Coherent optics, photonic line systems, and software agents are helping cloud providers and telco companies to manage the explosive AI workload demands on fiber networks – across cutting-edge interconnect systems, upgraded longhaul infrastructure, and emerging mobile access networks. As told to RCR by Ciena at MWC.
In sum – what to know:
Fiber-first scale – Ciena is unbundling its telco-grade optical fiber systems into interconnect pluggables for hyperscalers, helped by its Hyper Rail photonic line systems and $270m Nubis acquisition.
Software gains – Its Blue Planet OSS and Navigator controller layer telemetry, automation, and agentic AI on top of its WaveLogic optics to create programmable closed-loop fabric from cloud to edge.
Mobile access – Blue Planet support for last-mile 5G access means telcos can orchestrate AI workloads at the edge, supporting demand for dedicated spectrum and mission-critical performance.
Here’s a quick replay of a conversation with Ciena at MWC at the start of the month (March 2-5) – just to make the newsletter cut before today’s deadline (March 24). The 12-inch remix might appear later this week, featuring better lines and bigger hooks – just because some conversations are worth telling properly. Being MWC, the talk is mostly about fusty old telco networks, rather than trendy new hyper-scale networks. But it starts with the ‘scale-across’ mission, per the AI industry’s cloud priorities. “Right now, all eyes are on that. Because everything is happening there; it’s this gravitational force. And the big thing is the scale-across as data centers are more distributed.”
Scale up, out, and across – as the narrative goes; to connect AI compute as it multiplies inside and between racks (‘up’ and ‘out’) in data centers until land and power are used-up, and further multiplication has to be orchestrated ‘across’ longer distances between regional AI ‘factories’. I mean, that is the only AI-infrastructure story in telecoms, right now – surely? About high-capacity fiber-optic interconnects for frontier models in scattered data-center clusters? Mostly in the US, mostly by cloud providers – and nothing much to do with mobile at all? “Yes,” responds Rebecca Smith, senior vice president of marketing at Ciena. “Because it has to be, right?” But give it time, she goes on.
“The data has to leave the data center [at some point] to monetize anything-AI, right – as inference moves closer to where these models are used and consumed. Even at the edge, it has to go out to the user. Which is driving a lot of interesting trends – because the hyperscalers and neoscalers are investing massively in data centers, and some [telco] service providers are too; and the service providers, closer [to the edge] with regional networks, have access to the end users. You might think, you know, the hyperscalers will dominate and disintermediate. But we don’t see that; those regional networks, especially in markets like Europe, make telcos incredibly valuable partners in this.”
Still not mobile, though – not much, and not yet. Maybe ‘mobile’ should just be dropped from ‘MWC’ on the grounds that fixed telecoms matters more right now; all the talk about 5G and 6G on the stages and stands feels a little performative when hyperscalers are taking fiber build-outs into their own hands, and the most urgent business meetings in the back rooms on the show floor are about the flex and capacity of new fiber systems. Smith responds: “Yes, I’ve been coming for 20 years, and it is much less of a mobile show now; it is really a general telecoms show. But it has a different lens on the AI opportunity – versus OFC, say, where we see more hyperscalers.”
Fiber interconnects
At writing, the telco industry’s big optical-fiber conference, OFC, has just finished (March 7-11) in Los Angeles, of course. (RCR has an OFC interview with Ciena, pending.) But both ways, Ciena has a captive market – with urgent east-west hyper-scale interconnect projects, per the OFC agenda, and semi-urgent north-south telco upgrade works, per the MWC discussion; plus with mobile backhaul, and even carrier OSS orchestration via its Blue Planet subsidiary. And 5G/6G matters, very clearly; Nokia (selling as much optical componentry as mobile radios, these days) said on Sunday night at MWC that 53.5 percent of AI traffic starts there. “[AI] is mobile-native,” it stated.
But no question: fiber is the word in AI infrastructure in mid-2026, going forwards. Ciena’s customer split is between ‘enterprises’ (telcos mostly, and AT&T significantly, buying routing and switching products and software platforms), and cloud providers (hyperscalers and other AI factory builders, buying optical products and photonics). The trend is towards the latter, plus a stronghold of clients like AT&T. But it has seen a surge among a few large customers; just three contributed 43 percent in the final quarter of 2025, worth 10-plus percent of its full-year takings (up about 20 percent; its first 2026 quarterlies are here). Analysts cite “one of the strongest demand environments in its history” – for this hyperscale(-across) bump.
Which is fairly new, actually – and has prompted Ciena to change its product model, away from integrated optical systems that slot into telco networks, to disaggregated modular building blocks and pluggable optics that match how hyperscalers build infrastructure. It has just (September 2025) acquired Nubis Communications for $270 million to bring in compact optical and electrical interconnect tech for ultra‑high capacity inside AI clusters.
“Optical transport is our bread-and-butter… We are known for our systems-play on the telco side,” says Smith.
“The data center is a different opportunity – about how to take our technologies and break them down into components to get ultra-high capacity in and around the data center.” So, Ciena has unbundled its telco-grade optical tech in other words so cloud providers can custom-build at hyper-scale. Because capacity is exponential, and efficiency (power, space, latency) is everything. Which is where assets like its Hyper Rail system come into play: a new reconfigurable (RLS) photonic line architecture that shifts the approach from expanding a wavelength at a time to lighting up entire fiber pairs (‘rails’) simultaneously to meet the massive bandwidth demands of AI workloads.
Wholesale models
As such, its move to deconstruct its telco proposition has come full circle; less about dismantling capability, more about reformatting it – to align with the pluggable software-driven infrastructure models of the hyperscale era, now going faster and wider with new AI dollars. But we’re at MWC, and the talk is about telcos in all of this. It is funny because AT&T, Colt, Lumen, Verizon, Vodafone, Zayo – etc – all talk like they are connecting to data centers, and even handling some of the scale-across. “It is not like a hard line – like hyperscalers at this many kilometers, service providers from here. It depends where it is, as well – North America versus Europe, say,” responds Smith.
She adds: “But hyperscalers are definitely investing in those scale-across architectures, and, then, there are corridors in Asia Pacific where see a lot of these MOFN networks across very large distances – which, you could argue, is scale-across. So they definitely have a role in that value chain, right?” We are going south, quickly, into terrestrial and subsea longhaul systems – which (bar most new subsea builds) remain the provenance of traditional telcos, including some with mobile RAN setups attached to the ends. This is the mostly-fiber discussion at MWC.
Intriguingly, some here (Deutsche Telekom, say) are also setting up new hyperscale-homes, full to the rafters with GPUs, in their own regional network strongholds. “Especially in markets like Europe, where sovereignty and security are such big topics,” says Smith. “Where the regional service providers have networks that the hyperscalers don’t – which, again, makes them incredibly valuable partners.” As such, there is rising interest in wholesale-style MOFN (managed optical fibre network) leasing models to secure wavelengths for AI workloads – extending to dark fibre and managed spectrum, and even outsourced work (where telcos take charge of the first-mile to get the facility online).
“Those models have been around for years, but we are seeing lots of variations because of AI, where hyperscalers want access to these regional networks to bring everything to the edge. We are seeing lots of investment there; it is good for service providers,” says Smith. Dynamic SD-WAN controls for throttling bandwidth up and down are also in demand. But how do ‘regional’ telcos – with metro infrastructure, backhaul and longhaul, some subsea pairs, wholesale deals all over – limber up, and even match the programmable performance that deep-pocketed cloud firms are lighting-up between data centers?
Twin technologies
The answer is this combination of coherent optics and photonic line systems, where the transmission technology encodes data onto light signals with advanced modulation and error correction (allowing huge volumes to travel long distances efficiently), and the transmission infrastructure carries and routes the signals across the network (as the ‘highway’ for the coherent ‘traffic’, to use a standard throughput analogy) – from longhaul north-south routes to shorter-haul interconnect links in the cloud and metro systems at the edge. Ciena frames it in fundamental terms: “what matters” in any AI network, says Smith, is still power-efficiency per kilometre and cost per bit.
The difference – suddenly, increasingly – is how those twin requirements, technical and economic, are being met across fragmented global fiber infrastructure that was designed for the internet. Its WaveLogic platform, the brand for its coherent optics solution, is being developed in two R&D tracks. Its latest (sixth-gen) Wavelogic 6 product is available as an embedded system for routers and switches in longhaul telco networks and as pluggable interconnect components for “shorter-reach” data center usage, now enhanced with the Nubis technology. “We’re seeing growing demand on both sides,” says Smith. “We are uniquely positioned in the market right now.” Ciena has more, of course.
Physical infrastructure is half the story, says Smith. The challenge is to run the network efficiently, at scale. “Which is a lot of what we have talked about here. Service providers have talked about efficiency for years. The difference is it’s not just about competitiveness, anymore; it is an EBITDA imperative for the c-suite. That’s the shift – which opens up conversations about automation and autonomy.” But it is a chicken-and-egg deal. “You need the network for that,” says Smith. “Scalability, capacity, programmability – which comes with the Wavelogic technology. Then you need the software on top, which has the telemetry, understands the demands on the network, and brings the intelligence.”
That software comes in two layers: an on-box domain controller (Navigator) for its own fiber infrastructure, and a multi-vendor fiber/mobile OSS layer (courtesy of Blue Planet). “We can sense what is happening across a network, whether it’s a Ciena optical network or a third-party routing network..” By layering in agentic AI, Smith says Ciena can now reason over its inputs, as well: interpreting telemetry, understanding SLAs and business objectives, and recommending or executing actions to reallocate resources, resolve alarms, or adjust traffic flows. “We now have a closed-loop programmable approach to manage the network. That’s the kind of conversation we’re having at MWC.”
Software automation
Telcos used to rely on manual spreadsheets and truck rolls, she says. “Now, it’s programmable, and increasingly autonomous.” The same software works for last-mile mobile access, too – where the towers join the fiber backhaul, going into the longhaul; where Ciena’s optics come into play. Blue Planet crosses both disciplines; beyond visibility and telemetry, it also supports 5G slicing (deployed with Deutsche Telekom, for instance). Smith says: “AI could be the killer app for… well, I mean, we’ve been trying to figure out the killer slicing app forever – but it could be AI. Think of healthcare, or lots of industries, where you need low latency and dedicated spectrum to run an AI application.
She goes on: “Those are the table stakes for a lot of AI –100 percent uptime, no interruptions; mission critical. That is where slicing comes into play. A lot of AI applications only make sense with next-to-no latency. So AI might drive that; it could be super valuable for enterprises, right? So it is not that AI is enabling that slice, but that AI will drive the need for the slice. And Blue Planet allows the service provider, assuming it has the bandwidth, to provision and orchestrate that slice.” There is agentic AI in there, presumably – so the Blue Planet proposition extends its parent’s “closed-loop, programmable approach” from the data center to the edge.
She says: “Agentic AI is changing the conversation because it allows telcos to go faster. The fact you could create an agent, tell an agent what to do in your operating environment without any coding is phenomenal, right? You don’t need a data scientist to do that. You don’t need an expert engineer. You could just have somebody in a NOC say, ‘this is my agent; go do this’. The potential for that is really unlimited for the service providers.” The conversation skirts around a whole lot more besides, but we have to bring it to a close. And, so that’s it – the long player; there never was a single version, after all. Just to finish, Smith notes the difference in telcos, MWC 2026 versus 2025.
She reflects: “Like I said, AI is this gravitational force, which is pulling hyperscaler investment and service provider investment behind it. If you’re part of the ecosystem, it is an interesting time. And the paths to opportunity are clearer, now. Last year, we were still like: what does it mean? We now understand that better – how it’s playing out. And not just in the data center, but with real opportunities for service providers – whether it’s a wholesale MOFN thing, or an industrial cloud in Germany. This industry is figuring it out, right? Which doesn’t mean that everything is figured out, of course, because it will evolve. It is like that iceberg cliche, where there’s so much more below the surface.”
