The AI build-out is rewriting the fibre comms order-book – between data centres and within data centres, right down to the rack itself. As power constraints, scale-out architectures, and interconnect bottlenecks push fibre ever closer to the data-centre edge, optical networking specialist Ciena has emerged as one of the standard bearers for the whole shift.
In sum – what to know:
Optical infra – AI is driving investment in fibre optic systems, not just between data centres but increasingly inside them, as scale-out and rack-level architectures expose the limits of copper interconnects.
Ciena story – Ciena’s surge reflects more than just market exuberance: strong demand for optical line systems, coherent pluggables, and short-reach photonics are all amplified by hyperscaler spending.
AI fragments – as data centres are distributed geographically and architecturally, to deal with power demand, value is moving into fibre, connectors, and optical packaging – triggering a land grab.
If you want a picture of the ripple effect from the AI build-out – of the waves crashing all around as mega-sized data centres rise up to house massive frontier language models – then you might look at the companies enabling the underlying infrastructure: in fibre and cabling, optical switches, cloud interconnects, and the software that orchestrates it all. You might also look at power and cooling, of course, plus security, construction, integration. (It is an infrastructure free-for-all, at least while the AI industry accepts a sketchy ROI in pursuit of killer applications.) But for traditional telecoms, the real spin-off story is about fibre systems and componentry, and few companies illustrate this better than US outfit Ciena, which has emerged from 2025 like the darling of optical networking.
Ciena’s stock is up by over 160 percent over the last six months (and the same over 12 months). It reached a record 20-year high of almost $260 per share last week, having hardly been above $60 since the tail end of the dot-com crash in 2000/01. Shares “raced” nearly 15 percent higher just in December, following an “excellent” fiscal fourth quarter (when revenues jumped 20 percent on the year-ago period to $1.35 billion) and an astonishing year (after posting total revenues of $4.77 billion at about the same rate). Its profit (EBITDA) was up by more than 32 percent in the year (to $637 million); its operating margin, about 11 percent, was marginally higher (by 1.5 percentage points). Ciena has raised its 2026 revenue forecast to $5.7-$6.1 billion, with a margin of 17 percent.
Heady days. As an aside, there is some minor intrigue about insider share sales around the turn of the year: the company’s chief executive, Gary Smith, sold 83,552 shares for about $19 million in early January; directors and “other insiders” have sold a total of 127,117 shares (for around $28 million) over the last few months; asset management firm Robeco Institutional, a relatively minor investor, has effectively bailed, selling most of its holding for $216,000. But telecoms analysts suggest these are discrete stock sub-plots. “The share price has more than doubled in the last year; I don’t think you can read too much into that,” responded Dimitris Mavrakis, research director at ABI Research, now covering data-centre fibre interconnect market.
Investment analysts, meanwhile, are bullish, citing “exceptional demand tailwinds”, and Ciena’s “expanding product footprint” and “favourable” price/earnings-to-growth (PEG) ratio, even considering a high-end stock valuation. Equity researcher Uttman Dey perceives a “phenomenal vote of confidence for a structural swing towards the adoption of optical connectivity solutions and components for the AI data centre”. He writes: “Many major players have reported seeing a strong pick up, not only in business interest for optical solutions, but also in bookings… Ciena is one… In all probability, [it] has just about turned the page towards one of the strongest demand environments in its history as more hyperscalers look at adopting optical solutions.”
Connecting data centres
Mavrakis says the same: “It’s not just Ciena that’s doing well; it’s all these optical companies. I mean, most of the growth for Nokia, say, is from its infrastructure business – which is all optical stuff, right? It is triggered by all the investments in AI data centres… So the AI market is lifting the optical industry, as seen in Ciena’s results. If you look at its 2024 numbers, there was a slight dip because [telecoms] service providers were not spending so much. But that has been completely negated by hyperscalers in 2025, investing much more than service providers ever did. And that will continue; investment in data centers is increasing, which will register for the optical firms too. And Ciena is one of the leaders. So the outlook is very positive; Ciena is very well placed.”
Indeed, he notes a developing trend in data centre construction. Where large training clusters have been concentrated in large venues until now, they will be progressively broken into smaller regional centers over the next few years, he says. The logic is to take advantage of energy availability – essentially to split power supplies between regions, just to be able to draw enough to power these monster AI factories. It’s not about sovereignty or latency, he responds; just about power. “It is just energy,” he says. “I mean, consider that most of these developments are in the US, where sovereignty is not such a priority – especially for training these large foundation models. Apart from supply chain constraints around getting GPUs from Nvidia, energy is the biggest challenge.”
He says: “There’s just not enough power in one place to support these massive sites. So they will be split into smaller regional clusters.” A forced-fragmentation will require distributed data centres to perform “as one”, he says – so the fibre between them is highly available and reliable, and fast. They need “thousands of fibres between”, he says. “So it is a very positive outlook for Ciena, and everyone else.” That’s the forecast; what about the weather outside? It is instructive to consider Ciena’s customer/product segmentation, split between enterprise customers (mobile operators 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 solutions).
The trend is towards the latter, and also towards a stronghold of clients, like AT&T and certain others. On the latter score, Ciena detailed some dependence on a few large customers in its 2025 results, particularly towards the back-end of the year, when three customers contributed 43-percent of revenue in the quarter worth 10-plus percent in the full year – and just two contributed 28.4 percent in the quarter and 10-plus percent, again, in the fiscal year. Dey’s analysis is good on this; he cites a full-year filing that a “large and unknown cloud provider” accounted for around 18 percent of its total sales in 2025. Which is where the wind is blowing, about this rising swell around AI data centres. “The turnaround [in 2025] was led by an acceleration in [its] optical networking segment,” he writes.
He explains: “Management attributed the strong growth in optical networking to rapid pick up in sales of photonic optical line systems… [which are] typically used to simplify and automate the deployment and operation of flexible, high-capacity networks. Coherent pluggable transceivers were the other product segment that contributed to accelerating growth which probably incentivized Ciena to acquire Nubis and expand its optical networking solutions portfolio for scale up and scale out networks.” Which is an interesting sub-plot in optical networking, tied to AI data centres, as well. This focus on high-capacity optical systems also helps explain Ciena’s acquisition of Nubis, which expands its portfolio deeper into short-reach and near-compute interconnects for AI-driven networks.
Connecting data systems
The theme extends into AI data centres as well. “With density in GPU clusters already crossing 100,000 GPUs on a single fabric, it makes sense that the industry is looking at deploying optical line systems to simplify and manage… these scale-up and scale-out networks,” says Dey.
That dynamic also explains why Ciena is rocking out — and why the fibre-optic divisions of companies such as Nokia and Cisco are also seeing an uplift. Fibre systems are required not only between data centres, but increasingly inside them, as system-level ‘fabrics’ span many racks and rows and on-site cabling connects thousands of GPUs across single facilities. At that scale, the fabric becomes optical by necessity, and is increasingly managed using telecom-style optical systems rather than treated just as oversized Ethernet clusters. Nubis delivers the short-reach and medium-reach optical interconnects – including co-packaged and near-packaged optics – to reduce power, latency, and complexity as these intra-facility comms scale.
The story goes deeper; the same logic extends inside the rack. As AI clusters are decomposed geographically, they are also decomposed architecturally, with processors and accelerators no longer confined to a single board or chassis, exposing the limits of traditional electrical interconnects even at very short distances. Mavrakis at ABI Research says copper is the pinch-point not just across racks, but within them. He says: “If you look at technologies like Nvidia’s NVLink, connecting GPUs on the same board, or the link switches that connect boards in a rack, they typically support one or two metres of reach at most. Which is driving discussions around silicon photonics – where everything becomes fibre rather than copper – moving deeper into the data centre, and ultimately inside the rack.”
For decades, the guiding principle has been “copper if you can, fibre if you must”, he says. Copper has been cheaper and simpler, and deeply entrenched – particularly in Ethernet-based systems. Fibre, by contrast, is expensive and complex, and notoriously power-hungry. But the calculus is changing. “It’s not that fibre is suddenly easy,” he says. “It’s still more expensive and complicated. But AI systems demand it – not everywhere, but increasingly so. In five to 10 years, fibre will dominate [inside the rack]; fibre is the growth area.” Which is reflected in the market, and in this narrative of course. A growing share of AI infrastructure investment is flowing into optical connectivity at ever shorter reaches – from data-centre interconnect, to intra-facility fabrics, to rack-level inter-GPU comms.
Recent moves underline the trend: Marvell’s acquisitions of XConn and Celestial AI, as written about last week; Qualcomm’s purchase of Alphawave in December; even Amphenol’s deal to acquire complementary operations from CommScope this week – all of these trades point to a race to control the interconnect layer closest to fragmenting data-centre compute functions. And the reason is plain: as AI systems become more distributed, performance bottlenecks increasingly sit in the interconnect – between facilities, across the data hall, in the rack. Value is shifting into cabling, connectors, fibre management, and optical packaging technologies that can move data efficiently at scale. And Ciena looks like a poster child for it.
