Optical Connections Magazine - Autumn 2026

Bringing the World the Latest in Optical Communications News

ISSUE 44 | Q3 2026

SCALING OPTICS FOR AI DATA CENTRES ANTHONY SAVVAS | p8

RETHINKING TESTING FOR HOLLOW-CORE FIBRE

Douglas Clague | p16

AI FOR OPTICAL NETWORKS Pino G. Dicorato | p24

PHOTONICS IN THE AGE OF AI Dr Helena Diez-y-Riega & Dr. Jon Pugh | p18

NEXT-GEN FREE-SPACE OPTICS  NEW 1000KM FIBRE ROUTE  ADVANCED MONITORING PLATFORMS

Media Partners to

Scan to learn more

WWW.USCONEC.COM

CONTENTS

4 8

FROM CONCEPT TO REALITY AT ECOC EXHIBITION 2026

Industry News

Scaling optics for AI data centres Anthony Savvas

Welcome to the Autumn 2026 edition and ECOC Exhibition 2026 special of Optical Connections.

14 Reducing fibre connection losses Pietro Bernasconi 16 Hollow-core fibre testing Douglas Clague 18 Photonics in the age of AI Dr Helena Diez-y-Riega & Dr. Jon Pugh 20 The fibre bottleneck Thomas Ritz 22 AI in broadband networks Manuel Paul & Tony Zeng 24 AI-driven automation Pino G. Dicorato 26 Cleaning for stable fibre connections Masayuki Murakami 31 ECOC2026 Preview 33 Product News

A year ago, we were highlighting the industry’s impressive momentum, as it accelerated new optical technologies and initiatives to meet the emerging demands of AI‑driven data centre connectivity. This year, we are starting to see that progress move from concept to reality. Some of the solutions we discussed in 2025 will now be on display and demonstrated live across the Exhibition’s Market Focus sessions and Product Focus showcases. In this issue, we echo some of these developments, with a strong focus on how optical networks are laying the foundation for the AI era, and vice versa. Hollow‑core fibre (HCF) is one of the technologies now moving firmly from promise to deployment. Its distinct physical properties unlock transformative performance gains but also introduce new complexity in testing and certification, which must be addressed before it becomes a stalwart in optical infrastructure. We also look at other key technologies shaping the AI data‑centre era - from photonics to ultra‑low‑loss fibre connections - and the physical preparation needed to ensure they operate seamlessly when the system goes live. It is not only optical connections shaping the future of AI, but AI is now reshaping optical connections themselves. In this edition, we look at how AI is being embedded deeper into the network, from access and in home environments to the automation frameworks that make optical communications more efficient, predictable, and reliable. I hope you enjoy reading this edition and the sold out ECOC Exhibition as our industry continues to explore the technologies, challenges, and opportunities shaping the next chapter of optical innovation and building a more intelligent, connected world.

Brian Dolby Editor, Optical Connections

READ ONLINE/SUBSCRIBE: www.opticalconnectionsnews.com FOLLOW US: @opconsnews

MEDIA PARTNERS

EDITORIAL: editor@opticalconnectionsnews.com ADVERTISING: sales@opticalconnectionsnews.com DESIGN: Antonio Manuel

Optical Connections is published by NEXUS MEDIA EVENTS LTD Suite 3, Building 30, Churchill Square, Kings Hill, West Malling, Kent ME19 4YU t: +44 (0) 1732 752 125

www.opticalconnectionsnews.com

3

ISSUE 44 | Q3 2026

INDUSTRY NEWS

Enabling next-generation free- space optical communications

Taara, a graduate of X, Google’s Moonshot Factory, and a leader in high-speed, high- capacity wireless optical communications, has launched the industry’s first link planning product developed specifically for next-generation free-space optical communications. Taara Link Planner brings nearly a decade of research and development originating at X into a publicly accessible platform for network operators, infrastructure teams, and network planners. Its models have been validated against one of the largest real-world datasets on free-space optical links ever assembled, drawing on deployments and atmospheric observations spanning diverse

geographies, climates, and operating conditions around the world. The platform allows users to plan deployments of Taara Lightbridge, the company’s wireless optical communications technology, which uses narrow, invisible beams of light to deliver high-capacity connectivity through the air between two fixed and easily deployable terminals. By entering an address or coordinates, or by placing link endpoints manually on a map, users can assess link distance, identify potential line-of-sight obstructions, adjust mounting heights and estimate expected availability based on local atmospheric conditions. Taara’s internal modelling currently shows availability with about 1% accuracy of

the predictions, on average, when local weather data is available and terminals are installed to specification. “Over the last decade, our deployments from dense cities and remote communities to some of the world’s most demanding atmospheric conditions have taught us how light behaves in the real world. Taara Link Planner brings that body of research together in a product that operators everywhere can use. It gives free-space optical communications the kind of dedicated planning capability that established network technologies have relied on for years, and represents an important step towards making light a mainstream part of global connectivity,” said Mahesh Krishnaswamy, Founder and

CEO of Taara.

Wireless optical communication uses

narrow, invisible beams of light to deliver high-capacity connectivity through the air. Unlike radiofrequency (RF) systems, it operates in unlicensed optical spectrum and avoids many of the congestion, licensing, disruption, and costs associated with traditional wireless infrastructure and fibre deployments. Because wireless optical communications transmit data using highly focused beams of light, network planning involves different environmental and atmospheric considerations than traditional RF systems. Taara Link Planner has been designed specifically with those optical characteristics in mind.

‘First-ever’ automated error detection for network records launched

euNetworks deploys 1000km fibre route between Paris and Milan

AI software developer IQGeo has announced an “industry-first” capability that automatically detects errors in telecom network records during field operations. The new capability marks a foundational step toward the accurate, continuously updated network data that autonomous network operations require, says the company. When engineers build or repair network infrastructure, the physical work rarely matches the original design exactly. Engineers may need to make slight variations, like changing routes, substituting equipment or changing connections.

Network operator euNetworks has announced the deployment of a new

network records to gradually drift out of sync with reality. Most operators currently reconcile these differences manually, a slow and error-prone process that leaves operational decisions, from planning future investment to diagnosing faults, increasingly dependent on data that no longer reflects the network as it exists in the real world. IQGeo’s visual AI application captures the results of a network build or repair and compares it against the original design. Any differences are immediately flagged, classified and routed to back-office teams for review, with follow-up work orders created where corrective action is needed.

within euNetworks’ existing network. The development follows the delivery of the company’s short, direct route connecting Frankfurt to Milan via Zurich in October 2025. The two developments combined provide the shortest cumulative routes between

1057km long-haul fibre route through The Alps between Paris and Milan. The new route offers the most direct path between the two regions, offering a shorter alternative to the more common routes running along the coast via Lyon and Marseille, according to the company. The route has been developed to provide customers with the most direct, diverse path between the key connectivity hubs of Paris and Milan, as well as creating opportunities for the further shortening and diversifying of interconnecting routes

Frankfurt and Milan, at 771km and 1,000km.

Customers utilising the route will have the ability to connect seamlessly into euNetworks’ owned and operated metro networks in Paris, which includes 38 on-net data centres, and Milan, which includes 18 on-net data centres, as well as benefitting from onward connectivity to the company’s 600+ connected data centres across Europe.

These variations are normal, but they lead

4

www.opticalconnectionsnews.com

| ISSUE 44 | Q3 2026

INDUSTRY NEWS

Brightspeed launches 8-gig fibre for AI businesses

Aylight raises €4.5M to reinvent AI chip lasers

Fibre builder Brightspeed expanded its fibre portfolio with the launch of its 8 Gig speed tier, providing organisations with the network infrastructure needed to support increasingly data-intensive, AI-powered operations. “AI is fundamentally changing how businesses work and it’s changing what businessowners should expect from their network,” said Jeff Lowney, president of Brightspeed Business. “From AI assistants and intelligent automation to cloud collaboration, real-time analytics and connected workplaces, today’s businesses are generating and moving more data than ever before, and connectivity has evolved from a utility to strategic infrastructure. Brightspeed Business 8 Gig

Swiss photonic chip developers Aylight has announced that it has

fibre offering delivers the performance, reliability and flexibility businesses need to build for the future.” AI workloads continuously access cloud-based models, process large datasets and generate high volumes of simultaneous network traffic. These workloads require faster download speeds and greater capacity. Additionally, business owners want lower latency and consistent performance across their increasingly connected organisations. “Our goal isn’t simply to offer faster internet,” Lowney added. “We’re building the digital infrastructure that enables businesses to confidently adopt AI, innovate faster and scale without limits. As technology continues to evolve, our network is built to evolve with it.”

supply for AI data centres — a public signal that moving data between chips, rather than raw compute, increasingly limits how far AI infrastructure can scale. Today each optical link requires its own laser, and that approach does not reach the throughput that data centres will demand. “We started from a problem, not a technology: the laser had become one of the real limits on how far AI can scale,” said Bahareh Marzban, co-founder and CEO of Aylight. “What convinced me was seeing something first demonstrated for a completely different application and realizing it was the answer — so we translated it into a new kind of laser built for AI data centres.

raised €4.5m in a pre-seed funding round to accelerate the development of chip- scale multiwavelength lasers – tackling one of the key bottlenecks limiting the scaling of artificial intelligence. The funding round – co- led by Elaia and Swisscom Ventures with participation from Verve Ventures and Plug and Play – will take Aylight’s laser from a research result to first prototypes, manufactured in a semiconductor foundry. It also lets Aylight expand its R&D team. In March 2026, Nvidia committed $4 billion across Coherent and Lumentum to secure advanced laser and optical-interconnect

www.opticalconnectionsnews.com

5

ISSUE 44 | Q3 2026

INDUSTRY NEWS

FiberCop and Nokia to transform fibre networks into advanced monitoring platforms

Wholesale operator FiberCop and Nokia are testing the use of optical fibre as a distributed sensor, capable of detecting climate events, anomalies and potential faults. The collaboration involves joint research and testing activities on solutions that enable fibre not only to transmit data, but also, thanks to artificial intelligence, to detect events and changes along the infrastructure in real time. “This collaboration is an important step in the evolution of fibre infrastructures, which can become ever more intelligent and capable of providing new information that is useful for their

management,” said Stefano Paggi, Chief Technology & Operations Officer of FiberCop. “The goal is to explore solutions that help strengthen network reliability and open up new application possibilities to support the economy.” Fibre sensing can support two main areas of application: network protection and monitoring of the surrounding environment, the partnership claims. In operational contexts, it enables the detection, location and classification of problems caused by events such as landslides, fallen trees, roadworks or acts of vandalism, supporting faster interventions, reduced field operations and greater

service continuity. The same infrastructure can also enable environmental monitoring services, detecting phenomena such as wind, temperature changes, seismic activity, flooding, leaks or traffic. The technologies at the heart of the collaboration use the network as a distributed sensing system, capable of detecting physical and environmental changes, such as vibrations, temperature variations, or mechanical stresses, through analysis of the optical signal. In this way, fibre evolves from being a connectivity infrastructure to a platform that can generate information useful for the management, security and resilience of infrastructures.

“AI is radically changing what networks need to do. Networks no longer simply carry data; they also help operators understand what is happening in real time,” said John Harrington, Executive Vice President and Head of Europe at Nokia. “By combining Nokia Bell Labs’ innovation with our AI-enabled fibre sensing, we can help FiberCop transform its fibre network into an intelligent monitoring platform that provides reliable infrastructure capable of sensing, understanding and acting in real time. Working together, we can support faster detection of network issues, enable new sensing- as-a-service applications, and create a more resilient network.”

Photon Design natively completes 10 second TFLN simulations

OIF to bring industry-wide interoperability to life at ECOC 2026

Optical networking interoperability forum OIF has announced a live, multi- vendor interoperability demonstration at the ECOC Exhibition taking place this September 21 – 23 in Málaga, Spain. The demo, featuring 39 member companies in booth 2126, will showcase how open interoperability is helping the industry transition next-generation optical networking technologies, from specification to real- world deployment for AI infrastructure. OIF’s demonstration will highlight interoperability across Optical Systems, 448G and 224G Common Electrical I/O (CEI), Common Management Interface Specification (CMIS), Co-Packaging, Energy Efficient Interfaces (EEI) and other interoperability initiatives. The pace of AI infrastructure deployment is placing

Photon Design has become the first company to natively simulate Thin Film Lithium Niobate (TFLN) bends, in just 10 seconds, with its FIMPROP simulation tool, which uses Eigen Mode Expansion (EME) processing. FIMPROP delivers results as rigorous as industry-standard FDTD tools, but hundreds of times faster. Because TFLN is a birefringent optical material that’s refractive index changes with voltage, so it is commonly used for the high-speed, phase switching needed for Mach-Zehnder Modulators (MZMs). These are central to coherent data communications, AI and co- located optics processors. Rigorous TFLN modelling is, therefore, essential for

greater emphasis on proven interoperability across the networking ecosystem. OIF’s live demonstration showcases

developing these emerging, high-data-rate, photonics applications. Dr Dominic Gallagher, CEO of Photon Design, said: “EME achieves its efficiency by modelling

how its Implementation Agreements (IAs) enable

interoperable solutions that simplify integration, preserve vendor choice and accelerate the deployment of next- generation optical networks. “The value of a specification is ultimately measured by how well products from multiple companies work together,” said Mike Klempa, OIF Secretary/Treasurer and Physical and Link Layer (PLL) Interoperability Working Group Chair (Qualcomm). “At ECOC, OIF members will demonstrate interoperability in action, validating technologies across the ecosystem to reduce integration risk, accelerate adoption and support the open, scalable networks required for the next generation of AI infrastructure.”

only the live material regions of the TFLN

adiabatic bend, rather than the full bounding volume. TFLN bends can have large bounding volumes so we used a 500um² ‘S’ bend as a benchmark, demonstrating rigorous simulation results in just 10 seconds. Not only does FDTD process the entire bounding volume, but its runtime also increases with both volume and simulation duration. This means that every time the bend dimensions double, the FDTD simulation runtime increases eightfold.”

6

www.opticalconnectionsnews.com

| ISSUE 44 | Q3 2026

RJ45 Planning and Installation — Easier and Faster than ever with R&M Harmonizing RJ45 connection modules into a universal design helps planners and buyers select the right option faster, manage fewer part numbers, reduce SKU complexity and deliver LAN infrastructure on tighter timelines. Easy Lock quick mounting streamlines termination by combining wire guiding, connection and housing closure into a single, tool-free step. Key installation features — including strain relief, colour coding and a dust cover — are integrated into the module housing, leaving fewer loose parts to drop, lose or rework. • Faster termination without specialist tools—in as little as 1–2 minutes per module. • Fewer errors—colour coding helps prevent mix-ups, while dust protection supports cleaner, safer patching. • Reliable long-term performance—integrated strain relief helps prevent cable movement from affecting transmission. • Simpler logistics—fewer variants and part numbers reduce purchasing effort and warehouse burden. • Less packaging waste—plastic is being replaced with paper and recycled-cardboard formats.

Find out more

www.opticalconnectionsnews.com

7

ISSUE 44 | Q3 2026

ANTONY SAVVAS DATA CENTRE INFRASTRUCTURE

OPTICAL SCALES FOR AI-FOCUSED HIGH-CAPACITY DATA CENTRES

With the needs of AI now driving new data centre infrastructure build outs, the optical interconnectivity technology to support the changing market has to evolve. Technology journalist Antony Savvas looks at the latest developments.

“We raised our full year 2026 forecast from 10% to 16% following standout results in the first quarter of 2026,” says Jimmy Yu, vice president at Dell’Oro Group. “We estimate the optical transport market grew 20% year-over-year in the first quarter of this year, driven by demand for data centre interconnect. However, many key optical suppliers are noting a growing backlog of orders as lead times continue to stretch out. We think supply may be the biggest factor keeping the optical transport market’s growth rate from being even higher this year due to all the AI data centre build- outs,” says Yu.

for higher throughput and lower latency interconnection. Ed d’Agostino, vice president at DE-CIX North America, says: “Our upgrades are about building and reinforcing reliability at scale. With our 4-node architecture spanning New York and New Jersey, and readiness for higher-capacity services, we’re building for how networks are actually evolving: more distributed, more performance-driven, and with greater degrees of visibility and control than ever before.” SUBSEA Lightstorm, a cloud and AI network infrastructure platform delivering high-performance terrestrial and subsea connectivity across the Asia- Pacific region, recently announced the successful quadrupling of client service capacity for its Japan–Guam–Australia (JGA) submarine cable system. JGA is the newest, low-latency subsea cable system connecting Tokyo to Sydney, delivering an advanced Pacific route optimised for cloud, AI, and data- intensive workloads. Powered by Ciena’s WaveLogic coherent optical technology, the move from 100Gbps to 400Gbps client traffic represents a significant increase in efficiency, scalability, and economics for customers supporting hyperscale cloud services, AI training and inference workloads, content distribution, and latency-sensitive enterprise applications. “Upgrading JGA to 400Gbps client traffic is a major milestone to us, delivering AI-ready, cloud-optimised connectivity across the Pacific,” says Amajit Gupta, Lightstorm group CEO and managing director. “As the newest and lowest-latency cable between Tokyo and Sydney, JGA provides a powerful foundation to support cloud and AI architectures.” “Subsea operators like Lightstorm require scalable, high-performance

optical solutions to meet surging cloud and AI bandwidth demands,” adds Amit Malik, vice president and general manager, Asia-Pacific, Japan and India, for Ciena. “With WaveLogic coherent optics, Lightstorm is transforming JGA into a platform capable of delivering 400G services today and to seamlessly scale to 800G services in the near future.” MOBILE EDGE GPU RAN (or AI-RAN) integrates graphics processing units into telecoms networks to process Radio Access Network (RAN) baseband workloads. By relying on massively parallel processing, it replaces traditional purpose-built hardware, allowing operators to run both standard 5G data traffic and edge AI applications on the same shared, programmable infrastructure. Instead of dedicating separate servers strictly to network traffic and separate ones for AI, operators can use shared GPU clusters to monetise the network edge. According to Dell’Oro Group, cumulative AI-RAN revenue is projected to reach $35 billion over the next five years (2026-2030). “AI-RAN is already happening and will scale ahead of 6G. These tools will enhance the RAN,” says Stefan Pongratz, vice president at Dell’Oro Group. He says AI- RAN is expected to become “an important technology enabler” as operators incorporate greater virtualisation, intelligence, and automation into their RAN roadmaps. Amdocs, the provider of software and services to communications service providers (CSPs), has completed a live AI-RAN field-validated blueprint in collaboration with 1Finity, a Fujitsu company, and Supermicro. The deployment showcased next- generation Open vRAN architecture, running 1Finity Open vRAN software on Supermicro ARM-based 1U servers, equipped with NVIDIA GH200 Grace Hopper Superchips and Red Hat

ADDRESSING AI DEMAND As traffic patterns evolve and AI

workloads demand more distributed infrastructure, DE-CIX has upgraded its New York metro platform to a quad- node network architecture, with two core nodes in New York and two in New Jersey. The design delivers higher redundancy, greater resilience, and enables true A/B connectivity strategies across the Hudson River. The upgraded architecture reflects the growing importance of New Jersey as a natural extension of the New York metro’s digital infrastructure. By operating 2-node cores on each side of the Hudson, DE-CIX enables customers to build geographically separated yet tightly integrated interconnection strategies. This supports enterprises, carriers, cloud providers, and AI-driven platforms seeking greater operational control, improved business continuity, and infrastructure diversity. As traffic shifts toward higher-capacity services, DE-CIX’s platform has been upgraded to support growing demand for 400 Gigabit Ethernet ports. It says the rise of GPU-intensive and AI- driven workloads, alongside content distribution and large-scale cloud connectivity, is accelerating the need

8

www.opticalconnectionsnews.com

| ISSUE 44 | Q3 2026

ANTONY SAVVAS DATA CENTRE INFRASTRUCTURE

Openshift software. The end-to-end setup brings together 1Finity radios, a 5G standalone core, and commercial devices in a fully integrated, end-to-end multi- vendor environment, “demonstrating seamless connectivity and reliable performance”, says Amdocs. The capabilities were codified as a Network Workflow in the Amdocs aOS (agentic operating system), enabling global CSPs to deploy and scale capabilities “simply and flexibly”. “Supercharging the network with AI capabilities unlocks significant benefits, from optimising utilisation to enabling edge-based enterprise applications,” says Anthony Goonetilleke, group president of technology and head of strategy at Amdocs. “This AI-RAN blueprint is a critical component in service providers’ transition from automated operations to autonomous, outcome-driven networks. The AI-powered RAN will see AI agents continuously sense network conditions, reason, and take action.” QUANTUM PHOTONICS In the age of AI, the communications industry needs to get more out of its optical hardware, and Quantum Pulse Ventures (QPV) says it can help it do so. It has announced a strategic expansion of its composite pulse photonics platform, to meet the needs for scalable, low loss integrated optical applications, and deliver “unmatched computation fidelity and accuracy” for quantum computers and quantum routers. Quantum Pulse 2.0 (QP2.0) extends the company’s offering beyond optical quantum computing components, with a broad class of photonic integrated circuits, and it is now available. “Until now, photonics has mostly meant point-to-point communication, with processing and decision-making performed electronically after the light was received,” says Ofer Shapiro, CEO and co-founder of Quantum Pulse Ventures. “Next-generation systems break that mould: we are no longer just transmitting light, we are processing it as part of the computation itself. This is a major shift in computing and network architecture.” Quantum Pulse 2.0 is said to represent a major boost for leading- edge quantum computation, quantum networking, optical AI infrastructure, or in-network photonic processing. The QP2.0 platform includes a universal directional coupler that delivers “an order-of-magnitude improvement” in operational fidelity. The updated system promises: • A 10x reduction in qubit requirement • Up to a 10x cost advantage in quantum computers • 4x speed for quantum routers • 10x more accurate polarisation control

Yaron Oz, chief scientist and co-founder of Quantum Pulse Ventures, says: “Once light becomes part of the computational fabric, every physical imperfection inside the photonic circuit directly impacts the accuracy of the operation. This creates an entirely new level of demand for accuracy, precision, stability, and fabrication tolerance in photonic integrated circuits. While we have several options for the physical implementation of qubits, for quantum routing, there is only one viable option, which is light itself. This is why we are expanding our platform to support these applications.” Without needing any changes to existing integrated photonics fabrication platforms, QPV says its composite pulse approach improves operational fidelity and robustness against fabrication variability. This enables “immediate adoption” across current silicon photonics, silicon nitride, thin-film lithium niobate, and related integrated photonics manufacturing processes. AI PHOTONIC SECURITY AI photonic security is an increasingly important consideration. euNetworks has developed Quantum Shield using Adtran’s optical transport technology to augment its broader architecture, which is designed to deliver secure, scalable data centre connectivity across its pan- European network. The new offering is built for enterprises with stringent security, performance and customer-controlled encryption requirements. The deployment combines high-capacity dedicated infrastructure with real-time fibre monitoring and enhanced optical- layer visibility to safeguard critical traffic. By integrating advanced encryption with continuous monitoring across the optical layer, euNetworks says it can deliver the highest levels of protection for sensitive data moving across Europe. This comes at a time when organisations across Europe are accelerating plans to secure data in transit in response to evolving cybersecurity regulations and post- quantum security guidance. The EU’s coordinated post-quantum cryptography roadmap targets migration of high-risk and critical infrastructure environments by 2030, while regulations, including DORA and NIS2, are increasing expectations for encryption, crypto-agility and the protection of sensitive traffic traversing private and third-party infrastructure. euNetworks will offer Quantum Shield to customers as an additional security layer for its Private Connect MOFN solution, which provides private, managed network infrastructure for organisations seeking enhanced security, scalability and control over their data. The addition of private

quantum-safe connectivity provides the security of the organisation’s dedicated fibre, plus quantum-resistant encryption at Layer 1, ensuring all traffic is automatically encrypted. The new infrastructure is built on Adtran’s FSP 3000 optical transport platform, incorporating its S-Flex technology to support high-capacity encrypted data centre interconnect (DCI) services. Marisa Trisolino, CEO of euNetworks, says: “Together, we’re providing connectivity that combines strong security, predictable performance and clear visibility into the underlying network, while customers retain control over how their data is encrypted. As customer expectations continue to evolve, having trusted partners and proven solutions is essential to supporting long-term digital growth across Europe.” THE FUTURE The potential for data centres to be located in space is in the news. MBRYONICS, a global player in satellite optical communications, has developed STARLIGHT, a 25-800G bi-directional coherent optical transceiver. STARLIGHT is purpose-built for space, and provides ultra-high speed, and free space optical links with dual lasers for separate transmit and receive wavelengths. Legacy radio-frequency links impose critical bottlenecks in terms of data rates and transmission speeds, limiting the effectiveness of satellite links. As a member of the STARLight project, which is supported by the EU Chips Joint Undertaking funding of €110m/$127m, MBRYONICS says it has developed its STARLIGHT coherent transceivers to provide “unprecedented connectivity in space”. The tech offering came as MBRYONICS hosted this June’s pan- European STARLight Consortium annual general meeting. The STARLight project brings together a consortium of leading industrial and academic partners to advance silicon photonics (SiPho) and aims to develop application-driven solutions focusing on key industry sectors such as data centres, AI clusters, telecoms, and the automotive market. “Our STARLIGHT coherent transceiver is a vital part of our satellite optical communications platform, providing the global infrastructure needed to power the next generation of direct-to-device, cloud, and AI services,” said John Mackey, CEO of MBRYONICS. As the space economy transitions from standalone satellites to complex, data-heavy mesh networks, Mackey says his firm’s 800G coherent transceiver “represents the critical infrastructure required to move massive workloads off-planet”.

www.opticalconnectionsnews.com

9

ISSUE 44 | Q3 2026

FADY MASOUD NEXT GENERATION COMPONENTS

SHAPING THE OPTICAL CONNECTIVITY FOR THE AI ERA WITH NEXT-GENERATION COMPONENTS AI is reshaping data centre design as GPU clusters scale to hundreds of thousands of accelerators. Ethernet is moving from 400G, 800G, 1.6T and beyond already on the horizon. Meeting this bandwidth demand requires more than faster optics; it requires the right optical architecture for each link, balancing reach, power, cost, latency, density and operational simplicity.

For more than 15 years, coherent optical technology has been the performance benchmark for high- capacity networking. By using advanced modulation, wavelength selectivity and powerful digital signal processing, coherent systems deliver exceptional spectral efficiency and optical reach. They can carry massive amounts of data across metro, long-haul and submarine networks while compensating for fiber impairments that would overwhelm simpler transmission methods. That performance, however, comes with trade-offs. Traditional coherent optics require sophisticated digital signal processors (DSPs), tunable lasers, high- speed modulators and precision optical components. These elements increase cost and power consumption, making full coherent technology ideal for transport and data centre interconnect

(DCI) applications, but overkill for many shorter links inside and around the data centre. Inside the data centre, intensity- modulated direct detection (IM/DD) has long been the preferred choice for short-reach connectivity. Its simpler architecture provides low-power and attractive cost for server-to-switch, leaf-spine, and campus-style links. Fully retimed IM/DD pluggables use DSPs in both transmit and receive paths to clean up signals and ensure broad interoperability, but that processing adds power, heat and latency as speeds rise.

margin. Full coherent optics can overcome those impairments but comes with cost and power penalties. This is where coherent lite is emerging. It brings the benefits of coherent optics into shorter-reach, high-loss data centre and campus environments, but strips the architecture down to the essentials needed for those applications. Rather than optimizing for thousands of kilometers, coherent lite is tuned for links where IM/DD is not enough and transport-class coherent is too much. COHERENT LITE: RIGHT-SIZING COHERENT FOR AI DATA CENTRES Coherent lite depends on purpose- built photonic integrated circuits (PICs) and optimized DSPs. The PIC must integrate lasers, modulators, detectors and supporting optical functions

AI networking is exposing the limitations of both ends of this

spectrum. IM/DD remains efficient, but optical circuit switching, higher connector counts, patch panels and larger cluster fabrics introduce loss and crosstalk that can exceed its practical

Figure 1: Attributes of coherent lite pluggables

10

www.opticalconnectionsnews.com

| ISSUE 44 | Q3 2026

FADY MASOUD NEXT GENERATION COMPONENTS

Figure 2: IMDD optical pluggable architectures

with low insertion loss and efficient modulation in standard QSFP-DD and OSFP form factors. The DSP must deliver the essential benefits of coherent detection while reducing power, latency and complexity through simplified processing, lighter compensation and application-appropriate forward error correction (Figure 1). The result is a new middle ground for optical connectivity. Coherent lite can support high-loss intra-data centre links, campus scale-out and short DCI- style applications without carrying the full cost and power burden of long-haul coherent systems. It gives operators another tool as AI fabrics expand beyond simple rack-scale topologies. LPO: REMOVING THE DSP TO SAVE POWER While Coherent lite addresses links that need more optical performance, another innovation is attacking the power problem from the opposite direction. Linear pluggable optics (LPO) removes the DSP from the pluggable module entirely. Instead of retiming and equalizing the signal inside the optic, LPO relies on high-linearity analog components and the host switch ASIC to maintain signal integrity. This architecture can significantly reduce module power, heat, latency and cost, which is why LPO is attractive for dense AI fabrics where thousands of short links operate inside tight power envelopes. The trade-off is tighter dependency on the host. Without a DSP in the optical module, the electrical channel, PCB traces, drivers, TIAs, modulators and photodiodes must be carefully designed as a system. LPO

works best where reach is short, link conditions are controlled and the switch ASIC can shoulder more of the signal- processing burden. CPO: BRINGING OPTICS TO THE ASIC As data rates move from 1.6 T to 3.2 T and beyond, even highly optimized pluggables face a fundamental challenge: electrical loss between the switch ASIC and the front-panel module. At 224 G and future 448 G electrical lane speeds, long PCB traces consume power, reduce margin and complicate thermal design. Co-packaged optics (CPO) change the physical architecture. Instead of placing optics at the faceplate, CPO integrates optical components into the switch ASIC, reducing electrical trace length and improving power efficiency, bandwidth density and signal integrity. This proximity is essential as switching capacity scales, but it creates a reliability concern: lasers do not like the heat generated near high-power ASICs. That is why CPO architecture uses external laser sources in ELSFP form factors. Moving continuous-wave lasers to the cooler, serviceable front panel improves thermal management and field replaceability while supplying stable light to multiple silicon photonics engines (Figure 2). BUILDING THE RIGHT OPTICAL FOUNDATION FOR AI CONNECTIVITY The AI era is expanding the optical toolkit. Conventional coherent remains essential where reach and performance dominate. Coherent lite adapts coherent

principles for high-loss, shorter-reach data centre and campus links. LPO strips power and latency from short, controlled pluggable connections by eliminating the module DSP. CPO integrates optics with the switch ASIC to support the next leap in switch capacity, enabled by external laser sources that improve reliability and serviceability. Together, these approaches form an innovation continuum, enabling operators to match each link to the right balance of performance, efficiency, cost, and scale.

Fady Masoud is a Senior Director for Solutions Marketing at Nokia focusing on next-generation Intelligent Coherent Pluggable optics (ICE-X) and cloud / data centre interconnect solutions. His area of expertise is the architecture and requirements of next-generation optical networking infrastructure.

www.opticalconnectionsnews.com

11

ISSUE 44 | Q3 2026

PIETRO BERNASCONI FIBRE CONNECTION LOSSES

FIBRE CONNECTION LOSSES MATTER MORE THAN EVER

From AI data centres and advanced sensing systems to biomedical instruments and quantum photonic platforms - where even small losses can degrade performance, reduce sensitivity, or compromise information integrity - the demand for ultra- low-loss fibre connections has become a critical requirement rather than a desirable feature. Yet fibre connectors remain indispensable for building flexible, modular, and maintainable optical systems writes Diamond SA Technical Key Account Manager Pietro Bernasconi .

A look at the mechanisms responsible for insertion loss in fibre connectors, in particular on fibre core misalignment reveals how high-accuracy connector manufacturing processes can impact system performances across the entire spectral range, especially at shorter wavelengths. WHEN LOW LOSS IS NOT LOW ENOUGH In optical networks, power losses can trigger a broad variety of undesired

situations may also be encountered in medical and scientific instrumentation like optical coherence tomography (OCT), distributed fibre sensing, or astronomical interferometry that operate at the limits of detection sensitivity. In all these cases, optical losses are not merely costly, they are true showstoppers. The takeaway message is clear: minimize optical losses from the very beginning, before they propagate into system-level complexity and cost.

the ferrule that encapsulates it. On the fibre side, the key parameters are the eccentricity of the core relative to the cladding, the uniformity of the outer diameter, and, possibly, its oval cross- section. On the ferrule side, the bore that houses the fibre must be precisely centred and sized closely enough to the fibre’s outer diameter to prevent lateral displacement during the glueing process. Each of these tolerances contributes independently, and their combined effect determines the residual axial offset of the fibre cores at the connector interface. For good-quality single-mode fibres operating at 1550 nm with typical MFD of approx. 10 mm, inside state-of-the-art ceramic ferrules the total core offsets typically fall in the range of 0.5–1.0 µm. When two ferrules are mated face-to- face, these individual offsets may push the effective misalignment further and yield losses of about 0.15–0.25 dB. Though such values appear modest, they may not be acceptable in many loss-sensitive systems. The situation deteriorates significantly at shorter wavelengths with much smaller fibres MFDs. There, the same 0.5–1.0 µm offset that could be acceptable at 1550 nm produces losses of 0.8–1.0 dB at 800 nm. This sharp penalty would make standard connector technology inadequate for many emerging applications that operate in the visible and near-infrared range. FIBRE CENTRING IN METAL- CERAMICS FERRULES To reduce core misalignment between mating connectors the manufacturing tolerances of both fibres and ferrules should be tightened. However this approach is often impractical or too costly. A more effective solution relies upon a two-part ferrule: a zirconia ceramic outer shell housing an internal titanium capillary that eventually holds the fibre. By plastically deforming the titanium, the fibre’s core can be displaced and moved to the very centre of the ferrule as shortly described in the following.

THE ORIGIN OF FIBRE CONNECTION LOSSES

consequences. Along any signal transmission line, losses cause

performance degradations that may need to be compensated for through infrastructure upgrades - such as more powerful transmitters, more sensitive receivers, or additional amplifiers - each carrying clear engineering and economic challenges. However, these remedies, though necessary, often give rise to a whole new set of problems related to an increased system energy consumption. Specifically, an increase in the power dissipation triggers thermal management challenges within and around opto- electronic systems that, in a vicious circle, can jeopardise or even undermine the effectiveness of the solutions. These issues have been accompanying the telecom and datacom industries for decades, and despite relevant technical progress, many concerns persist or have even worsened in modern data centres and AI computing clusters that often operate under very tight power budgets. In some other domains, countermeasures are not even an option. Quantum communication and quantum computing systems for instance often rely on the transmission of individual photons or delicate entangled states. In such environments, any loss is irreversible and directly corrupts the information integrity. Other examples based upon single-photon detection range from quantum key distribution (QKD) to fluorescence lifetime imaging (FLIM) and photon-counting LiDAR systems. Similar

Frequently underestimated contributors to optical loss are fibre connectors. These opto-mechanical components are key to systems’ flexibility since they allow optical networks to be modular, reconfigurable, and maintainable. Despite their deceptively simple function, the engineering behind them can be quite complex. For a connector to perform reliably, the cores of the two mated fibres must be mutually aligned with sub-micron accuracy, in a consistent and repeatable manner. Any deviation at this junction introduces insertion losses and, potentially, back- reflection, both of which may cripple the system performance. Achieving low insertion loss at a fibre connector interface requires precise control of two main factors: mechanical alignment of the fibre cores and optical surface quality. Assuming the fibre end- faces are properly polished and clean, the dominant mechanical contributions to loss arise from misalignment between the cores of the two mated fibres, which can manifest as a combination of lateral offset and angular tilt. In practice, lateral offset is the primary driver and typically dominates the loss since its effect scales exponentially with the ratio of displacement to fibre’s mode field diameter (MFD). In a fibre connector arrangement, these misalignments are the cumulative result of manufacturing tolerances across the two most critical components inside a connector body: the fibre itself and

14

www.opticalconnectionsnews.com

| ISSUE 44 | Q3 2026

PIETRO BERNASCONI FIBRE CONNECTION LOSSES

The capillary’s inner diameter is intentionally sized to accept standard manufacturing tolerances, removing the need for tight matching between fibre and ferrule. The core centring process starts with a circular wedge-shaped crimping tool pressed onto the front of the ferrule to plastically deform the titanium and to collapse the capillary wall uniformly around the fibre cladding. Any remaining offset is then precisely measured, before a second arc-shaped crimping tool selectively deforms the metal to nudge the fibre core toward the exact ferrule centre. The adjustment is repeatable and finely controlled, reducing the remaining eccentricity below 0.125 µm. With this technique pioneered and developed by Diamond SA and known as active core alignment (ACA), coupling losses between mated ferrules can be reduced by nearly an order of magnitude. Of course, when mated, two ferrules must be held in position very accurately. This is achieved by inserting them into a snug fit sleeve that suppresses any additional, lateral and/or angular misalignment. NEARLY AS GOOD AS SPLICES By applying ACA techniques to a variety of fibres, it has been possible to reliably and reproducibly demonstrate coupling losses well below the values specified by the international standards IEC 61753-1. As prescribed therein, connectors from a sufficiently large production lot assembled on standard telecom fibres (1310-1550 nm wavelength) and mated to each other generate a loss distribution characterised by average value ILavg ≤ 0.12 dB and a value at the 97% of the cumulative distribution IL97% ≤ 0.25 dB, so-called Grade B connectors. Since these loss values are still not suitable for many applications, higher- grade connectors have been offered based upon top-quality ferrules and fibres with tighter tolerances to achieve losses approximately halved, i.e. ILavg ≤ 0.05 dB and IL97% ≤ 0.15 dB. With ACA technologies and metal- ceramic ferrules, these values can be lowered even further to set new limits: ILavg ≤ 0.02 dB and IL97% ≤ 0.05 dB. Note

that the use of distributions to characterise the loss performance of connections is not simply dictated by the need of a realistic representation of what could be expected in the field, but it is now necessary to obtain a correct assessment of values very close to the accuracy of the measurement system. To verify the correctness of the characterisation method, the stability and accuracy of these results have been independently confirmed by the U.S. National Institute of Standards and Technology (NIST) using instrumentation specifically developed to resolve extremely low insertion losses while still following IEC 61300-3-34 test methodology. These values eventually show that mating two fibres by means of connectors can compete closely with permanent fibre splices in terms of power attenuations. FROM NIR TO UV Driven by many applications in the sensing and quantum photonics fields, the ultra- low loss performances demonstrated with SM fibres at 1550 nm have been extended to shorter wavelengths with fibres with much smaller MFDs and thus intrinsically more sensitive to fibre core misalignments. Since the residual core-to-ferrule eccentricity remains < 0.125 um after ACA regardless of the fibre’s MFD, exceptionally low loss values can be achieved through the visible and down to the near-UV spectral range. For example, ILavg ≤ 0.05 dB and ILavg ≤ 0.10 dB are available at 800 nm and 450 nm, respectively. Moreover, the advantages deriving from a better

centring extends beyond standard fibres and can encompass other specialty fibres such as polarisation maintaining (PM) or photonic crystal fibres (PCF). THE FUNDAMENTAL ENABLER OF FUTURE OPTICAL INFRASTRUCTURES: ULTRA-LOW- LOSS CONNECTIVITY As optical technologies continue to advance, the tolerance for power loss is shrinking across a broader range of applications. Even small connection losses can turn into higher costs, increased power consumption, reduced measurement sensitivity, or compromised system performance. This trend is particularly evident at shorter wavelengths, where conventional connector technologies often struggle to deliver the required levels of efficiency. ACA addresses this issue by dramatically reducing fibre core eccentricity and the resulting coupling losses at the fibre connector interfaces. By enabling connector losses that approach those of permanent splices while retaining the flexibility and serviceability of a connectorized interface, ACA technology removes a long-standing compromise in optical network design. As performance requirements continue to tighten and photonic systems expand into new application areas, ultra-low-loss connectivity is poised to become not just a desirable feature, but a fundamental enabler of next-generation optical infrastructures.

Pietro Bernasconi Technical Key Account Manager, Diamond SA

www.opticalconnectionsnews.com

15

ISSUE 44 | Q3 2026

DOUGLAS CLAGUE HOLLOW-CORE TESTING

RETHINKING TESTING AND CERTIFICATION FOR HOLLOW-CORE FIBRE Fibre‑optic cables deliver extraordinary data transmission capacity, with modern systems sending and receiving hundreds of terabits per second. But the bandwidth and latency demands of today’s AI‑driven data centres are stretching traditional networks to their limit. Enter the transformative potential of Hollow- Core Fibre (HCF), which is moving rapidly from promise to reality as lab research becomes field trials become early operational deployments, writes VIAVI Solutions Fibre Optic Field Solutions Marketing Manager Douglas Clague .

HCF’s ability to guide light through a predominantly air filled core rather than solid glass promises transformative performance gains through lower attenuation, reduced latency, and less signal distortion. However, the same physical properties that enable these benefits introduce new complexities when it comes to testing and certification. WHAT ARE THE BENEFITS OF HOLLOW-CORE FIBRE? Compared with traditional single mode fibre (SMF), HCF offers clear and measurable benefits. Because light travels through air rather than solid glass, HCF achieves significantly lower latency, has lower chromatic dispersion, and its reduced light and glass interaction means nonlinear effects are negligible. In certain wavelength bands, advanced HCF designs have even demonstrated attenuation below 0.1 dB/km. Individually, each of these benefits is attractive, but together they enable and drive specific applications for DCI and AI workloads and next-generation optical transport, where every microsecond and every fraction of a decibel of optical loss can influence overall efficiency and scale. HOLLOW-CORE FIBRE DEMANDS A DIFFERENT TESTING APPROACH As operators consider deploying HCF at scale, the industry must rethink how this new class of fibre is validated. Conventional fibre test methods and assumptions, developed for solid core single mode fibre (SMF), are inadequate when applied directly to HCF. Applying them without adaptation risks inaccurate measurements and misleading conclusions, which undermine the

confidence needed to deploy such a higher-cost infrastructure. HCF’s air guided design reduces interaction between light and glass, which, in turn, alters the optical signatures relied upon by traditional Optical Tome-Domain Reflectometer (OTDR) test techniques. An OTDR maps the condition of an optical fibre by sending short pulses of light into the fibre and analysing the tiny amount of light that returns, using Rayleigh backscatter and Fresnel reflections to identify loss, events, and distance. In HCF, Rayleigh backscatter is much weaker, typically around 14 to 20dB lower than in SMF. Therefore, to be able to perform OTDR measurement, it requires high dynamic at short pulse width. Unlike SMF, where backscattering is relatively uniform, HCF may show non-uniform backscattering along the fibre length due to microstructural variations. Infiltration of ambient air

components around the splice point could generate high reflection that produces different OTDR signatures compared to SMF. Standard OTDR settings may fail to detect key features, requiring different pulse widths and averaging. This complicates OTDR trace interpretation and requires different test settings than for SMF (index of refraction (IOR) and backscatter coefficient factor (K)). Variable backscatter coefficient also means that bidirectional OTDR test and analysis is the only method suitable for characterising fibre and splice losses. These effects make these familiar single‑ended OTDR measurements unreliable if they are interpreted in the same way as regular SMF results. Without selecting test equipment with the correct performance and adapting test methodologies, operators risk either overlooking real impairments or misclassifying benign features as defects.

Regular uni-directional OTDR trace showing the transitions from SMF to HCF, variation of RBS coefficient in the trace and reflective splices

16

www.opticalconnectionsnews.com

| ISSUE 44 | Q3 2026

Page 1 Page 2 Page 3 Page 4 Page 5 Page 6 Page 7 Page 8 Page 9 Page 10 Page 11 Page 12 Page 13 Page 14 Page 15 Page 16 Page 17 Page 18 Page 19 Page 20 Page 21 Page 22 Page 23 Page 24 Page 25 Page 26 Page 27 Page 28 Page 29 Page 30 Page 31 Page 32 Page 33 Page 34 Page 35 Page 36 Page 37 Page 38 Page 39 Page 40 Page 41 Page 42 Page 43 Page 44

Made with FlippingBook Annual report maker