Find out more about the sessions at Market Focus this year.
View the timetable and find out more about the speakers
Find the Market Focus theatre in Pavillion 2 in the Exhibition Hall.
Monday 21st September: Morning session |
Topic: Modules and sub systems (chaired by Abdul Rahim and Michael Lebby) |
| 10:00 -10:15 Strategies for Scale-Across and Scale-Out Networking |
| Tom Williams, Acacia |
 AI is driving tremendous growth in networking capacity with increased focus on reliability and power. The industry faces a challenge to simultaneously scale network bandwidth and drive architectural innovation. As networks are ramping 200G/lane technology, including 1.6T PAM4 and 800ZR/ZR+ for scale-across, development is already underway on 400G/lane for 3.2T PAM4 and 1600ZR/ZR+. Limitations in system cooling, power delivery, and signal integrity are driving proposals for multiple new architectures and form factors in the industry. This presentation will discuss key industry trends and provide insight into which offer the most promise for long term success. |
| 10:20 – 10:45 Optical Circuit Switching: Seeking New Applications in a Growing Market |
| Scott Wilkinson , Cignal AI |
 Optical Circuit Switching has moved beyond the science experiment / limited deployment phase, primarily based on Google integrating the technology into its TPU-based architecture. The TPU architecture will quickly grow the OCS business to a multi-billion-dollar TAM and keep several suppliers busy for many years, but what is the opportunity outside of Google and other TPU users? In this talk, we will cover the current status of OCS technology and the OCS market and provide some insights – and open questions – into what will be the next big opportunity for OCS. |
| 10:40 -10:55 Accelerating CPO Broad Market Adoption |
| Karen Liu, Ciena |
 The surging interest in co-packaged and near package optics in recent months is emerging as one of major technology headlines for 2026. For close followers of the data center networking industry, this is not unexpected as the first high volume deployments of CPO are forecasted to occur later this year. The big question for the industry is – how can the power and density benefits of CPO/NPO be made available beyond the realm of just a couple of huge suppliers? This presentation describes a clear path to broadly accessible CPO/NPO through a combination of standards and highly robust, proven technologies with an ultimate goal of delivering on CPO/NPO value propositions of high density and low power while simultaneously retaining important operational aspects of current pluggable optics ecosystem. |
| 11:00 -11:15 NPO vs. CPO: How to Balance Near-Term Deployment with Future AI Scaling? |
| Pan Cao, Huawei Technologies |
 As exponential AI growth triggers skyrocketing data traffic and unprecedented network scale, traditional pluggable transceivers are reaching their density and thermal limits. The industry stands at a pivotal crossroads: should we leap directly to Co-Packaged Optics (CPO), or adopt Near-Packaged Optics (NPO) as a pragmatic bridge for 3.2T+ interconnects? How do we balance 200G/lane deployments with future 400G/lane evolution? Comparing both architectures across power consumption, packaging complexity, manufacturing yield, reworkability, and multi-vendor ecosystem maturity, this presentation navigates this dilemma with a data-driven assessment charting the commercial deployability, optimal timelines, and techno-economic realities for both technologies. |
| 11:20 -11:35 Challenges and solutions to build components for 1600G+ |
| Julia Larikova, Nokia |
 This session will examine the key challenges in scaling optics to 1600Tb/s and beyond, with a focus on signal integrity and power efficiency, and thermal management. As lane speeds increase, crosstalk, insertion loss, and packaging density become critical barriers to reliable performance. The session will explore advanced solutions, including improved design, optimization of electrical and optical interfaces, and enhanced shielding techniques to mitigate interference. Attendees will also learn about innovative approaches to reduce power consumption and efficiently dissipate heat, such as optimized DSP architectures and low-power front-end optics, and advanced cooling strategies, enabling next-generation high-bandwidth, energy-efficient pluggable modules. |
| 11:40 -11:55 Cooling Smarter: How Advanced Materials Unlock Data Center Efficiency |
| Sanjai Parthasarathi, Coherent Corp |
 As AI and high-performance computing drive unprecedented power densities in data centers, thermal management has become a critical challenge for energy efficiency and sustainability. Traditional cooling solutions based on copper cold plates and silicon substrates are increasingly constrained by thermal conductivity limits, resulting in higher energy consumption and reduced performance. This presentation explores advanced materials, including diamond-based cold plates and silicon carbide (SiC) substrates, which deliver up to 5× the thermal conductivity of conventional materials for superior heat removal. It also highlights thermoelectric technologies that convert waste heat into usable energy, transforming thermal management into a strategic efficiency advantage. |
| 12:00 -12:15 Scaling AI Interconnects with Advanced Optical Connectivity |
| Benoit Fleury, Corning |
 As AI-driven networks push data center performance to new levels, co-packaged optics is emerging as a potential path to higher bandwidth density and improved energy efficiency. Benoit Fleury will present Corning’s perspective on how optical connectivity innovation can support this transition. The discussion will focus on the infrastructure needed to enable scalable, reliable, and manufacturable CPO adoption, including advances in fiber, connectivity, and integration. From Corning’s viewpoint, success with CPO will depend not only on chip and package design, but also on building an optical ecosystem ready for real-world deployment. |
| 12:20 -12:35 Scaling the Backbone Fabric: Physical Infrastructure for the AI Era |
| Giuseppe Rizzelli, Meta |
 The growth of AI traffic is pushing the Backbone into a regime where no single lever—IP capacity, fiber, or site power—is enough on its own; scaling requires moving all three together. In this session, we share our approach across three coordinated fronts. On the IP side, we are moving from chassis-based nodes to a disaggregated 2-tier Clos fabric to lift the per-site capacity by an order of magnitude. From a physical infrastructure perspective, we are redesigning the inside fiber plant around centralized Optical Distribution Frames, pre-deployed on day one to collapse IP-to-optical patching. In the outside plant, we are introducing an in-house designed, vendor-unified Universal ILA hut, standardized for multi-rail growth. Together, these elements create a backbone that is denser, faster to deploy, and built for the future. |
|
|
Monday 21st September: Afternoon session |
Topic: Modules and sub systems (chaired by Bill Ring and Sanjai Parthasarathi) |
| 13:00 -13:15 The Evolution of Optical Interconnects: From Pluggables to XPO and CPO |
| Andy Bechtolsheim, Arista |
 Recent Multi-source agreements focussed on next generation pluggable optics and next generation optical standards have garnered significant industry interest. I will review the outlook for these standards in the context of the rapidly increasing bandwidth demands of AI data centers. |
| 13:20 -13:35 Lessons learned from GW-scale AI deployments |
| Mark Filer, Oracle |
 When you’re training at the scale of 10,000+ GPUs, the “laws of large numbers” become your primary adversary. Single link flaps or other network instabilities can stall a multi-million dollar training job at huge expense to the AI cloud operator. We’ll share some of the hard-won operational learnings from OCI ’s large-scale deployments supporting its top AI customers, including some production statistics on one of the world’s largest deployments of 800G 2xDR4 LPO transceivers and how that translates to deployments at 1.6T. We’ll additionally explore some forward-looking paths toward achieving high uptime and resilience in massive-scale distributed systems. |
| 13:40 -13:55 Optics for NVIDIA Spectrum-X Multiplane Network Architecture |
| Craig Thompson , Nvidia |
 AI factories require consistent, deterministic high-performance networks for scaling to large cluster sizes. Network architectures, including the optical physical layer, have evolved to address this challenge. This talk will review current state of the art for AI network architecture and how it impacts the deployment and roadmap for optical interconnects. |
| 14:00 -14:15 Optical Scale Up Vs. Scale Out Beyond 200G PAM4 |
| Hacene Chaouch, Nexthop AI |
 Finding an alternative to copper has become increasingly urgent for scale-up systems and optics is expected to play a central role. While scaling beyond 200G PAM4 optically appears to have a clear path forward for scale-out networks, meeting the much stricter power, reliability, and cost requirements of scale-up remains a significant challenge. In this presentation, we will review the leading optical technology candidates for the scale-up domain and discuss their trade-offs, maturity, and readiness for deployment at scale. |
| 14:20 -14:35 Architecting the Future: Strategic Sourcing for Scalable Optical Ecosystems |
| Nikita Parikh, Google |
 This presentation explores the critical intersection of cutting-edge optical innovation and global supply chain resilience. As 800G/1.6T deployments accelerate, the industry faces unprecedented pressure on lead times and material sustainability. I will outline a strategic framework for sourcing next-generation network infrastructure that prioritizes multi-vendor interoperability and circular economy principles. By aligning R&D roadmaps with strategic procurement, operators can mitigate geopolitical risks and optimize Total Cost of Ownership (TCO). This session provides actionable insights into building agile, sustainable value chains capable of supporting the massive capacity demands of the AI-driven era. |
| 14:40 -14:55 From Pluggable Optics to CPO: The Emerging Role of Near-Packaged Optics in AI Infrastructure |
| Jianwei Mu, LIGENT |
 The rapid growth of AI clusters is driving optical interconnects toward unprecedented bandwidth density and power efficiency requirements. As link speeds scale to 1.6T and beyond, conventional pluggable optics face increasing challenges in signal integrity, channel loss, power consumption, and thermal management, while Co-Packaged Optics (CPO) introduces significant deployment and serviceability complexities. Near-Packaged Optics (NPO) has emerged as a compelling intermediate architecture that balances performance and maintainability. This talk examines the key bandwidth, noise, thermal, packaging, and reliability considerations of NPO, and discusses its role as a practical migration path for future AI optical infrastructures. |
| 15:00 -15:15 Next-Generation Fiber Technologies for AI Datacenters |
| Jan Vydra, Heraeus Covantics |
 Next generation optical fibers for AI datacenters are redefining performance expectations in bandwidth, latency, and energy efficiency. Building on a long heritage and strong expertise in fused silica technologies for optical fiber manufacturing, Heraeus Covantics contributes to advancing next-generation fiber technologies for emerging interconnect needs. In this market-focused contribution, we will highlight key trends in hollow-core fiber, multicore fiber, and high-density multicore fibers for µLED interconnects. Emphasis will be placed on industry direction, evolving requirements, and collaborative ecosystems, showing how innovation and partnerships enable scalable, future-ready optical infrastructures for AI-driven networks. |
| 15:20 -15:35 Beyond 1.6T: Enabling AI Datacenter Scale with Coherent-Lite Optics |
| Shawn Esser, Coherent Corp |
 As AI datacenter networks scale up, out, and across link speeds are rapidly advancing toward 1.6 Tb/s and 3.2 Tb/s. In this regime, coherent optical transceivers are becoming increasingly compelling relative to traditional IMDD solutions. Their advantages – enabling higher optical power budgets, extended reach, improved spectral efficiency, and more flexible network utilization – position coherent-lite technologies as a strong candidate for next-generation datacenter interconnects. This presentation will explore emerging AI-driven data center network architectures and examine the role that coherent-lite optics can play in enabling scalable, high-performance, and cost-effective interconnect solutions. |
| 15:40 -15:55 Illuminating the Squeeze: How AI’s Super-Cycle and Geopolitics are Rewiring the Optical Component Supply Chain |
| Mingyang Lyu, Omdia |
 Much like the memory chip sector, the AI super-cycle is fundamentally disrupting the optical component market. This unprecedented surge in demand, coupled with intense geopolitical friction, is drastically reshaping the industry’s landscape. This discussion explores the critical upstream variables dictating the future of optical components. We will examine how geopolitical maneuvers impact global supply and demand networks, with a specific focus on material vulnerabilities, including wafers, optical fiber and rare earth elements. Furthermore, we will analyze the fierce capacity competition among optical component manufacturers as they navigate this volatile, high-stakes environment driven by AI infrastructure expansion. |
| 16:00 -17:00 XPO and System Scaling. Panel session hosted by Jose Pozo (Optica) |
    Moderator: Jose Pozo, Chief Technology Officer, Optica
XPO and System Scaling. AI infrastructure is pushing optical connectivity past the limits of current pluggables, and XPO has emerged as one of the leading answers, with more than 100 companies now signed up to the multi-source agreement and live multi-vendor demonstrations shown at OFC 2026. This session opens with a presentation by Andy Bechtolsheim, co-founder of Arista and the architect behind both OSFP and XPO, framing the roadmap and where the architecture is heading. A panel follows, bringing together the system builders and module vendors actually deploying it. The discussion covers integration and thermal challenges, scaling strategies across coherent and IMDD, and what XPO means for data centre design. The aim is a deployment-grounded view of XPO rather than a roadmap alone. Opening presentation: Andy Bechtolsheim, Co-Founder and Chief Architect, Arista Networks Panel: Andy Bechtolsheim, Co-Founder and Chief Architect, Arista Networks System perspective Sunil Priyadarshi, Senior Director, AI Hardware Architecture, Arista Networks System perspective Bilal Riaz, Senior Director, Interconnects Product Line Management, Ciena Coherent XPO modules Rang-Chen Yu, Vice President of Marketing, TeraHop, IMDD XPO modules |
|
|
Tuesday 22nd September: Morning session |
Topic: Components/IC/PIC/Fiber (chaired by Tiger Ninomiya and Guangcan Mi) |
| 10:00 -10:15 Scaling the Optical Future: Technologies Enabling Scale-Up, Scale-Out, and Scale-Across AI Infrastructure |
| Julie Eng, Coherent Corp |
 Artificial intelligence is reshaping the requirements for datacenter and communications infrastructure, driving unprecedented demand for bandwidth, power efficiency, and network scalability. As AI clusters grow in size and complexity, optical technologies are becoming increasingly critical to sustaining performance and enabling future growth. This presentation will explore the key market trends and technology innovations transforming optical networking, from advanced lasers, modulators, and high-speed transceivers to co-packaged optics and optical circuit switching architectures. It will also examine the evolution of coherent optics and transport technologies that are extending the reach and capacity of data center interconnects and communications networks. Together, these innovations are laying the foundation for the next generation of AI-scale infrastructure. |
| 10:20 -10:35 Crossroad again, what optics solution(s) will be prevailing for post 1.6T era? |
| Frank Chang, Source Photonics |
 With AI DC build out for 2026, rapid deployment of 800G is happening with close to 100M ports annually, while 1.6T is emerging soon as clear roadmap to reach prime time in next 2-3 years. The next era of networking is here for post 1.6Tb/s. In order to sustain the AI and cloud‑driven bandwidth surge, the industry enters into another crossroad again with many MSA solutions available, similar to the last situation with transition to 100Gb/s for 10 year ago. From industry perspective, while CPO is commonly considered as a long‑term enabler, many other solutions like NPO, CPX, XPO, 300/400G per lane, high‑order modulation & Coherent lite are proposed. In this talk, we will provide an overview of various development status/readiness, and discuss these high capacity/high density optical solutions per use cases for how to meet AI scaling requirements in supporting AI scale up, scale out and scale across. |
| 10:40 -10:55 High-Stakes Mitigation for Optical Component Players in Scale-Up |
| Mark Lutkowitz, fibeReality, LLC |
 Despite the extraordinary growth opportunity for optics vendors in scale-up, there are still challenges, even with short-reach optics at 400G/lane. An accurate determination of the design plans of both system firms and end-users will be vital to the success of those component suppliers. This presentation will closely examine the prospects for seven industry options: 1) staying longer with 200G/lane; 2) implementing OCI MSA and slow and wide; 3) settling for NPO (socketed); 4) estimating reasonable timeframe for large deployment of CPO; 5) returning to VCSELs at 1060nm; 6) exploring radio over waveguides; and 7) alleviating the threat of copper alternatives. |
| 11:00 -11:15 Co-Packaged Optics: From Industry Promise to Reality |
| Justin Abbott, Coherent Corp |
 As AI clusters scale toward millions of accelerators and unprecedented bandwidth densities, optical interconnect architectures are facing power, cost, and density challenges. While Co-Packaged Optics (CPO) has long been positioned to address these challenges, as deployment becomes a reality, the industry is now entering a phase where deployment readiness, manufacturability, reliability, serviceability, and ecosystem maturity are becoming the primary decision factors. This presentation examines the advantages of the evolving CPO ecosystem and discusses how external laser source strategies may help accelerate adoption for the first generation of large scale CPO deployments, while preserving a migration path toward more highly integrated solutions. |
| 11:20 -11:35 Photonics Packaging for High-Volume Manufacturing in the AI Era |
| Massimo Leo, Jabil |
 As AI data centers scale to 1.6T and beyond, photonics packaging has emerged as both the critical enabler and the key bottleneck for high-volume manufacturing. This talk will address the key challenges and latest advances in photonics packaging that enable large-scale production of photonic solutions deployed in AI data center racks. Drawing on real-world manufacturing experience across NPI and high-volume production, the session will cover the packaging advances driving yield, reliability, and scalability for pluggable transceivers through to near-packaged and co-packaged optics (NPO/CPO), and what the industry must address to meet the demands of 1.6T and 3.2T AI infrastructure at scale. |
| 11:40 -11:55 The Impact of Scale Across on the Coherent Market |
| Andrew Schmitt, Cignal AI |
 Examines the magnitude of the scale across market and the impact of this new application on shipments of line systems and coherent optics. |
| 12:00 -12:15 Trend of large scale Silicon Photonics Interconnect-Break through the bottleneck of mass production |
| Xu Sun, Crealights Technology |
 Compatible with SiPh processes, advanced packaging supports high-density integration of electronic and photonic chips. It shortens RF and power delivery paths to improve power efficiency, while fulfilling requirements for 200G/lane and higher rates as well as high-density integration. Accordingly, SiPh engines with advanced packaging have been extensively researched for high-density applications, such as NPO and CPO. Beyond NPO and CPO applications, can advanced SiPh packaging technologies also deliver substantial performance enhancements and cost benefits for conventional high-speed SiPh modules? Which optical advanced packaging methods can be applied in optical interconnect products in able to break through the bottleneck of SiPh scalable mass production? In this talk, we will discuss the latest optical and electrical co-design results toward low power, low cost and scalable mass production. |
| 12:20 -12:35 EBO MSA: Enabling Reliable and Interoperable Optical Connectivity for AI Data Centers |
| Richard Ward, EBO MSA |
 AI data center networks are moving toward higher bandwidth density, high-radix switching, and higher fiber-count interconnects, creating new challenges for deployment, reliability, and field operation. As fiber counts scale, connector cleanliness, inspection, and repeatable mating become increasingly important operational considerations. Expanded beam optical (EBO) connectivity is gaining industry attention because it can reduce sensitivity to dust and contamination while supporting more resilient high-density optical connections. In response to these market trends, the EBO MSA is developing open, interoperable specifications to enable multi-vendor connector solutions and a scalable supply ecosystem. This presentation will discuss the market drivers, EBO benefits, and the MSA’s role in accelerating adoption. |
| 12:40 -12:55 Laser innovations enabling the new photonics ecosystem optimized for AI system interconnects |
| Loukas Paraschis, Lumentum |
 The increasing adoption of optical interconnects in the AI system has proliferated the number of optical designs, including next generation pluggable modules, embedded, CPO, and more recently NPO, each optimized for specific requirements in scale-up, scale-out, scale-across, and soon also scale-in. Moreover, some of these AI systems already employ or plan to employ some form of OCS. This presentation will review how these evolving AI system requirements have motivated innovations in lasers technology and manufacturing capacity to enable the new AI photonics ecosystem to optimally meet to the AI infrastructure demands. |
|
|
Tuesday 22nd September: Afternoon session |
Topic: New and emerging technologies (chaired by Frank Chang and Mark Watts) |
| 13:00 -13:55 Fireside chat with speakers from Lumentum, Marvell and Coherent |
| Giovanni Barbarossa (Coherent), Rafik Ward (Lumentum), Xi Wang (Marvell) with Jose Pozo (Optica) and Loukas Paraschis (Lumentum) |
   This session will discuss the increasing role of photonics in next generation AI/ML interconnects, the related challenge and opportunities, and plus the related new photonics ecosystem that has emerged. |
| 14:00 -14:15 So Co-Packaged Optics here, let’s now talk about OCI! |
| Rajiv Pancholy, Broadcom |
 Multi-trillion parameter models demand massive aggregate bandwidth to sustain usable inference latencies. While copper interconnects limit coherent clusters to roughly 144 GPUs, optical interconnects enable “Super-Nodes” of 576 GPUs by overcoming physical reach constraints. This presentation will highlight Broadcom’s CPO programs and technology platform, present reliability data to facilitate the adoption of optics into the scale-up domain and address the migration to 400G/lane. It will further describe how Broadcom is working to implement optics into the Scale-up domain and provide an update on the progress of the OCI-MSA consortium that is working to standardize optical links for scale-up connectivity. |
| 14:20 -14:35 Beyond 1000 Lanes: What Optical Architecture Will Enable the Next Generation of Ethernet Switch? |
| Youxi Lin, Huawei |
 AI infrastructure is pushing Ethernet network into a new scaling regime, where switch radix must exceed 1,000 lanes while per-lane bandwidth advances toward 400 Gb/s and beyond. This raises a fundamental question: what optical architecture can sustain the next generation of AI networks? This presentation examines that challenge from a switch-system perspective. It explores how the differing requirements of AI scale-up, scale-out, and scale-across networks influence interconnect choices, and how switch architecture increasingly drives the adoption of pluggable IMDD optics, coherent links, and future high density optical I/O architectures. The session presents a framework for understanding where each approach creates system value as Ethernet switch evolves toward multi-hundred-terabit systems. |
| 14:40 -14:55 CPX: Interchangeable 224G Copper and Optical Interconnects for AI Scale-Up/Scale-Out |
| Matthew Burns, Samtec |
 AI Scale-up and Scale-out interconnects demand both reduced power consumption, cost, and latency and increased bandwidth density, capacity, reliability, and scalability. Interchangeability between copper and optical interconnects simplifies system architectures and offers flexibility in scale-up/scale-out implementations. Field replaceable CPO solutions are a must as well. In this Market Focus presentation, technical experts from Samtec will detail the Si-Fly® HD 224 Gbps PAM4 CPX Systems supporting high-performance, electrically pluggable CPC/CPO solutions. In addition, presenters will summarize the benefits of an interoperable ecosystem of optical engines enabling various AI scale-up/scale-out use cases and existing CPX proof-of-concept platforms. |
| 15:00 -15:15 NPO as a Key Enabling Technology for High-Density Optical Interconnects in AI Infrastructure |
| Zhao Wenyu, CAICT |
 From the perspective of China Academy of Information and Communication Technology (CAICT), telecom operators’ expansion into AI token monetization requires ultra-high performance computing and networking infrastructure. AI token generation and delivery heavily rely on low-latency, high-bandwidth networks powered by advanced optical interconnects. CAICT identifies Near-Packaged Optics (NPO) as a critical technology that offers superior integration density, energy efficiency, and scalability for 6.4T/12.8T-class systems. This presentation highlights NPO’s technical advantages in supporting massive parallel AI workloads and accelerating the development of a robust AI token economy. |
| 15:20 -15:35 IPEC Perspective: High-Density Low-Power Interconnect Technology Standards and Deployment for AI Scaling |
| Yu Xu, Li Xu, IPEC |
 Driven by AI scaling, traditional electrical interconnects face severe power and bandwidth limitations. This presentation reveals the IPEC perspective on high-density, low-power interconnect technologies, focusing on Near-Packaged Optics (NPO) solutions. We explore core technical innovations in optical packaging, analyze industrial market timelines, and highlight the critical role of standardization in mitigating market fragmentation. Ultimately, this presentation demonstrates how IPEC’s open standards accelerate commercial adoption and harmonize the ecosystem to sustain next-generation AI infrastructure. |
| 15:40 – 15:55 Scale Across Architectures: Advancing Coherent Connectivity with Next-Generation Pluggable Optics |
| Josef Berger, Marvell |
 The transition from chassis-based coherent transport systems to compact pluggable optics is transforming how operators scale metro, regional, long-haul, and DCI networks. Driven by network disaggregation, lower cost per bit, improved power efficiency, and growing bandwidth demands, coherent pluggables enable more flexible and scalable architectures. This presentation explores advances in coherent DSPs, photonic integration, and pluggable form factors that support 400G, 800G, and higher-capacity services across diverse applications and reaches. It highlights innovations in performance, power, interoperability, and open interfaces, while examining the roadmap toward next-generation of coherent optics, including 800G and 1.6T ZR/ZR+ solutions and tighter optical-network integration. |
| 16:00 – 17:00 Panel session: Realistic check of AI interconnect scaling solutions for post 1.6T era |
| Scott Wilkinson (Cignal AI), Andy Bechtolsheim (Arista), Mark Filer (Oracle), Craig Thompson (Nvidia), Sanjai Parthasarathi (Coherent), Ryan Yu (TeraHop). Moderated by Frank Chang Source Photonics |
 |
Wednesday 23rd September: Morning session |
Topic: Networking/systems/Service provider (chaired by Stephan Neidlinger and Karen Liu) |
| 10:00 -10:15 Carrier-Grade Optical Networks for Distributed AI Computing: From DCI to Reliable AI Interconnect |
| Dr. Yuyang Liu, China Telecom |
 Generative AI is reshaping traffic patterns and reliability requirements across cloud and carrier networks. This presentation shares a carrier perspective on optical networks for distributed AI computing, evolving from traditional DCI to reliable AI computing interconnect. It will discuss scale-out and scale-up traffic requirements, ultra-high-capacity DWDM/OTN transport, ROADM/WSON-based fast recovery, embedded monitoring, and AI-assisted planning. Based on real network trials, the talk highlights how 800G/1..2T-class transport, C+L-band systems, and sub-50 ms optical restoration can support efficient training, secure inference, and multi-cloud computing, and outlines a roadmap for resilient, energy-efficient optical networks in the Generative AI era. |
| 10:20 -10:35 Scaling ground station connectivity for satellite constellations |
| Jessica Wang , Coherent Corp |
 The rapid expansion of low earth orbit satellites (LEO) is motivating a shift in how terrestrial networks scale to support space-based connectivity. As constellation sizes grow to 10’s of thousands of satellites, the need for efficient ground station interconnect intensifies. This presentation examines the scaling challenges and architectural implications for evolving ground station interconnect and proposes an entirely pluggable optical line system as a path towards supporting high density, power efficient and cost-effective connectivity in next generation Satcom systems. |
| 10:40 -10:55 Driving a 448Gbps Signaling Spec for AI |
| Nathan Tracy, TE Connectivity |
 AI training and inference applications require power-efficient, low-latency, and high-speed interconnects. The industry must go beyond the current OIF Common Electrical I/O (CEI) 224Gbps spec to align on a 448Gbps solution that optimizes electrical interfaces to meet these conflicting requirements. This session provides an overview of the OIF CEI-448Gbps Framework projects that address next-generation architectures and applications. Key challenges including modulation format, target loss budgets, density requirements and packaging options are identified. Learn how OIF is driving interoperable interfaces for AI application spaces to enable cost-effective components, subsystems, and system development and deployment. |
| 11:00 -11:15 Developing AI Optical Interconnect Standards on the AI Timeline |
| Jeff Hutchins, RANOVUS |
 This presentation explores the development of dense, reliable, and energy-efficient interconnect standards for AI, focusing on the evolving needs of large-scale AI compute systems. It addresses the need to develop standards on the AI innovation timeline that meet application requirements and ensure predictable supply chains. The session considers the key technical challenges for optimizing solutions for scale-up and scale-out networking. It also considers ASIC escape density, reliability, cost sensitivity, and other innovations such as bi-direction optics. Finally, it outlines criteria for compelling, standard AI scale-up solutions in this fast-moving domain and highlights OIF’s foundational work for future AI interconnects. |
| 11:20 -11:35 Decoding OIF CMIS to Simplify Network Management for Coherent Pluggables |
| Gary Nicholl, Cisco |
 As the complexity of AI interconnects expands with demands for higher speeds over an ever-increasing number of links, module management capabilities must evolve to meet network scalability, performance, and security requirements. From a module perspective, this complexity is expressed in higher lane counts, more complex configuration options, and advanced telemetry and security features. Learn how the OIF Common Management Interface Specification (CMIS) meets the need for managing current interconnect modules and the plans to support evolving applications. |
| 11:40 -11:55 The Future of Coherent Optics in the AI Era |
| Frank Chang, Ciena |
 AI-driven traffic growth is accelerating demand for higher-capacity, more efficient optical networks, driving the industry toward 1.6T coherent pluggables and beyond. As coherent optics continue to evolve, network operators face important architectural decisions regarding power, density, reach, cooling, and spectral efficiency. This session will explore the diverging paths of coherent optics and examine where pluggables, transponders, and emerging architectures each deliver the greatest value. Topics will include 1.6T and 3.2T coherent innovations, liquid-cooling readiness, XPOs, full-spectrum transponders, and the network requirements shaping future deployments. Attendees will gain practical insight into how optical networks are evolving across multiple parallel vectors to meet the demands of the AI era. |
| 12:00 -12:55 Panel: The Ethernet Evolution of AI |
| Panel organized by John D’Ambrosia, Ethernet Alliance with Kent Lusted (Synopsys), Mark Nowell (Cisco) |
   AI networks are becoming the next great catalyst in Ethernet’s evolution. As back-end scale-up architectures push past the limits of today’s 200Gb/s-per-lane technologies, they’ll demand ultra-high-radix solutions. At the same time, scale-out networks will surge beyond the boundaries of 1.6TbE. The industry is already moving—efforts to define 400Gb/s electrical and optical signaling are underway.Join leading experts from across the Ethernet community to explore how this next leap will shape the intelligent networks of tomorrow—where unprecedented speed and massive scale converge.Moderator: Michael Lebby |
Wednesday 23rd September: Afternoon session |
Topic: Networking/systems/Service provider (chaired by Daryl Innis and Stephan Neidlinger) |
| 13:00 -13:15 Hollow Core Fibre and the Path to the AI Grid |
| Carl Delaney, euNetworks |
 Carrier networks have a built-in AI advantage: metro reach and capillarity. As distributed AI drives agent-to-agent and machine-to-machine communication, bandwidth demand will become less predictable, with strict requirements for latency, jitter and determinism. |
| 13:20 -13:35 Rethinking Optical Interconnect for the AI Agent Era |
| Mingwang Mao, Meituan |
 The rapid growth of AI infrastructure is driving exponential demand for optical interconnect. With AI training compute scaling at 5× per year and context windows expanding 30× annually, networks can no longer evolve incrementally — they must leap forward. This presentation draws on Meituan’s large-scale deployment experience to share technical insights and forward-looking perspectives across three dimensions. First, we present field experience deploying 400G/800G optical interconnects at scale, including system-level LPO validation on our in-house 51.2T whitebox switch, and our contribution to the IPEC eSR multi-mode standard, now widely adopted by industry peers. Second, we compare next-generation interconnect technologies — FRO, LPO, NPO, and CPO — across Scale-up and Scale-out scenarios, and share Meituan’s technology selection rationale for AI-oriented deployments. Beyond hardware selection, we introduce our intelligent optical O&M system, validated across 6,000+ modules in production, achieving 60× faster fault detection and a 200% improvement in detection coverage. Third, we examine the emerging challenges introduced by AI Agent workloads. Unlike traditional LLM inference, Agent workflows generate mixed large/small flows, stateful KV Cache migrations of tens of gigabytes and unpredictable traffic patterns — exposing fundamental limitations in conventional ECMP routing and compute scheduling. We propose network evolution directions including traffic-aware routing, topology-aware scheduling, and optical-electrical hybrid networks leveraging OCS. We conclude that optical interconnect is becoming the core competitive differentiator of AI infrastructure. |
| 13:40 -13:55 From Architecture to Deployment: End-to-End Optical Connectivity Solutions for Scale-Up AI Infrastructure |
| Josh Wilken, Corning |
 Next-gen AI models require optical connectivity for larger GPU nodes, replacing traditional interconnects. AI scale-up architectures are driving unprecedented in-rack fiber density, increasing installation complexity, extending deployment timelines, and elevating connectivity risk. This talk outlines these challenges while positioning Corning as an end-to-end solution provider. Corning’s approach starts with architecture-level understanding and extends through fiber, cable assemblies, structured cabling (e.g., Flexweave), PRIZM® TMT connector platforms, and hardware for blind-mate systems. We also emphasize ecosystem collaboration with system integrators, enabling deployers to tailor solutions that optimize TCO while accelerating scalable, repeatable AI infrastructure deployment across evolving data center environments. |
| 14:00 -14:15 Hollow-Core Fiber at Scale: Operational Insights from Multi-Span Transmission Systems |
| Andrew Kam, Ciena |
 Hollow-core fiber (HCF) is widely recognized for its potential to reduce latency and unlock new levels of optical network performance, yet questions remain regarding manufacturability, scalability, and practical deployment. This session will explore the benefits and deployment considerations of HCF based on multi-year evaluations spanning several generations of fiber technology. Ciena will share results from one of the industry’s most extensive HCF transmission demonstrations, including a multi-span 400 km system built using commercial coherent transceivers and line systems. The session will examine measured performance, system integration considerations, and lessons learned, providing a practical assessment of the progress, opportunities, and remaining challenges on the path toward operational HCF networks. |
| 14:20 -14:35 AI-driven data center investment creates new opportunities for European service providers |
| Stefan Voll, Adtran |
 Global AI investment has triggered unprecedented data center expansion in the United States, reshaping connectivity architectures and contributing measurably to economic growth. From 2026 onward, more than half of all new data center capacity will be built outside the US, with Europe emerging as the primary destination for billions of dollars in capital from international hyperscalers, regional champions, and institutional investors. This session explores how these “AI factories” change connectivity requirements — from AI scale across, high-capacity DC interconnect and cloud on ramps to more distributed metro and possibly even edge deployments for AI inference. We examine the resulting opportunities for European service providers, including roles as infrastructure partners, AI enablement specialists, and vertical-focused solution providers, and how AI assisted operations can help to evolve from reactive to responsive support models, enabling faster resolution of customer issues and requiring fewer truck rolls. |
| 14:40 -14:55 Scaling Optical Reliability with Bandwidth Across AI Clusters |
| Rajan Pai, Credo |
 As AI clusters have grown significantly in size relative to GPU counts and power consumption, so has the optical transceiver count per cluster. The number of link flaps has increased significantly, making the Mean Time Between Failures (MTBF) much more frequent. Even a single unstable link can cascade into widespread disruption. This presentation will address the impact that link instability has on a cluster, the new class of optical transceivers that bring advanced telemetry to predict—and prevent—link flaps before they occur, and the role of standardization in improving the reliability bottleneck as AI scales. |
| 15:00 -15:15 The impact of Multi-Data Center Compute & the Need for Direct Optical Server Interconnection |
| Francois Moore, 1Finity, a Fujitsu Company |
 Many jurisdictions impose power limits on an individual data center and as a consequence, it also limits computing resources at the location. As a solution, we witness the emergence of distributed computing, which combines the resources of geographically separated smaller data centers. This requires a new optical interconnection: A direct point-to-point connection between two compute devices with Remote Direct Memory Access (RDMA) capability, where the far-end server is treated as a memory extension to the near-end server. This is a new optical application for the industry. This presentation will describe this application, outline its key requirements and provide its roadmap. |
| 15:20 -15:35 The Talent Bottleneck: Why Photonics Won’t Scale Without Humans (and How to Fix It) |
| Ilse de Graaf, SMART Photonics |
 While the photonics and optical communications industry is scaling at record speed, talent is not. This talk flips the usual technical lens and focuses on the human infrastructure behind innovation. From competing with Big Tech for engineers to redefining what “qualified” really means, I’ll explore how fast-growing deep tech companies can attract, develop, and retain the talent needed to sustain growth. Expect practical insights, surprising data points, and a few uncomfortable truths delivered with energy. Because without people, even the most advanced technology doesn’t ship. |
| 15:40 -15:55 Beyond Connectivity: Real-World Network Sensing with Existing Optical Infrastructure |
| Helen Xenos, Ciena |
 Optical networks are evolving beyond connectivity into intelligent sensing platforms capable of enhancing security, resiliency, and public safety. This session will examine the growing market drivers for network sensing, including fiber intrusions, submarine anchor incidents, transient network impairments, and environmental events. It will compare different approaches to fiber sensing and describe how sensing capabilities can be enabled using existing optical infrastructure powered by advanced coherent technologies. Ciena will also share real-world results and operational insights from sensing trials conducted across deployed terrestrial and submarine networks, demonstrating the practical value and increasing maturity of network sensing technologies. |
View the timetable and find out more about the speakers