Is Nokia the new Cisco?

Nokia’s AI Networking Lab announcement is a bid to make its “validated” transport designs the new standard for AI-enhanced networking.

Keeping track of the AI industry is a little like watching a hockey game. If you glance at the scoreboard for half a second, you might miss Sidney Crosby scooping the puck off the ice from behind the net and tucking it into the top corner past a helpless goalie. That highlight is all you remember, but it only exists because of the things you barely noticed. Spectacular goals in hockey depend on passing lanes, positioning, timing, and, above all, the smooth sheet of ice beneath the players that is the foundation that decides which plays are even possible.

Something like that may have happened on May 20, 2026, when Nokia announced the launch of an AI Networking Lab in Sunnyvale, California, with partners including AMD, Lenovo, and Supermicro. Nokia and its partners appear to be working on the hidden networking fabric required to keep massive AI systems from choking on their own data.

This is not an article meant to pump Nokia stock. It is an article about the hidden systems forming beneath the AI boom—and why networking may become just as important as agents, chips, and compute.

Disclosure: I hold a small personal position in Nokia, but I have no professional relationship with the company and received no compensation for this piece; my aim here is to understand what its networking lab reveals about the future of AI, not to offer investment advice or a recommendation to buy or sell any security.

Because…

AI systems are not simply powerful chips running in isolation. They depend just as heavily on what’s happening unnoticed behind the action.

They are sprawling physical systems made up of processors, memory, routers, fiber-optic cables, switches, cooling equipment, and power infrastructure—all working together at tremendous scale and speed. In fact, the new optical networking systems can push data at speeds measured in hundreds of terabits per second. Training and running large AI models means constantly moving substantial amounts of data between thousands of processors at once. If the network slows, expensive hardware sits idle. And that costs money.

This scenario helps explain why Nokia’s new AI Networking Lab in Sunnyvale may matter more than it initially appeared to. The company is not trying to build the next frontier AI model. It operates farther down the AI stack—in the hidden networking layer that connects the rest of the system.

Think of the physical AI stack as five rough layers:

A close‑enough AI stack illustration

· AI models and products—OpenAI, Anthropic, and Google; the layer users see.

· Compute—NVIDIA, AMD, and Cerebras, where raw training and inference horsepower lives.

· Memory and storage—Micron, Samsung, SK Hynix; the layer that determines how much data stays close to the chips and how fast it can move.

· Networking and optics—Nokia, Ciena, Corning, and Lumentum; the layer that moves information between processors, manages congestion, and keeps large systems synchronized. Nokia’s lab sits in the transport layer—the part of the stack that decides whether bits arrive on time—and is trying to push some of its assumptions up into the control layer as “validated” blueprints.

· Power and cooling—Vertiv, GE Vernova, Schneider Electric; plus GaN and SiC power‑conversion chip makers like Navitas that decide how efficiently power, and lots of it, gets from the wall into useful compute.

Most public attention focuses on the top of this stack, where the visible AI products and headlines live: Microsoft, Meta, Google, Nvidia. But each layer depends on the physical systems beneath it. GPUs require memory. Memory requires networking fast enough to move large amounts of data between processors. The entire system depends on power, cooling, and increasingly dense webs of fiber-optic connections.

As AI systems grow larger, the lower layers of the stack, like networking, become more important. The challenge is no longer simply building smarter models and better algorithms. It keeps enormous machine systems working together to keep data moving.

Fun Fact: Most of this networking still runs on some version of Ethernet—the same broad networking standard used across much of the modern internet and enterprise world. But AI workloads are pushing Ethernet into unfamiliar territory as companies race to move enormous amounts of data between thousands of processors with minimal delay. Specialized networking systems, such as NVIDIA’s InfiniBand, also compete in this space, underscoring how central networking has become to the future of AI.

So what is the purpose of Nokia’s AI Innovation Lab? Indeed, what does Nokia get out of organizing a tech‑bro playdate? Nokia’s own press language talks about “technology innovation,” “ecosystem collaboration,” and “real‑world validation,” but that undersells what a serious interoperability lab does. A lab like this exists to break networks on purpose—in topology design, traffic management, and congestion control—until you know where they fail. It is where engineers observe real GPU communication patterns, stress-test optical systems, and shave microseconds off latency so that networks transporting AI workloads behave like a single, coherent machine rather than a pile of parts.

Nokia is unusually well-suited to run that kind of torture chamber. Long after its phone business faded from view with the introduction of the iPhone, the company kept absorbing pieces of Bell Labs and the old telecom research world. Between its R&D and the Bell Labs legacy it acquired with Alcatel‑Lucent, Nokia controls one of the biggest telecom patent portfolios on the planet: more than 20,000 patent families and tens of thousands of individual patents in networking and wireless systems.

Nokia inherited a culture built around reliability, queuing theory, and physical networks that are not allowed to go down. It deploys 5G worldwide and leads early efforts in 6G systems, tightly coordinating radios, fiber, and compute. Nokia has even run a small “network in a box” on the Moon under NASA’s Tipping Point program—an extreme-edge rehearsal for making systems in hostile environments behave like normal infrastructure on the first try—think data centers in space, Mr. Musk.

If you wanted a company to turn AI networking from a messy engineering problem into a set of constraints the rest of the industry could quietly live inside, Nokia is not a bad candidate. Instead of copper and fiber as simple physical conduits for moving data, Nokia wants to build AI-enhanced “networks that sense, think, and act.”

To be sure, Nokia is not the only player chasing this network validation layer—Cisco, Arista, Broadcom, NVIDIA, Juniper, and others all want a say in how the AI fabric is wired—but the more compelling story is the contest itself. It is a fight over who gets to define what a “real” AI network looks like and, therefore, what kinds of systems are even thinkable.

But we have seen this movie before in other industries. Rail systems only scaled once countries agreed on track gauges; electrical grids needed standard voltages and plug shapes; the internet needed TCP/IP; global trade needed standardized containers; cloud architecture required a few shared assumptions about APIs and compatibility. Each began as a contingent technical choice that eventually sank into the background as a hidden structure—constraints everyone had to live inside. In every computing era, many firms sold into the structural layer, but a few—Cisco with its certifications and reference designs, VMware with virtualization, Intel with its “Intel Inside” architectures, and NVIDIA with CUDA—ended up defining what “real” looked like. Nokia’s AI Networking Lab is a conjunctural move aimed at the same destination for the physical and optical layer of AI: to turn “Nokia Validated Designs” into one of those quiet blueprints the rest of the industry measures itself against.

In hockey terms, Nokia is not trying to be the star forward; it is trying to be the ice—defining the surface everyone else must skate on.

The Innovation Lab, in other words, is not a simple hardware playground. It is Nokia embedding its networking IP directly into the data center fabric, allowing the network itself to sense, think, and act to maintain and coordinate network throughput. By opening this lab, Nokia is setting itself up as a testing ground to prove that its software can orchestrate a chaotic, multi‑vendor environment under real‑world pressure. They are turning their theoretical network architecture into Nokia Validated Designs—concrete blueprints designed to prevent a multi-trillion-dollar AI build-out from collapsing under its own weight. The “validation” comes from partners who serve as both proof of concept and a map for future network products and systems. If a design can survive Nokia’s lab, other vendors will trust the blueprint.

Most importantly, Nokia’s Lab illustrates the growing reach of AI as a constitutive technology becoming embedded across multiple societal and industry domains while changing the conditions for action. AI is not just changing chips, connectors, software, energy regimes, and data centers at the edge of the map. It is slowly transforming what “real” networking will become.