The old new thing: Cisco and retro tech are back and forming the AI buildout's backbone

By Hannah Pedone

'Everybody forgot about them for 20 years. And now suddenly they're important again.'

Needham analyst Ryan Koontz told MarketWatch: "So many of these old-school... industrial tech telecom companies [are seeing] a total new life in supporting the needs of AI."

Under the ground and below the ocean's surface lie millions of miles of glass strands thinner than a human hair. Data runs through the glass, in the form of pulses of light, at the speed of over 100,000 miles per second.

At the dawn of the internet, a select group of companies made a killing creating the equipment and business model for the system of glass strands bundled in cables, known as a fiber-optic network.

Shares of Cisco Systems (CSCO), which designed the routers and switches that connected the cables, rose over 100,000% from when the company went public in 1990, to their peak in the early 2000s. For a brief period, Cisco became the world's most valuable company by market cap. Other companies that operate fiber-optic networks, like AT&T (T), or that make the glass that runs through the cables, like Corning (GLW), also saw their shares soar at the height of the dot-com bubble.

In March 2000, as the bubble burst, shares of companies in the business of transmitting data, or networking, plummeted too.

But now, the optical fiber that forms the basis of telecom networks is being used for a different purpose: connectivity inside and between data centers, hundreds of miles apart.

As demand for data centers, and the networks that connect them, booms, the companies that manufacture the cables, routers, switches and lasers essential for the connections that make up a fiber-optic network, are clawing back stock-market losses from a quarter-century ago.

"So many of these old-school... industrial tech telecom companies [are seeing] a total new life in supporting the needs of AI," Needham analyst Ryan Koontz told MarketWatch. "Telecom [is] seeing its day in the sun again... and this is the technology that was developed for the AT&Ts and Verizon's (VZ) of the world." It's now "sewing together these different data centers," he said.

In a fiber-optic network, transceivers and optical components help turn data into pulses of laser light of various colors, which are sent through cables. Routers and switches then direct data traffic to different networks, data centers or homes, for instance, Koontz explained.

David Mazza, chief of Roundhill Investments, an investment adviser, told MarketWatch that demand has outrun supply for optical components. Earlier this year, Nvidia (NVDA) invested $4 billion into Lumentum Holdings (LITE) and Coherent (COHR), two foremost optical suppliers to fund laser capacity. "Customers do not fund their vendors' factories unless the constraint is real," Mazza said.

Roundhill launched a photonics-themed ETF last month that traded $76.7 million in volume on its first day in August, according to FactSet. But one of the most prominent market indicators has been Cisco. It took over 25 years for Cisco to win back its gains from the dot-com era, and in December 2025 the stock surpassed its high from March 2000. Since its low in 2002, its market capitalization has more than sextupled and now stands at $429 billion.

Similarly, Corning's market cap peaked at $100 billion in September 2000, plunged, and in February 2026 it finally reached $100 billion again.

"It's like [opticals] did something in the dot-com era and it was huge and then everybody forgot about them for 20 years. And now suddenly they're important again," said Seaport Research analyst Jay Goldberg. "Back in the '90s all these companies built big fiber networks either from city to city or within cities ... Now it's coming back, but instead of connecting internet exchange points it's connecting hyperscaler data-center campuses."

19 million fiber miles

Outside Sacramento, Calif., dozens of workers in yellow vests do stretches before going out to lay a continuous, five-mile-long fiber-optic cable under the city. These crews are supporting a network-building project for Lumen Technologies (LUMN), which has posted videos about their progress on YouTube.

In one video, the workers unwind a line of black tubing, or conduit, from a spool the size of a small car, before feeding it into the ground. They then use specialized compressed-air-jetting equipment to propel fiber-optic cabling through the conduit, which then runs across a river and into the city.

The fiber, historically used for telecoms' connectivity, is being used to connect data centers.

In the late 1990s, Lumen saw a need for networking given the growth of the telecom industry as the internet took off. At the time, the company laid more conduit than was needed, Chris Stansbury, Lumen's CFO, told MarketWatch.

"We have excess capacity, which was actually driven by a bet over 25 years ago, a quarter of a century ago, on the internet," he said.

As of June, Lumen had put down approximately 19 million "fiber miles," which is calculated by multiplying the length of the route by the number of fibers inside each cable. The company said that each cable can contain anywhere from 96 to 1,728 fibers. And now that capacity is finding a new use, as Lumen leases capacity to hyperscalers for AI training.

"It still takes years to deploy a full network, but it's far fewer years than if you were starting from scratch and you actually had to go get permits to go dig the trenches and actually put this stuff in the ground," Stansbury said.

Lumen's stock garnered meme-like status as the artificial-intelligence boom got underway and investors anticipated opportunities for the company around connectivity and fiber optics as a growth sector. Still, while the company has scored a number of deals with hyperscalers and AI companies, those deals haven't yet markedly increased its revenue, noted Ishan Majumdar, an analyst at Baptista Research, a financial advisory.

Lumen has signed $13 billion of deals cumulatively with Microsoft (MSFT), Meta Platforms (META), Amazon Web Services (AMZN), Google Cloud (GOOGL) (GOOG), Anthropic and additional undisclosed customers. But the company's revenue has fallen by billions since AI began to proliferate. The company reported $12.4 billion in revenue for the 2025 fiscal year ending December 2025, down from $19.7 billion in the 2021 fiscal year ended December 2021.

Lumen's case serves as "evidence that AI can revalue an old asset before it meaningfully reaccelerates the underlying company," Majumdar wrote in an email. "The $13bn validates demand; whether it produces a durable financial comeback is still being tested," he added.

Yet Lumen is still banking on the market for data-center networking becoming more lucrative. "We've been planning out our routes to be able to cover where we believe most of the AI traffic [will be]," Jim Fowler, Lumen's chief technology and product officer, told MarketWatch. "We know that by 2030 there's gonna be a 10x increase in the number of data centers across the United States, and we know where almost every one of those data centers are."

'Dot-com poster child'

John Chambers was the CEO of Cisco from 1995 to 2015, presiding over the company's rise and fall.

Cisco has become a prominent example of an old-school tech company that has rapidly found a role to play in the AI frenzy. It makes networking hardware and telecom equipment like routers and switches, which are used in telecommunications networks as well as in data centers.

A router is a box that routes data traffic between different networks, and to the internet. Meanwhile, a switch connects computers and servers to create a network. The routers and switches in a telecom network are the same underlying technology as those in a data-center network, though the speeds and volume of traffic are drastically different.

Kevin Wollenweber, Cisco's general manager of data-center and internet infrastructure, started as an intern at Cisco nearly three decades ago and for much of that period the company mostly sold networking systems that served as a core backbone for telecommunications companies. Roughly 10 years ago, Wollenweber said, the company figured out it could sell new and bigger systems to rapidly growing cloud businesses of the Big Tech companies.

Google, Amazon and Microsoft "kind of came out of nowhere" he said, and they "started building out these massive, massive data centers for cloud." They demanded different systems than what the company had built for service providers, due to higher capacity demands, he said. Cisco developed partnerships with hyperscalers to build systems for their data centers and started investing in optical technologies and chips.

"We also went to all of the large hyperscalers and said, 'not only are we investing in these technologies and we will be competitive in the market, but we'll sell it to you however you want it,'" Wollenweber explained.

Cisco now sells custom-designed networking chips as raw components, as well as complete systems and switches and routers with chips built in.

But Seaport's Goldberg argues that Cisco's shift to design and sell systems for cloud providers came late.

Cisco is the "ultimate dot-com poster child trying to claim relevance in today's bubble," Goldberg said. "They missed the shift to the cloud and now they are trying to make up for lost time by claiming relevance in AI."

Since 2022, the year OpenAI released ChatGPT, Cisco's annual revenue has grown roughly 23%, to $63.3 billion in 2026.

Morningstar analyst William Kerwin told MarketWatch that Cisco is seeing AI growth supplement its core business. "But its core campus market is also seeing higher growth, as enterprise networking bandwidth demands rise from AI needs," he said.

"I think it's less that old equipment is now more in vogue, and more that formerly 'sleepy' low-growth end markets are now seeing higher demand," he said, referring to networking systems Cisco originally made for telecoms.

Chokepoints of the AI supply chain

At a plant in Sherman, Texas, roughly the size of a football field, Cisco competitor Coherent makes lasers for iPhones - and for data centers.

According to the company, the facility is home to the largest-volume production site in the world for the manufacturing of 6-inch indium phosphide - a semiconductor material used in lasers - one of the most critical chokepoints currently in the AI supply chain.

The material is essential for turning electrical signals into laser light. The indium phosphide-based lasers are used inside optical transceivers, which connect networking equipment and transmit data across data centers. With demand for its data-center components surging, Coherent reported revenue of $7.1 billion in the fiscal year ending June 2026, up 114% from its fiscal year ending June 2022, the year the AI frenzy began. Lumentum, a Coherent competitor, generated $3 billion of revenue in the year ending June 2026, up 76% from its fiscal year ending June 2022.

"Wherever you look in the data center, between the racks, within the rack, data center to data center...we have the entire portfolio of technologies," Sanjai Parthasarathi, chief marketing officer of Coherent, told MarketWatch.

The company, which originally focused on photonics for industrial applications and eventually moved into telecommunications, is now focusing on AI.

And with a mix of Nvidia's funding, along with a $50 million federal grant from the U.S. Chips and Science Act, among other funding sources, Coherent plans to double the size and quadruple the production capacity of the site in Sherman.

Coherent reported that 79% of its quarterly revenue came from its data-center and communications businesses for its June quarter, which Baptista Research's Majumdar said is a case in point that AI has created a "massive need to move data faster and more efficiently between chips, servers and data centers."

"What is really happening is that some of these companies spent decades building things like fiber networks, optical technology, networking expertise and manufacturing capacity," he said. "A lot of that infrastructure was not especially exciting to investors for years," he added.

But now, there's been a "revaluation of infrastructure that is difficult and expensive to replicate," he said.

Jeff Hecht, a science and technology writer and author of "City of Light: The Story of Fiber Optics," explained to MarketWatch: "Corning will supply the glass, and other people [like Coherent and Lumentum] would supply the lasers, and then Cisco provides the software and the transmission electronics, and some place like Lumen will pick up on other things."

The AI supply chain is effectively a "whole communication system that's built around glass fibers and lasers that are producing the light to go through the glass fibers," he said.

-Hannah Pedone

This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.