demands latency and bandwidth that are dozens of times higher than traditional north-south traffic.

Previously, most high-speed internal connections in data centers were copper cables. Copper is cheap, easy to install, and stable—default for data centers. But the geometric structure of AI training clusters is exactly what copper hates. Tens of thousands of GPUs distributed across dozens of racks, often tens of meters apart, cannot be connected with copper. Fiber optics, on the other hand, have no distance limit.

Suddenly, the sparse networks that sufficed before are no longer enough. Cloud giants need to re-lay fiber optics, more densely than ever before.

This scale of re-laying is reflected in their capital expenditures. By 2026, the combined capital expenditure of the six largest cloud giants is expected to exceed $600 billion. The number of ultra-large data centers in operation worldwide has reached 1,297, nearly three times the figure at the start of 2018. In 2026 alone, over 150 new data centers are expected to be built, with AI infrastructure spending exceeding $400 billion.

Market research estimates that AI clusters’ total fiber demand is 10 to 100 times that of traditional cloud services. This is the fundamental reason why Corning can sign four $6 billion deals now.

Between data centers, between racks, all fiber must pass through something called cable conduits. Usually plastic or metal pipes with 2 to 4 inches in diameter, buried underground or running along racks. These conduits are difficult to add to once laid: expanding a city’s pipeline requires reapplying for road rights and digging up roads, taking years. Adding a conduit in an active data center means downtime and renovation, measured in months.

Upcoming underground cable conduit, source: internet

In recent years, Corning has focused on enabling existing conduits to hold more fiber without adding new pipes, especially for AI data centers.

Besides making fibers thinner, Corning changed the arrangement from loose “spaghetti” style to a rollable flat ribbon, which can be flattened when used and rolled up when not. This design allows more fibers to be packed into the same 2-inch conduit—over three times as many, from about 1,000 to over 3,000. Using a 4-inch conduit with six such cables side by side can hold over 20,000 fibers, six times the traditional capacity.

Corning’s rollable fiber, source: Corning

Not only more fibers, but also easier terminations. A 3,456-core cable traditionally takes over 200 hours to terminate, but Corning’s ribbon design reduces this to under 40 hours, saving about 30% in preparation time. Considering the US already faces a shortage of optical communication engineers, this efficiency is significant.

In building a large AI factory, each month of delay means massive GPU depreciation and training postponements, costing hundreds of millions of dollars. Products that can cut months and millions in engineering costs are extremely cost-effective when adding a 30% to 70% premium on fiber.

Huang Renxun’s “Unprecedented Scale”

On May 8th, NVIDIA CEO Jensen Huang emphasized in an interview that the next-generation AI infrastructure requires massive optical connectivity, and copper lines can no longer meet the demand. He also said NVIDIA will expand the application of optical technology on an unprecedented scale.

The recent details of NVIDIA’s investment in Corning clearly reflect this “unprecedented scale.” Among the 18 million shares of equity certificates, 3 million are “gifted.” This structure is rare in NVIDIA’s ecosystem investments over the past year, meaning NVIDIA immediately gained significant equity exposure in Corning without using cash—more like a long-term partnership signing fee.

Corning is not NVIDIA’s only strategic move. Since September last year, NVIDIA has adopted a new investment rhythm: larger scale, and increasingly using financial instruments like “framework agreements,” “options,” and “prepaid warrants” to lock commitments and disburse funds gradually. Besides the $100 billion framework for OpenAI, NVIDIA has also invested tens to hundreds of billions of dollars in AI infrastructure companies like Anthropic, Intel, and CoreWeave.

The most overlooked part is its investment in optical communications. Besides Corning, NVIDIA has also invested $2 billion each in Lumentum and Coherent, two of the world’s largest optical device companies. Including Corning’s initial $500 million and the $3.2 billion in options, NVIDIA has poured about $7.7 billion into optical communications alone.

If you lay out this investment list, it looks exactly like a blueprint for building an AI factory: computing power, network, optics, power, cooling, software, customers, models—each layer locked with at least one key supplier. At this year’s GTC conference, NVIDIA released a comprehensive design diagram, including a hardware reference architecture called Vera Rubin DSX and a digital twin scheme called Omniverse DSX Blueprint, which are essentially “AI factory construction blueprints.”

A GW-scale (powering 1 million households) AI factory from planning to commissioning takes 18 to 24 months, involving over 100 suppliers. In the past, data center operators did this themselves, repeatedly verifying interfaces. Now, NVIDIA’s Omniverse DSX systematizes this process, with all partners’ products tested in NVIDIA’s digital twin, parameters aligned, interfaces standardized, so cloud giants can directly buy according to NVIDIA’s blueprint.

Jensen Huang unveils the AI factory blueprint platform at the 2026 GTC, source: NVIDIA

This marks a key step for NVIDIA’s transformation from a chipmaker to an “AI factory general contractor.” With increased integration and higher gross margins, even if AMD or Broadcom produce equivalent GPUs tomorrow, replicating NVIDIA’s supply chain coordination from chips to fiber to power grids would still take years.

Therefore, NVIDIA’s $3.2 billion options to Corning essentially lock in a key player for “localizing optical communication capacity” within its own AI factory blueprint. Only NVIDIA currently has the capability to draw this comprehensive plan.

Click to learn about Lu Dong BlockBeats’ job openings

Join the official Lu Dong BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Twitter Official Account: https://twitter.com/BlockBeatsAsia