Nokia’s James Watt discusses technologies addressing hyperscalers and data centers networks at OFC50

Hi, thanks for having me, Phil. Great to be here.

I think the biggest single impact is scale in a bunch of dimensions. Obviously the business is larger, which has its own qualities. The set of great people we have on a whole bunch of different topics—whether product, whether project execution, whether technical pre-sales—is now much, much bigger. And more importantly, we've got a broader portfolio to really bring a much wider set of solutions to people, not just today but increasingly quickly as we move on. So scale, scale, and more scale.

So if we take that broadly, there are a couple pieces of that answer. The first is the data center fabric itself, which is something we've been talking about for the last few quarters, where we've significantly bulked up our investment in that space, delivering a wide range of solutions on that data center fabric—both the switches themselves but also the automation to make that work. The other side of that equation, which is really part of a discussion dominated by power, is the interconnect between that fabric, and that's somewhere where we believe we would apply the expanded technology portfolio to be part of solving that power problem as the speeds go up. And so I think it's really the combination of those two for Nokia, and the part obviously the acquisition brings is the second part.

Yeah. And if you look at some of those pieces, data center interconnect is something we've been doing—both the classic Nokia, but also Infinera—for a number of years, but increasingly so. Data center gateways as well on the IP routing side, the fabric we have been doing but that's somewhere where we ramped up our efforts. And then the interconnect—so the data centers themselves are changing, there's a lot of evolution there, putting a lot of pressure on how you fill them up. And obviously that's not only us, there's GPUs and other things that go in. But how do you connect those things together, and then how we connect those data centers both to themselves but also to all the other networks in the world—and that even trickles through to how those networks evolve.

Yeah. So the line system—you framed it well because there's really two parts to the optical networking, right? The WAN part of it, if you will. One is the photonic infrastructure, which is really about being able to get the wavelengths to where you need them, and enough of them or as many of them depending on the topic. And then, how do you get light over them, which we'll come back to in a sec. Line systems are under pressure in different parts. Certainly interconnecting data centers—either interconnecting campuses, sets of them, or whatever—the push there is capacity. We've been doing C+L really for the last 8 to 10 years, and now we're actually even working beyond that. But the other pressure that the growth is bringing is a simplification of that infrastructure. So if you look in the booth, you'll see a basically a very high port count 66-port ROADM to allow the customers to construct the topologies they need in a very simple way, given having more data centers of certain size in a particular region as well, but still getting that massive capacity. And then beyond that, we've got work to do on multirail to get more fibers because we can now fill up a fiber. So a lot of evolution there on the capacity. By the way, on the non-hyperscaler customer, a lot of evolution deep in the network to be able to get bandwidth as deep as we can, which is a completely different but equally important set of trade-offs.

Yeah. So that's actually—on top of all the other great things as you mentioned, the fact that it fits in a power and physical envelope that's so dramatically smaller than when we started coherent optics is one part. The interoperability part is something that really hasn't been part of the optical networking lexicon ever. Really, the line side of the optics—whether coherent or before that—people have been pushing hard to get the most bits on the wavelength and haven't worried about doing that in a way that somebody else can do. But if you look at the volumes that are expected, the needs of customers for secure sources of supply—a thing we got a bitter reminder of a few years ago—there's been a lot of interest in making sure we can interoperate, not just in general but with probabilistic constellation shaping, which is a reasonably both technologically aggressive but complex technique to maximize the capacity of a wavelength. And so the significance of this is: yes, it's interoperable, which has all the good properties, but in a very complex technical domain—which is a testament to a reasonably large amount of work in a short period of time in the standardization setting to get something that gives great results on one hand but can be done by a bunch of people on the other. Those two have historically been in opposition and they've been pulled together here. That's what that's about.

So yeah, it was a collective effort across a number of vendors to get both specification, but also to get implementations done in a time frame where we're actually—it's being shown here.

Yeah. So if you look at inter-data center, in some ways it's a very similar, just not identical, problem space to what we've done in the long haul forever. And we talk about long haul and we immediately, as you know—I've tortured you before with—oh well this is really regional and this is ultra long. But it's a very similar problem space in that it's a capacity game, right? There's a finite set of resources—fiber—and how do you make the most of them. I think if you look at the inter-DC space today, the difference and the evolution has been that the pressure, the need for capacity there is just ever so much more. And as you get there, you continue to push—we've got to the point where now we can fill up a fiber. And obviously if you think about running a fiber across Texas or maybe across the whole US, that's a very significant investment. If you think about other places, even just across the state, it's an investment, but maybe it's cheap—how exotic do you want your technology? Can I do another fiber, right? But when you're—it's a place where you're both fiber-constrained and the capacity demands are growing. So the pressure there is on capacity. Also going back to the collection of data centers and how they're being built out is shifting due to power constraints, and that starts to put pressure, as I was alluding to earlier, on how you interconnect them and the architecture and how you need to support that, on latency and how you want to do it. So it adds some new pressures to what we've been doing. In some ways, if you think about many of these pieces—and this is true even for the service provider—the core technologies are the same. The key thing is packaging the right pieces into the right solutions for the customer. And for sure, inter-data center stuff starts to look very similar to what we do, maybe only for the biggest customers. But if I look back say 10 years ago when we would have talked about C+L, we would have said yeah, that's only for a very small set of customers that have this problem. If I look at now, I have dozens and dozens of customers for that, and so it's part of the capacity thing—it just happens to be at the high end. By the way, we have a lot of CSP customers who provide inter-DC connections, either of their own or for end customers who are not ones who are building their own. So the inter-DC space is also across different customer segments as well, albeit with different pressures and different choices.

No, I think it's a great summary. There's an awful lot going on, as you say—an awful lot of pressure—and we've got a broader portfolio, broader set of people, and are in a position to move faster, get innovation that matters into the hands of our broad customer base sooner than we ever could together.