The CEO of Ciena on how AI is fueling a global subsea cable boom

Under the ocean’s surface lies the true backbone of the internet: an estimated 870,000 miles of submarine cables that ferry over 95% of global intercontinental data traffic across every major ocean. These cables are critical infrastructure, carrying everything from our daily video calls to multinational financial transactions, and they’re facing new challenges as both technology and geopolitics reshape global connectivity.

Few companies are more embedded in that infrastructure than Ciena, the 33-year-old networking firm behind much of the gear that keeps global data flowing. With over $4 billion in annual revenue and a multifaceted customer base, Ciena is now navigating a more turbulent era, one where AI and geopolitics increasingly shape what’s built, and where.

Fast Company spoke with Ciena CEO Gary B. Smith about how subsea networks are adapting to shifting global demands. The conversation has been edited for length and clarity.

What’s changing right now about how subsea cables are being built and used?

Submarine cable has been around for a long time. When you were phoning your grandmother in Australia or wherever, they went through that. And it was mainly the phone companies that buried them, and they did consortia on it, because they were very expensive even then. That’s how that infrastructure started, and then the evolution has been one of the coming of the internet, and then the hyperscalers coming on top of that, requiring global connectivity. 

They’ve all got different business models, for sure, but it all relies on connectivity and getting as many users as possible, and then you’ve got all the AI stuff piling on top of that. Submarine cables are a critical part of that, and they’ve had to scale up over the years.

AI workloads are building this surging demand for bandwidth. How are you modeling the AI-driven capacity needs?

Great question. The challenge in answering it is that no one really knows. 

What we’ve seen is an absolute uptick in cloud traffic across networks connecting these data centers. And Ciena is number one in the world at connecting these data centers; that’s what we do for Meta, Google, Microsoft, Amazon, and Oracle. And we’ve seen an absolute uptick in cloud traffic. How much of that is driven by AI? We don’t know. Clearly, ChatGPT and the inquiry models are driving traffic, but also, most enterprises are also now completely cloud based. So you’ve got this confluence of traffic growth. 

Some of that is absolutely AI, and that’s before really you’ve got a lot of inference going on. There are a couple of exceptions to that. We are seeing for the first time—this is very U.S.-centric—the building out of an AI infrastructure for training and learning as a network. Before, that had always been inside the data center. Now, the GPUs, they can’t get everything they need inside a single data center; we’re seeing that architecture now evolve to where you’ve got dedicated networking for training and learning. And that needs to be very low latency, super high speed, etc.

If you kind of step back from it all, I think the sort of hierarchy of flow here around GPU and power—and those two things are the really defining dimensions of the scaling—I think what’s happened in the last six to nine months, we’ve seen a massive uptick in demand. Because we don’t want to strand all of that data center power, GPU investment, because we didn’t put enough of the network in. And relative to the spend on GPUs, the network is smaller; it’s a few billion. But it’s complicated: You’ve got submarine cables, you’ve got fiber, you’re going into countries like India, which have different regulatory schemes.

How close are we to fully autonomous adaptive networks?

We’re pretty close to it. And you’ve got a lot of automation now on these networks, so we’ll automatically pick up degradation and move traffic, and do those kinds of things. I mean, that we’re doing. You’ve got an opportunity now with agentic AI to take that to the next level. But you’ve got a lot of sophistication now in these networks around things like GeoMesh, where you’ve got multiple routes, so if you have a cable cut, it switches automatically.

You’ve spoken before about prior backlog issues and delivery delays. How’s the supply chain holding up for subsea equipment these days?

The whole supply chain whiplash of 2022 and 2023 had a big impact. That was really a shortage of tiny analog devices, and we were, like everybody else, affected by that. What we’re seeing now in this scale up, we’re able to scale. The amount of demand that we’re seeing continues to increase, so we’re building more and more capacity to be able to deliver on that. There’s not a chip shortage as before. Generally speaking, notwithstanding geopolitical challenges and all the other things going on in the world, the supply chain is pretty agile and robust. Now, we’re fortunate to be able to scale our capacity up. We’re a very focused player; this is all we do, basically. And we’re very vertically integrated as well: We own our own chips, and a lot of the technology around the chips is our stuff. 

In the last nine months, we’ve seen two things. One, the main hyperscaler players really invest in their network. And this is infrastructure: It needs to go under the sea, and needs to be going into all these other countries. So, this is a multiyear investment cycle.

And the second thing we’ve seen is that the list of data center players has gotten much larger. Anthropic, Stargate, xAI, etc.: They’re actually now also investing in the network. But back to what I said before, it’s not “I’m going to build my own network and nothing else.” Lumen is a good example. They have enormous amounts of fiber assets in the United States, and they are working very closely with all of these folks, who will take fiber from them but they’ll also take wavelengths from them, because they want to have multiple sources and resiliency on their network. 

And when you go into different countries, there is a thing called managed optical fiber networks, MOFAN, where they can’t—and don’t want to—become a carrier in India, so they they go to local companies that have the fiber, and they build out essentially a network that they use and that attaches to their global network. We’ve seen an enormous uptick in these managed optical fiber networks. Again, because we are pretty much in all the major carriers around the world, we’re able to help facilitate all of that as well. 

We used to basically just ship the technology to [companies], and they said, “You’ve got great technology, we’ll deal with the rest.” Now, as these networks have gotten massive, it’s become much more of a partnership. We design some of the line systems with them, for example, and we do a lot of co-development and design and architecture. We’ve got a big operation in places like India, so we can do installation and other projects.

Given the rise in geopolitical scrutiny and action around all this infrastructure with that rise, how do you navigate risk in contested regions or routes?

A lot of these submarine cables go through the Middle East, and certain parts of that are a little bit challenging. Everybody’s kind of figuring out how to navigate that. And then you’ve got this decoupling from China, and that’s a challenge from a telecom point of view. But China’s never been a big market for us; one of our major competitors earlier on was Huawei. 

Huawei’s first foray into telecom infrastructure was actually fiber optics in about 2003, when we came back from the sort of telecom nuclear winter where no one wanted to buy anything. We were just crawling out from that first coming of the internet, as I like to think about it, in 2000 and 2001, where all these networks were built out, and no one had any traffic on it. I think that accelerated a lot of these business models, like Amazon, Uber, etc.; you’ve got to have low-cost bandwidth for all that to work, and connectivity over your mobile phone. But when we first started to come out in 2003, we bumped into Huawei. And the economics, we couldn’t figure out. We had the leading edge technology, we were vertically integrated. And, of course, as it turns out, it was massively subsidized. 

It made us think, “Hey, China’s a massive market for our stuff, but is that really where we want to go?” They give you a little bit of that market, but it’s very centrally controlled. And we decided strategically not to enter that China market. Even though it was the fastest-growing market for our stuff in the world, we focused on other areas, submarine cables being one of them.

If you’re looking 10 to 15 years ahead, how are things going to change, and how are you thinking about those changes?

I think people are fundamentally underestimating AI. And I know that sounds a weird thing to say, but other industrial revolutions are basically about automated muscle and practical things; this is about automating intellect. Yeah, the internet has changed a lot of business models and revolutionized the connectivity of the world, but this is going to have a much greater impact than even that. 

So, that’s my starting point. And then I think about—and this is completely self-serving, so understand the source on this—I think we’ve grossly underestimated the amount of traffic and network that’s going to be required to support this. When we talk about AI right now, I mean, it’s largely data. You start talking video, that exponentially changes the amount of capacity you’re talking about by 50 or 100 times. It’s got a lot of scaling up to do to support what’s going to be required.

https://www.fastcompany.com/91368041/ciena-gary-b-smith-interview?partner=rss&utm_source=rss&utm_medium=feed&utm_campaign=rss+fastcompany&utm_content=rss