The AI build-out is rewriting the fibre comms order-book – between knowledge centres and inside knowledge centres, proper right down to the rack itself. As energy constraints, scale-out architectures, and interconnect bottlenecks push fibre ever nearer to the data-centre edge, optical networking specialist Ciena has emerged as one of many customary bearers for the entire shift.
In sum – what to know:
Optical infra – AI is driving funding in fibre optic programs, not simply between knowledge centres however more and more inside them, as scale-out and rack-level architectures expose the boundaries of copper interconnects.
Ciena story – Ciena’s surge displays extra than simply market exuberance: sturdy demand for optical line programs, coherent pluggables, and short-reach photonics are all amplified by hyperscaler spending.
AI fragments – as knowledge centres are distributed geographically and architecturally, to cope with energy demand, worth is transferring into fibre, connectors, and optical packaging – triggering a land seize.
If you would like an image of the ripple impact from the AI build-out – of the waves crashing throughout as mega-sized knowledge centres rise as much as home huge frontier language fashions – then you definately may take a look at the businesses enabling the underlying infrastructure: in fibre and cabling, optical switches, cloud interconnects, and the software program that orchestrates all of it. You may also take a look at energy and cooling, after all, plus safety, development, integration. (It’s an infrastructure free-for-all, at the least whereas the AI trade accepts a sketchy ROI in pursuit of killer purposes.) However for conventional telecoms, the actual spin-off story is about fibre programs and componentry, and few corporations illustrate this higher than US outfit Ciena, which has emerged from 2025 just like the darling of optical networking.
Ciena’s inventory is up by over 160 p.c during the last six months (and the identical over 12 months). It reached a report 20-year excessive of just about $260 per share final week, having hardly been above $60 because the tail finish of the dot-com crash in 2000/01. Shares “raced” almost 15 p.c larger simply in December, following an “glorious” fiscal fourth quarter (when revenues jumped 20 p.c on the year-ago interval to $1.35 billion) and an astonishing yr (after posting whole revenues of $4.77 billion at about the identical charge). Its revenue (EBITDA) was up by greater than 32 p.c within the yr (to $637 million); its working margin, about 11 p.c, was marginally larger (by 1.5 proportion factors). Ciena has raised its 2026 income forecast to $5.7-$6.1 billion, with a margin of 17 p.c.
Heady days. As an apart, there may be some minor intrigue about insider share gross sales across the flip of the yr: the corporate’s chief govt, Gary Smith, bought 83,552 shares for about $19 million in early January; administrators and “different insiders” have bought a complete of 127,117 shares (for round $28 million) over the previous couple of months; asset administration agency Robeco Institutional, a comparatively minor investor, has successfully bailed, promoting most of its holding for $216,000. However telecoms analysts recommend these are discrete inventory sub-plots. “The share value has greater than doubled within the final yr; I don’t suppose you’ll be able to learn an excessive amount of into that,” responded Dimitris Mavrakis, analysis director at ABI Analysis, now protecting data-centre fibre interconnect market.
Funding analysts, in the meantime, are bullish, citing “distinctive demand tailwinds”, and Ciena’s “increasing product footprint” and “beneficial” value/earnings-to-growth (PEG) ratio, even contemplating a high-end inventory valuation. Fairness researcher Uttman Dey perceives a “phenomenal vote of confidence for a structural swing in the direction of the adoption of optical connectivity options and parts for the AI knowledge centre”. He writes: “Many main gamers have reported seeing a robust decide up, not solely in enterprise curiosity for optical options, but additionally in bookings… Ciena is one… Perhaps, [it] has nearly turned the web page in the direction of one of many strongest demand environments in its historical past as extra hyperscalers take a look at adopting optical options.”
Connecting knowledge centres
Mavrakis says the identical: “It’s not simply Ciena that’s doing properly; it’s all these optical corporations. I imply, many of the development for Nokia, say, is from its infrastructure enterprise – which is all optical stuff, proper? It’s triggered by all of the investments in AI knowledge centres… So the AI market is lifting the optical trade, as seen in Ciena’s outcomes. In the event you take a look at its 2024 numbers, there was a slight dip as a result of [telecoms] service suppliers weren’t spending a lot. However that has been fully negated by hyperscalers in 2025, investing way more than service suppliers ever did. And that can proceed; funding in knowledge facilities is growing, which is able to register for the optical corporations too. And Ciena is among the leaders. So the outlook could be very optimistic; Ciena could be very properly positioned.”
Certainly, he notes a creating development in knowledge centre development. The place giant coaching clusters have been concentrated in giant venues till now, they are going to be progressively damaged into smaller regional facilities over the following few years, he says. The logic is to make the most of vitality availability – primarily to separate energy provides between areas, simply to have the ability to draw sufficient to energy these monster AI factories. It’s not about sovereignty or latency, he responds; nearly energy. “It’s simply vitality,” he says. “I imply, contemplate that the majority of those developments are within the US, the place sovereignty will not be such a precedence – particularly for coaching these giant basis fashions. Aside from provide chain constraints round getting GPUs from Nvidia, vitality is the most important problem.”
He says: “There’s simply not sufficient energy in a single place to help these huge websites. So they are going to be cut up into smaller regional clusters.” A forced-fragmentation would require distributed knowledge centres to carry out “as one”, he says – so the fibre between them is very out there and dependable, and quick. They want “1000’s of fibres between”, he says. “So it’s a very optimistic outlook for Ciena, and everybody else.” That’s the forecast; what concerning the climate outdoors? It’s instructive to contemplate Ciena’s buyer/product segmentation, cut up between enterprise prospects (cell operators principally, and AT&T considerably, shopping for routing and switching merchandise and software program platforms), and cloud suppliers (hyperscalers and different AI manufacturing unit builders, shopping for optical merchandise and photonics options).
The development is in the direction of the latter, and in addition in the direction of a stronghold of purchasers, like AT&T and sure others. On the latter rating, Ciena detailed some dependence on a number of giant prospects in its 2025 outcomes, notably in the direction of the back-end of the yr, when three prospects contributed 43-percent of income within the quarter value 10-plus p.c within the full yr – and simply two contributed 28.4 p.c within the quarter and 10-plus p.c, once more, within the fiscal yr. Dey’s evaluation is nice on this; he cites a full-year submitting {that a} “giant and unknown cloud supplier” accounted for round 18 p.c of its whole gross sales in 2025. Which is the place the wind is blowing, about this rising swell round AI knowledge centres. “The turnaround [in 2025] was led by an acceleration in [its] optical networking phase,” he writes.
He explains: “Administration attributed the sturdy development in optical networking to fast decide up in gross sales of photonic optical line programs… [which are] sometimes used to simplify and automate the deployment and operation of versatile, high-capacity networks. Coherent pluggable transceivers have been the opposite product phase that contributed to accelerating development which most likely incentivized Ciena to accumulate Nubis and increase its optical networking options portfolio for scale up and scale out networks.” Which is an fascinating sub-plot in optical networking, tied to AI knowledge centres, as properly. This deal with high-capacity optical programs additionally helps clarify Ciena’s acquisition of Nubis, which expands its portfolio deeper into short-reach and near-compute interconnects for AI-driven networks.
Connecting knowledge programs
The theme extends into AI knowledge centres as properly. “With density in GPU clusters already crossing 100,000 GPUs on a single cloth, it is smart that the trade is deploying optical line programs to simplify and handle… these scale-up and scale-out networks,” says Dey.
That dynamic additionally explains why Ciena is rocking out — and why the fibre-optic divisions of corporations reminiscent of Nokia and Cisco are additionally seeing an uplift. Fibre programs are required not solely between knowledge centres, however more and more inside them, as system-level ‘materials’ span many racks and rows and on-site cabling connects 1000’s of GPUs throughout single services. At that scale, the material turns into optical by necessity, and is more and more managed utilizing telecom-style optical programs relatively than handled simply as outsized Ethernet clusters. Nubis delivers the short-reach and medium-reach optical interconnects – together with co-packaged and near-packaged optics – to scale back energy, latency, and complexity as these intra-facility comms scale.
The story goes deeper; the identical logic extends contained in the rack. As AI clusters are decomposed geographically, they’re additionally decomposed architecturally, with processors and accelerators not confined to a single board or chassis, exposing the boundaries of conventional electrical interconnects even at very brief distances. Mavrakis at ABI Analysis says copper is the pinch-point not simply throughout racks, however inside them. He says: “In the event you take a look at applied sciences like Nvidia’s NVLink, connecting GPUs on the identical board, or the hyperlink switches that join boards in a rack, they sometimes help one or two metres of attain at most. Which is driving discussions round silicon photonics – the place all the pieces turns into fibre relatively than copper – transferring deeper into the info centre, and finally contained in the rack.”
For many years, the guideline has been “copper for those who can, fibre for those who should”, he says. Copper has been cheaper and less complicated, and deeply entrenched – notably in Ethernet-based programs. Fibre, in contrast, is pricey and complicated, and notoriously power-hungry. However the calculus is altering. “It’s not that fibre is all of the sudden simple,” he says. “It’s nonetheless dearer and complex. However AI programs demand it – not in every single place, however more and more so. In 5 to 10 years, fibre will dominate [inside the rack]; fibre is the expansion space.” Which is mirrored available in the market, and on this narrative after all. A rising share of AI infrastructure funding is flowing into optical connectivity at ever shorter reaches – from data-centre interconnect, to intra-facility materials, to rack-level inter-GPU comms.
Latest strikes underline the development: Marvell’s acquisitions of XConn and Celestial AI, as written about final week; Qualcomm’s buy of Alphawave in December; even Amphenol’s deal to accumulate complementary operations from CommScope this week – all of those trades level to a race to regulate the interconnect layer closest to fragmenting data-centre compute features. And the reason being plain: as AI programs develop into extra distributed, efficiency bottlenecks more and more sit within the interconnect – between services, throughout the info corridor, within the rack. Worth is shifting into cabling, connectors, fibre administration, and optical packaging applied sciences that may transfer knowledge effectively at scale. And Ciena appears to be like like a poster baby for it.
