Ethernet & Switching
Switch silicon and systems from 800G to 3.2T and beyond — Ethernet roadmaps, scheduled fabrics, SONiC, and the switching layer of AI and cloud data centers.
159 video interviews · 2019–2026
How Distributed AI Workloads Are Reshaping Network Architecture
- Ethernet is now a first-class citizen in AI networks, for both training and inference on heterogeneous hardware
- One unified fabric targets telcos, colos, hyperscalers, enterprises, and neoclouds
- Fabric is scalable, secured, fully programmatic, and orchestrated on demand across data centers
Full summary
Sanjay Kumar, VP of Products and Marketing, and Keyur Patel, CTO at Arrcus, discuss how the distributed nature of AI workloads is fundamentally changing network architecture, with Ethernet becoming a first-class citizen in AI networks for both training and inferencing across increasingly heterogeneous hardware environments. They explain how Arrcus addresses the needs of telcos, colocation providers, hyperscalers, enterprises, and neoclouds by providing a unified fabric that is scalable, highly secured, fully programmatic, and orchestrated on demand across different data centers.
Optical Interconnects for AI
- Optical engines fabricated on TSMC's platform; multi-chip packages integrate eight engines to replace copper
- Chip-to-system approach scales both GPU and switch connectivity
- Partner ecosystem (FOCI, Browave, Wiwynn, Alchip, GUC) targets 2028 deployment
Full summary
Vishal Chandrasekar, Director of Product Management at Ayar Labs, presents an optical interconnect solution that scales GPU and switch connectivity through a chip-to-system approach, featuring optical engines fabricated on TSMC's platform and multi-chip packages that integrate eight optical engines to replace copper implementations. The solution, developed with partners including FOCI, Browave, Wiwynn, Alchip, and GUC, targets 2028 deployment.
Ethernet for AI Networking at Scale: Building 100K+ XPU Clusters
- Ethernet has become the industry standard for AI networking, enabling clusters of 100,000+ XPUs
- Scale-up, scale-out, and scale-across are all addressed with Ethernet architectures
- Portfolio spans Tomahawk Ultra, Tomahawk 6, Thor Ultra, and Jericho 4
Full summary
Hasan Siraj, VP of Product Management for the Core Switching Group at Broadcom, explains how Ethernet has become the industry standard for AI networking at scale, enabling clusters of over 100,000 XPUs through scale-up, scale-out, and scale-across architectures. He discuses Broadcom's solution portfolio including: Tomahawk Ultra, Tomahawk 6, Thor Ultra and Jericho 4.
Rethinking Data Center Network Architecture
- Ethernet is advancing from 800G to 3.2T with co-packaged optics and scheduling layers that remove lossiness for AI fabrics
- GPU clusters are scaling from 8,000 toward one million units
- DriveNets' scheduled fabric claims better time-to-first-token and lower cost-per-million-tokens than proprietary alternatives
Full summary
In this Director's Cut video, Dudy Cohen, VP of Product Marketing at DriveNets, explains how Ethernet is evolving through advancing speeds (800G to 3.2T), co-packaged optics (CPO) integration, and scheduling layers that address lossiness to support AI infrastructure with GPU clusters scaling from 8,000 to nearly one million units. He describes how DriveNets has adapted its scheduled fabric technology, delivering superior time-to-first-token performance and lower cost-per-million-tokens compared to proprietary technologies.
Ethernet Fabrics for AI Data Centers
- Networking is ~10% of AI infrastructure cost but causes ~80% of deployment challenges
- Scheduled Ethernet fabrics can exceed InfiniBand performance
Full summary
Dudy Cohen, VP of Product Marketing at DriveNets, explains that while networking represents only 10% of AI infrastructure costs, it causes 80% of deployment challenges, advocating for Ethernet-based solutions with scheduling capabilities that can exceed InfiniBand performance. He emphasizes the importance of working with innovative vendors who have deep expertience.
Industry's First 224G Interoperability Testing
- Second 224G Plugfest: 20+ companies testing real product interoperability at Keysight
- Nine compliance stations feed results back into IEEE specification development
Full summary
David Rodgers, Technical Business Development Manager with the Ethernet Alliance, and Sam Johnson, Link Applications Engineering Manager at Intel and High-Speed Networking Chair, discuss the second 224-gigabit Plugfest event at Keysight in Santa Clara, where over 20 companies test actual product interoperability across multiple vendors while nine compliance stations feed critical data back into IEEE specification development.
Rethinking Data Center Network Architecture
- Scale-out is bandwidth-driven (800G to 1.6T+); scale-up fabrics unify xPU pods needing memory sharing and ultra-low latency
- Marvell roadmap: 100T and 200T scale-out fabrics plus 115T and 57T scale-up fabric devices
- Scale-up support spans NVLink Fusion, UALink, and ESUN
Full summary
In an extended discussion, Rishi Chugh, VP and GM, Data Center Switching at Marvell, shares how AI workloads are driving data center networking, distinguishing between bandwidth-driven scale-out networking (800Gpbs to 1.6Tbps and beyond) and scale-up fabrics that unify xPU pods requiring memory sharing and ultra-low latency. Chugh shares Marvell's commitment to deliver 100T and 200T scale-out fabrics along with 115T and 57T scale-up fabric devices supporting NVLink Fusion, UALink, and ESUN.
Linear Pluggable Optics for Data Center Efficiency
- 1.6T optical modules consume nearly double the power of 800G — up to +1kW on a 64-port switch
- Linear Pluggable Optics (LPO) cuts power by leaning on the switch ASIC's DSPs
Full summary
Neel Patel, GM of Optical Networking Component Solutions at Nokia, discusses the power consumption challenges of advancing optical modules in data centers, where 1.6T modules consume nearly double the power of 800G modules, potentially adding 1KW to a 64-port switch. He explains how Linear Pluggable Optics (LPO) addresses this by leveraging switch ASIC DSPs reducing power consumption.
Integration is the Real Race in AI Data Center Networking
- Hyperscaler AI demand is reshaping the data center networking market
- Integration — not any single component — is the real competitive race
- Nokia's portfolio spans switching silicon, DSPs, and Infinera-derived optics
Full summary
Mike Bushong, Vice President of Data Center at Nokia, discusses how AI workloads are fundamentally changing the data center networking market, with hyperscalers driving unprecedented demand and technical challenges. He explains that Nokia's broad portfolio across switching silicon, DSPs, and differentiated optics from the Infinera acquisition positions the company to handle complex integration challenges.
NVIDIA's AI Factory Networking Stack
- AI factories require four distinct network layers, each purpose-designed
- Scale-up on NVLink; scale-out on InfiniBand or Spectrum-X Ethernet
- Scale-across on Spectrum-XGS; storage on the BlueField-4 STX architecture
Full summary
Gilad Shainer, SVP of Networking at NVIDIA, shares why building AI factories requires designing four distinct infrastructure layers—scale-up using NVLink, scale-out using InfiniBand or Spectrum X Ethernet, scale-across using Spectrum XGS, and storage with NVIDIA Bluefield-4 STX architecture.
UCIe Chiplet Connectivity for Performance & Efficiency
- UCIe connects chiplets in-package and package-to-package, including over co-packaged optics
- Bandwidth density is 1-4 orders of magnitude better than PCIe or Ethernet
- Enables composable systems with dynamic resource allocation
Full summary
Debendra Das Sharma, UCIe Consortium Chair at UCIe Consortium, explains how UCIe technology enables chiplet connectivity within packages and package-to-package connections using co-packaged optics for composable systems with dynamic resource allocation. UCIe delivers bandwidth densities with 1-4 orders of magnitude improvement over PCIe and Ethernet.
Meter's Unified Network Stack:
- Meter runs a vertically integrated stack spanning firewalls, switches, access points, PDUs, and cellular equipment
- A cloud backend automates network operations via real-time data processing
- The integrated approach is designed to drive intelligent, automated operational decisions
Full summary
Sunil Varanasi, CTO and Co-Founder of Meter, in this extended interview, shares details on Meter's vertically integrated stack that includes firewalls, switches, access points, PDUs, and cellular equipment. He explains how their powerful cloud backend can automate network operations through real-time data processing for intelligent decisions.
Building at Scale: Ethernet & Optics
- Ethernet for Scale-Up Networking Group 1.0 spec expands racks from 64-128 XPUs to 1K XPUs
- Tomahawk 6 in volume production: first with 64 ports of 1.6Tbps, 200G SerDes and CPO
- New Jericho 4 enables cross-data center cluster deployment; Broadcom joins OCI MSA
Full summary
Hasan Siraj, VP of Product Management in the Core Switching Group at Broadcom, highlights the Ethernet for Scale-Up Networking Group's 1.0 specification and rack expansion from 64-128 XPUs to 1K XPUs. He also announces Tomahawk 6 shipping in volume production as the first to support 64 ports of 1.6Tbps with 200G SerDes and CPO, the company's participation in the OCI MSA, and the new Jericho 4 enabling cross-data center cluster deployment.
Broadcom's 400G/Lane Taurus Optical DSP
- Taurus is the industry's first 400 Gbps-per-lambda PAM4 optical DSP
- Enables low-cost, low-power 1.6T transceivers and a path to 3.2T transceivers
- Targets 200 terabit switch networking systems for AI infrastructure
Full summary
Gang Qiu, Product Line Manager, Marketing at Broadcom, presents Taurus, the industry's first 400 gigabit per lambda PAM4 optical DSP designed to help customers scale AI infrastructure at OFC 2026. The solution enables next-generation low-cost, low-power 1.6 terabit transceivers and paves the path to 3.2 terabit transceivers in 200 terabit switch networking systems.
Ethernet Evolves to Meet AI's Massive Demands
- IEEE 802.3dj project is nearly complete, with future work on 400G signaling
- Ethernet is adapting across electrical chips, optical chips, connectors, cabling, and packaging
- Standards-driven evolution positioned as key to keeping pace with AI demands
Full summary
John D'Ambrosia, Chair, IEEE P802.3dj Task Force and Chair Emeritus of the Ethernet Alliance describes how the industry is advancing through standards including the nearly complete 802.3 DJ project and future 400G signaling. He emphasizes that Ethernet continues to adapt across electrical chips, optical chips, connectors, cabling, packaging, and heat management.
Ethernet's Critical Role for AI Infrastructure
- AI-driven demand is straining supply chains for chips, power, and fiber
- Ethernet Alliance argues Ethernet's reliability track record makes it the only viable AI connectivity solution
Full summary
Peter Jones, Chair of the Ethernet Alliance, presents at OFC 2026 on how Ethernet technology must address unprecedented AI-driven infrastructure demands that are straining supply chains for chips, power, and fiber. He emphasizes that despite these immense challenges, Ethernet's proven track record of reliability makes it the only viable solution for enabling the industry's AI connectivity requirements.
Ethernet Alliance: Ethernet Simply Works
- Six Ethernet Alliance leaders across product, technical, and standards roles deliver one message: Ethernet simply works.
- Ethernet's dependable performance across diverse applications keeps it the default choice as speeds and use cases multiply.
Full summary
Keith Cole, GM & VP of Product Marketing, Peter Jones, Chair, Tim Crawley, Product Marketing Engineer, David Rodgers, Technical Business Development Manager, Anabel Alarcon, Product Line Manager, and Vijay Murthy, Senior Technical Product Manager, from the Ethernet Alliance present a unified message about Ethernet technology's reliability and effectiveness. The presenters collectively reinforce that Ethernet consistently delivers dependable performance across diverse applications.
Ethernet for AI — It Just Works… David Rodgers
- Ethernet Alliance booth demonstrates multi-vendor interoperability of Ethernet for AI
- Interop spans 1.6 terabit solutions alongside 800 gig and 400 gig technologies
Full summary
David Rodgers, Technical Business Development Manager at Ethernet Alliance, presents from the organization's booth at the Optical Fiber Conference 2026 in Los Angeles, where member participants demonstrate interoperability of Ethernet for AI with 1.6 terabit solutions alongside 800 gig and 400 gig technologies.
AI Infrastructure Form Factors & Coherent Transceivers at OFC 2026
- OSFP-RHS form factor targets higher-density AI deployments
- Coherent transceiver technology gaining adoption for GPU-to-switch connectivity
Full summary
Anabel Alarcon, Product Line Manager at EXFO, presents key trends at the Ethernet Alliance booth at OFC 2026, including OSFP-RHS for higher density AI deployments and the growing adoption of coherent transceiver technology for GPU-to-switch connectivity.
Silicon Photonics OCS: 1000x Faster Switching
- iPronics silicon photonics OCS reconfigures 1000x faster than previous solutions
- 32-radix switch ramping in production; 64x64 port OCS chip already fabricated
- Establishing a Thailand production line, with plans to double radix annually
Full summary
Christian Dupont, CEO of iPronics, presents the company's silicon photonics-based optical circuit switches that achieve reconfiguration times a thousand times faster than previous solutions, enabling rapid network switching for improved reliability and performance in data centers. iPronics is establishing a production line in Thailand to meet volume demands, with their 32-radix switch currently ramping up in production and a 64x64 port OCS chip already fabricated, while planning to double radix annually to improve port density and integration.
Multi-Vendor Ethernet Fabric Proving AI Workload at Scale
- Keysight demonstrates Ethernet for AI over a multi-vendor fabric network
- Fabric shown supporting massive AI training and inferencing workloads at scale
Full summary
Vijay Murthy, Senior Technical Product Manager at Keysight Technologies, demonstrates Ethernet for AI at the Ethernet Alliance booth, showing how Ethernet supports massive AI training and inferencing workloads through a multi-vendor fabric network.
1.6T PAM4 TRO Electriccal
- Marvell demos a 1.6T PAM4 TRRO electrical system with RT DSP at 200 Gbps per lambda
- Delivers 21 dB output TCQ and 0.995 linearity with wide-open eye patterns
- Targets high-speed data center, AI networking, and switching fabric interconnects
Full summary
Rittik Shah, Senior Staff Application Engineer at Marvell, demonstrates a 1.6T PAM4 TRRO electrical system at OFC 2025, featuring RT DSP technology operating at 200 Gbps per lambda with 21 dB output TCQ and 0.995 linearity. The demonstration showcases wide open eye patterns designed for high-speed data center, AI networking, and switching fabric interconnect applications.
All-Optical Circuit Switches: 2026 the Breakthrough Year?
- Salience Labs demonstrates a 32-port all-optical switch with a fully non-blocking optical crossbar
- OCS technology targets rack-level scale-up where traditional solutions fall short
- Positions 2026 as a breakthrough year for all-optical circuit switching
Full summary
Jim Finch, Vice President of Commercial Strategy at Salience Labs, presents developments in all-optical switches for 2026, highlighting how OCS technology addresses data center scale-up challenges across racks where traditional solutions cannot meet performance demands. Salience Labs demonstrates a 32-port optical switch with a fully non-blocking optical crossbar.
Why Testing Ensures Market Success
- VeEX leads the tech team for 800 gig to 10 gig Ethernet interconnects at the Ethernet Alliance Interop
- Live interoperability demos, planned a year in advance, validate the technology works as intended
- Interoperability framed as essential to successful market delivery of new technology
Full summary
Keith Cole, GM & VP of Product Marketing at VeEX, leads the tech team for 800 gig to 10 gig Ethernet interconnects at the Ethernet Alliance Interop at OFC, demonstrating successful interoperability testing through live demonstrations planned since the previous year. The testing confirms that interoperability is essential for successful delivery of new technology to the market and validates that the technology works as intended.
Solving High-Speed Data Transmission Challenges
- Ethernet Alliance tackles fast, efficient data delivery via robust testing and verification
- VeEX participates in the Alliance, positioning Ethernet as a reliable high-speed solution
Full summary
Tim Crawley, Product Marketing Engineer at VeEX, explains how the Ethernet Alliance addresses the critical challenge of delivering data between links quickly and efficiently through robust testing and verification solutions. He emphasizes that as demand for high-speed data transmission increases, Ethernet proves to be a reliable solution, and VeEX's participation in the Alliance.
AI Data Centers & 1.6T Testing
- VIAVI's award-winning D2 enables 802.3dj 1.6T testing with GPU emulation capabilities
- GPU emulation supports testing next-generation AI application workflows
- VIAVI collaborating with vendors on 1.6T and Ultra Ethernet interoperability
Full summary
Steve Rumsby, Senior Director of Platform and Architecture at VIAVI, presents the company's award-winning D2 product that enables 802.3dj 1.6T technology with GPU emulation capabilities for testing next-generation AI application workflows. Viavi is collaborating with vendors on interoperability to ensure successful deployment of 1.6T technology and Ultra Ethernet.
1.6Terabit Ethernet Demos
- 1.6 terabit Ethernet demonstrated using 200G-per-lane technology
- Multi-vendor demo includes Arista, Cisco, HPE, Keysight, and Teledyne LeCroy equipment
- Covers DSP-based, LPO, and LRO optical interconnect configurations
Full summary
Dave Estes, Hardware Engineer at VIAVI Solutions, presents demonstrations at the Ethernet Alliance booth at OFC 2025, showcasing the evolution toward 1.6 terabit Ethernet capabilities using 200G per lane technology with equipment from major vendors including Arista, Cisco, HPE, Keysight, Teledyne LeCroy, and others. The demonstrations feature various optical interconnects including DSP-based, LPO, and LRO configurations.
Ethernet Dominance, Multi-Site Clusters & Co-Packaged Optics
- Ethernet becomes the standard for scale-up networking
- AI clusters expand across multiple data centers with Ethernet as the interconnecting fabric
- Co-packaged optics go mainstream on the back of 70% power efficiency gains
Full summary
Hasan Siraj, Head of Software Products/Ecosystem at Broadcom, presents three networking predictions for 2026 centered on AI infrastructure, including Ethernet becoming the standard for scale-up networking, AI clusters expanding across multiple data centers with Ethernet as the interconnecting fabric, and co-packaged optics going mainstream due to 70% power efficiency gains. Broadcom is investing heavily in these technologies while collaborating with customers and partners to enable these infrastructure transitions.
GPUs diversity and Ethernet in Scale-Up
- The GPU market grows more heterogeneous, moving beyond a single dominant vendor
- Networking becomes a critical AI infrastructure design consideration
- Ethernet emerges as the leading technology across all connections, from scale-up to scale-out
Full summary
Dudy Cohen, VP of Product Marketing at DriveNets, presents three predictions for 2026 AI infrastructure: a more heterogeneous GPU market, networking becoming a critical design consideration, and Ethernet emerging as the leading technology across all AI infrastructure connections (from scale-up to scale-out).
Ethernet's Path to AI Dominance
- Ethernet will succeed in AI applications despite uncertainty over the specific path forward
- The industry understands its objectives and will determine the approaches needed to meet customer requirements
Full summary
Peter Jones, Chair of the Ethernet Alliance and Cisco Systems employee, presents predictions for Ethernet in AI applications, expressing confidence that Ethernet will succeed despite uncertainty about the specific path forward. He emphasizes that while the exact methods remain unclear, the industry understands its objectives and will determine the necessary approaches to meet customer needs.
The Future of Ethernet: 200G to 3.2T and Beyond
- The 802.3dj standard nears completion, introducing 200 Gbps signaling for speeds up to 1.6T Ethernet, with products already shipping
- The industry moves to define 400 Gbps signaling as the foundation for 3.2T Ethernet and beyond
Full summary
John D'Ambrosia, Chair of the IEEE P802.3dj Task Force at IEEE, discusses the future of Ethernet, focusing on the completion of the 802.3dj standard introducing 200 Gbps signaling for speeds up to 1.6 terabit Ethernet, with products already shipping. He highlights the industry's move toward defining 400 Gps signaling as the foundation for next-generation speeds including 3.2Tbps Ethernet and beyond.
OIF on 448G, Coherent Lite, CMIS & Power Efficiency
- The industry pushes toward 400/448 Gbps implementation while tackling reliability, power efficiency, and reach
- Coherent lite emerges as a developing market segment
- CMIS becomes critical firmware infrastructure for AI applications
Full summary
Nathan Tracy, President of OIF, discusses 2026 predictions focusing on the industry's push toward 400/448 Gbps implementation while addressing reliability, power efficiency, and reach challenges, with particular emphasis on coherent lite market development and CMIS as critical firmware infrastructure for AI applications.
AI Networking: Open Standards & 800G Switches Proven Through Live Testing
- Arista's new AI networking lineup features 800 gigabit switches and its latest XPO optics
- Open standards are central to Arista's AI networking strategy
Full summary
Alex Nichol, Principal Engineer at Arista Networks, showcases the company's new AI networking equipment including 800 gigabit switches and latest XPO optics while emphasizing their commitment to open standards.
Arista's Vision for Agentic AI in Enterprise Networks
- AI is reshaping the enterprise WAN, and Arista is positioning its cloud platform to meet those emerging demands.
- Arista's cloud platform embeds AI for network troubleshooting and application management.
- The approach spans both wired and wireless environments for unified operations.
Full summary
Kishan Ramaswamy, Director, Product Management and Brendan Gibbs, Vice President, AI Routing & Switching Platforms at Arista Networks share their views on the impact of AI in the enterprise WAN and also touch on how Arista's cloud platform integrates AI capabilities for network troubleshooting and application management while supporting enhanced operations across wired and wireless environments.
Aviz on Scaling AI Workloads with Open Networks
- Aviz Networks targets GPU resource allocation and monitoring across multi-tenant AI environments.
- Partners with Nvidia on the Spectrum X platform for deployment and analytics.
- Supports both Nvidia and AMD GPU infrastructures on SONiC-based networks.
Full summary
Vishal Shukla, Founder and CEO of Aviz Networks, leads the company's development of network management solutions that enhance GPU resource allocation and monitoring across multi-tenant environments. Through their collaboration with Nvidia on the Spectrum X platform, Aviz Networks delivers comprehensive system deployment capabilities and analytics for both Nvidia and AMD GPU infrastructures in SONiC-based networks.
Broadcom’s Ram Velaga on Scaling AI Networks
- Broadcom argues networking must span multiple scaling domains to support AI infrastructure.
- Positions Ethernet as the optimal networking solution for AI workloads.
- Cites Ethernet's clean interface, proven reliability, and consistent bandwidth gains.
Full summary
Ram Velaga, GM and SVP of Core Switching Group at Broadcom, discusses how networking must evolve to support expanding AI infrastructure and machine learning systems across multiple scaling domains in data centers. He explains why Ethernet technology stands out as the optimal networking solution for AI infrastructure, highlighting its clean interface, proven reliability, and consistent bandwidth improvements that align with industry needs.
d-Matrix on AI Efficiency and Scale
- d-Matrix's Jetream IO accelerator pairs with its Corsair compute accelerator for multi-node scaling.
- PCI cards support up to eight Corsair accelerators per node.
- Cross-node scaling is achieved through an Ethernet switch.
Full summary
Sid Sheth, CEO of d-Matrix, introduces their Jetream IO accelerator product that works alongside their Corsair compute accelerator to enable multi-node AI workload scaling. The solution uses PCI cards supporting up to eight Corsair accelerators per node, with the Jetream IO accelerator enabling cross-node scaling through an Ethernet switch.
Hedgehog on Open Networking for AI Data Centers
- Hedgehog's open-source networking software runs on whitebox switches at significantly lower cost.
- Automates network operations for AI data centers.
- Supports AI accelerators beyond Nvidia, including AWS and AMD.
Full summary
Marc Austin, CEO of Hedgehog, showcases the company's open-source AI networking software that delivers superior performance on whitebox switches compared to traditional solutions at significantly lower costs. The solution enables automated network operations while supporting diverse AI accelerators beyond Nvidia, including offerings from AWS and AMD.
NeuReality’s 1.6T AI NIC — Moshe Tanach
- NeuReality's 1.6T NIC lifts GPU active time from 16% to nearly 80%.
- Slated for release in late 2026 with in-network compute and Ultra Ethernet support.
- Positioned as a low-latency alternative to InfiniBand.
Full summary
Moshe Tanach, Co-Founder & CEO of NeuReality, unveiled their upcoming 1.6 terabyte NIC product that enhances GPU efficiency in AI workloads by improving active time from 16% to nearly 80%. The new NIC, set for release in late 2026, features in-network compute capabilities and Ultra Ethernet support, providing an alternative to InfiniBand while maintaining low latency performance.
UALink - Accelerating AI Interconnect Innovation
- UALink Consortium has over 110 members implementing its specs in switches and accelerators.
- The open standard builds on existing Ethernet infrastructure.
- AWS, Google, Meta, and Intel are cited as deploying UALink in data centers.
Full summary
Kurtis Bowman, Chairman of UALink Consortium, outlines how their open standard technology meets scaling requirements with over 110 members implementing specifications in switches and accelerators. The technology builds on existing Ethernet infrastructure while enabling companies to maintain their focus, with major tech firms like AWS, Google, Meta, and Intel actively deploying UALink solutions in their data centers.
AI Infrastructure Evolution
- AI is progressing from training toward distributed inference, and networking fabric must follow the workload.
- Arrcus offers IPsec offloading and cost-management for AI data centers atop a multi-platform fabric and OS.
- Ethernet's role is expanding to connect GPU stacks all the way out to edge computing.
Full summary
Shekar Ayyar, CEO of Arrcus, outlines how networking infrastructure supports AI's progression from training to distributed inference applications through their multi-platform networking fabric and operating system. He discusses Arrcus's IPsec offloading for AI data centers and cost management solutions, while highlighting Ethernet's expanding role in connecting GPU stacks to edge computing.
Scheduled Ethernet Fabric for Data Center Infrastructure
- DriveNets Network Cloud AI uses scheduled Ethernet fabric to eliminate jitter and packet drop at low latency.
- The fabric connects thousands of GPUs through a single hop.
- It runs on cost-effective white box hardware rather than proprietary switches.
Full summary
Dudy Cohen, VP, Product Marketing at DriveNets, outlines how data centers can support AI workloads through advanced networking solutions that eliminate jitter and packet drop while maintaining low latency. His showcases the DriveNets Network Cloud AI solution, which uses scheduled Ethernet fabric to connect thousands of GPUs through a single hop while operating on cost-effective white box hardware.
Ethernet Matches InfiniBand for AI Clusters
- Juniper's AI-lab and hyperscaler benchmarks show Ethernet matching InfiniBand performance in AI clusters.
- Leading AI providers pick Ethernet for its broad ecosystem and cost advantages.
- Power efficiency is pursued through liquid cooling and advanced switch designs.
Full summary
Amit Sanyal, Senior Director of Data Center Product Marketing at Juniper, shares how Ethernet performance matches InfiniBand in AI clusters through benchmarks from Juniper's AI lab and major hyperscalers. His analysis shows how leading AI providers select Ethernet for its extensive ecosystem and cost advantages, while highlighting their focus on power efficiency through liquid cooling and advanced switch designs.
Active Electrical Cables for AI Server Architecture
- AI servers are reshaping data center architecture via new switch placement and rack configurations.
- Marvell's 7-meter 28-gauge active electrical cables support 800G breakout connections.
- Its 1.6T product uses 32-gauge cables optimized for GPU-to-GPU connectivity.
Full summary
Winnie Wu, Senior Director of Product Marketing at Marvell, describes how data center architectures are adapting to AI server needs through changes in switch placement and rack configurations. She outlines their 7-meter 28-gauge cables supporting 800G breakout connections and their 1.6T product with 32-gauge cables optimized for GPU-to-GPU connectivity.
Network Architecture for Scaling AI
- NVIDIA spans scale-up and scale-out with GB200 NVL72, InfiniBand, and the Spectrum-X Ethernet platform.
- Spectrum-X, its optimized Ethernet offering, is positioned to support 100K+ GPU clusters.
- NVIDIA embraces both InfiniBand and Ethernet rather than betting on a single fabric.
Full summary
Kevin Deierling, Senior Vice President of Networking at NVIDIA, outlines the company's scale-up and scale-out networking technologies, focusing on their GB200 NVL72 architecture, InfiniBand offerings, and Spectrum-X Ethernet platform. He discusses scale-up and scale-out scenarios, highlighting how their optimized Ethernet offering can support 100K+ GPU clusters.
SNEAK PEAK on How AI Reshapes Data Center Networks
- AI workloads are pushing data center networks toward flatter architectures with far higher reliability.
- AI training's massive power draw and long runtimes make network failures extremely costly.
- The industry is rethinking traditional Ethernet and splitting designs into distinct front-end and backend networks.
Full summary
Rishi Chugh, VP and GM, Data Center Switching at Marvell, joins Roy Chua, Principal and Founder of AvidThink, to explore how AI workloads are fundamentally changing data center networking by demanding flatter architectures and unprecedented levels of reliability. The discussion covers how AI training's massive power consumption and time requirements make network failures extremely costly, driving the industry to reconsider traditional Ethernet approaches and adopt more complex designs with distinct front-end and backend networks.
Building the Next Gen AI RAN
- Arrcus unveiled the TGAX telco-grade AI switch.
- New partnerships span Broadcom, Liberty Global, and Fujitsu.
Full summary
Shekar Ayyar, CEO of Arrcus, announced multiple strategic initiatives at MWC 2024, including the TGAX Telco-grade AI switch and partnerships with industry leaders like Broadcom, Liberty Global, and Fujitsu.
Scale Out, Scale Up
- GPU and XPU connectivity is driving 10x to 100x more network bandwidth than historically seen.
- New certifications cited at 24 gigs equating to 1.6T for switches and transceivers.
- 448 gig demos target both scale-out and scale-up, delivering 10x the bandwidth of scale-out domains.
Full summary
Alan Weckel, Founder and Technology Analyst at 650 Group, discusses the significant bandwidth increases observed at OCP 2025, where networks are experiencing 10x to 100x more bandwidth than historically seen due to GPU and XPU connectivity requirements. He highlights new certifications including 24 gigs equating to 1.6T for switches and transceivers, along with demonstrations of 448 gigs technology for both scale-out and new scale-up applications that deliver 10x the bandwidth of scale-out domains.
Smarter Fabrics for AI
- Arrcus ACE AI fabric delivers congestion-free, lossless Ethernet spanning training to edge inferencing.
- New collaboration with Quanta Cloud Technologies enables Tomahawk 5 switches for AI-ready racks.
- Turnkey rack offerings aim to cut deployment friction for AI at scale.
Full summary
Sanjay Kumar, VP of Products and Marketing at Arrcus, presents the company's ACE AI distributed networking fabric that provides congestion-free, lossless Ethernet connectivity for AI workloads spanning from data center training to edge inferencing. Kumar announces a collaboration with Quanta Cloud Technologies to enable their Tomahawk 5 switches for AI-ready rack solutions, creating turnkey offerings that reduce deployment friction for customers implementing AI at scale.
How to Maximize GPU ROI in AI Infrastructure Buildouts
- Cornelis targets weak AI infrastructure ROI across compute cycles, power efficiency, and space utilization.
- Its technology increases GPU utilization to improve the economic model.
- Current 400 gig tech already offers Ultra Ethernet features like credit-based flow control and adaptive routing.
Full summary
Lisa Spelman, CEO of Cornelis Networks, discusses how companies struggle to extract sufficient value from their AI infrastructure investments, particularly regarding compute cycles, power efficiency, and data center space utilization to make their economic models viable. She explains that Cornelis Networks addresses these economic challenges by developing technology that increases GPU utilization and delivers enhanced performance, with their current 400 gig technology already providing ultraethernet compliance features like credit-based flow controls and adaptive routing.
OCP-Compliant AI Inference at Rack Scale
- d-Matrix showcases a rack-scale AI inference accelerator using ODSA, Bunch of Wires, and block floating-point numerics.
- New SquadRack product, built with Arista, Super Micro, and Broadcom, follows OCP Open Rack Spec V3.
- Roadmap adds support for EON, the Ethernet scale-up network.
Full summary
Max Sbabo, Senior Staff Engineer at d-Matrix, presents the company's rack scale AI inference accelerator at the OCP Summit, highlighting their collaboration with OCP through ODSA workgroups, Bunch of Wires standard implementation, and pioneering block floating-point numerics. He announces d-Matrix's SquadRack product developed with Arista, Super Micro, and Broadcom following OCP's Open Rack Specification V3 standards, along with support for the new EON (Ethernet scaleup network) on their roadmap.
Why High-Performance Networks Choose Ethernet Now
- DriveNets argues high-performance Ethernet now surpasses InfiniBand via scheduled fabric.
- Ethernet is moving into the demanding scale-up domain with better low latency and predictability.
- The pitch: Ethernet is winning leadership across all AI networking use cases.
Full summary
Dudy Cohen, VP, Product Marketing at DriveNets, discusses Ethernet's growing dominance across all networking use cases at the OCP Global Summit, explaining how modern high-performance Ethernet solutions now surpass InfiniBand performance through technologies like scheduled fabric. He emphasizes that Ethernet is successfully pursuing leadership in the demanding scale-up domain with its improved low latency and predictability capabilities, making this an exciting time for AI infrastructure networking.
HPE Demos Ultra Ethernet Transport & RoCE v2
- HPE demoed a live Ultra Ethernet Consortium spec implementation at the OCP Innovation Village.
- QFX 50240 switches handle Ultra Ethernet Transport and RoCE v2 traffic simultaneously.
- A packet trimming feature is used for congestion management.
Full summary
Mahesh Subramaniam, Sr. Director of Product Management at Hewlett Packard Enterprise, demonstrates the Ultra Ethernet Consortium specification implementation at the OCP Innovation Village, showcasing advanced networking technologies for AI data centers. The demonstration features Hewlett Packard Enterprise's QFX 50240 switches handling both Ultra Ethernet Transport and RoCE v2 traffic simultaneously, along with an advanced packet trimming feature for congestion management.
Transforming Data Center for AI with HPE
- HPE positions AI-native networking built from the ground up for AI operational challenges.
- Innovations include liquid-cooled switches and Ultra Ethernet solutions with packet trimming.
- Solutions support OCP-specified OB3 racks, underscoring an open networking commitment.
Full summary
Amit Sanyal, Senior Director, Data Center Product Marketing at HPE, presents the company's AI native networking solutions at OCP, which are designed from the ground up to solve operational challenges and optimize networking performance for AI applications. HPE showcases key innovations including liquid cooled switches, UltraEthernet consortium solutions with packet trimming capabilities, and support for OCP-specified OB3 racks, reinforcing the company's commitment to open networking solutions purpose-built for AI.
Lumentum Optics Powering AI Infrastructure
- Lumentum tackles network power consumption with optical circuit switches and external laser sources for co-packaged optics.
- Portfolio includes 1.6T partial retimed optics modules.
- Lumentum participates in the OCP OCS working group to align vendors and hyperscalers on common standards.
Full summary
Michael DeMerchant, Sr. Director, Product Line Marketing at Lumentum, explains how the AI infrastructure buildout drives hyperscalers to focus on optical technologies, with Lumentum addressing network power consumption challenges through optical circuit switches, external laser sources for co-packaged optics, and 1.6T partial retimed optics modules. DeMerchant also highlights Lumentum's participation in OCP OCS working group standardization efforts to bring industry vendors and hyperscalers together for common standards that accelerate technology adoption.
Marvell's CPO Switch Demo
- Marvell's CPO reference platform is a liquid-cooled 1 OU system with 16 6.4T silicon-photonics light engines.
- Scaling to 32 units per rack means managing over 36,000 fibers.
- Rack-level success hinges on industry solutions for backplane design and fiber routing.
Full summary
Kishore Atreya, Sr. Director, Platform Product Management at Marvell, presents the company's co-packaged optics reference platform featuring a liquid-cooled 1 OU system with 16 6.4T light engines that integrate silicon photonics technology for electrical-to-optical signal conversion. He addresses the significant deployment challenge of managing over 36,000 fibers when scaling to 32 units per rack, emphasizing the need for industry solutions in backplane design and fiber routing for successful rack-level implementation.
AI Needs Fully Photonic Networks - No Electronic Switching
- Oriole Networks argues AI infrastructure demands a full rethink with far more optical components.
- Its thesis: AI networks must be fully photonic, eliminating electronic packet switching entirely.
- The goal is meeting the demanding performance requirements of modern AI workloads.
Full summary
Joost Verberk, VP of Marketing and Business Development at Oriole Networks, explains how AI infrastructure requires a complete rethinking of network architecture with significantly more optical components than current implementations. He emphasizes that Oriole Networks believes AI-focused networks must be fully photonic, eliminating electronic packet switching entirely to meet the demanding performance requirements of modern AI workloads.
UALink for Scale-Up AI Interconnects
- UALink is pitched as a scale-up fabric with lower power, reduced latency, and efficient data movement.
- It achieves nearly 200 Gbit/s on a 200 Gbit/s link.
- Initial devices are expected in 2026, with widespread data center deployment in 2027.
Full summary
Kurtis Bowman, Chairman of UALink Consortium, presents UALink as a scaleup fabric solution that offers technical advantages including lower power consumption, reduced latency, and efficient data movement achieving nearly 200 Gbits per second on a 200 Gbit per second link. He reports that consortium members have IP available for designing switches and accelerators, with initial device availability expected in 2026 and widespread data center deployment anticipated in 2027.
VIAVI's 800G AI Network Testing
- VIAVI's B3 and M1 appliances generate 800-gigabit traffic with smaller footprint and lower power than traditional methods.
- A Juniper collaboration demonstrates real-time network performance testing.
- VIAVI showcases early Ultra Ethernet Consortium specs, enabling piece-wise testing of emerging layers and 800-gigabit NICs.
Full summary
Kevin Chang, Engineering Director, Hardware Platforms at VIAVI, presents advanced Ethernet test appliances at the OCP Global Summit that provide efficient testing for AI networks through B3 and M1 appliances generating 800-gigabit traffic with smaller footprint and lower power consumption than traditional GPU data center methods. VIAVI collaborates with Juniper to demonstrate real-time network performance testing and showcases early Ultra Ethernet Consortium specifications, enabling piece-wise testing of emerging network layers and 800-gigabit NICs as technologies move from specifications to hardware implementations.
Next-Gen Optical from VCSELs to CPO
- Broadcom debuts 200Gbps per lane VCSEL technology plus production-ready EML lasers for AI networks.
- Its co-packaged optics reduce power usage by 70%.
- The Tomahawk 5 Bailly 51T Ethernet switch connects to Thor 2 NIC for XPU-to-switch optical connectivity.
Full summary
Manish Mehta, VP Marketing, Optical Systems Division of Broadcom, presents the company's optical interconnect advances for AI networks at OFC, including their groundbreaking 200Gbps per lane VCSEL technology and production-ready EML laser solutions. The presentation showcases how Broadcom's co-packaged optics reduce power usage by 70% while enabling enhanced integration through the Tomahawk 5 Bailly 51T Ethernet switch connected to Thor 2 NIC and XPU-to-switch optical connectivity solutions.
800G Testing Challenges & Solutions for High-Speed Networks
- EXFO's Anabel Alarcon covers 800G advances and interconnection solutions at OFC 2025.
- 800ZR technology enables transmission up to 120 kilometers for metro and data center links.
- Thorough testing and thermal management grow critical as AI drives cooling requirements.
Full summary
Anabel Alarcon, Product Line Manager at EXFO and High-Speed Tech Lead at Ethernet Alliance, discusses the advancement of 800G technologies and interconnection solutions at the OFC 2025 booth. She explains how 800 ZR technology enables data transmission up to 120 kilometers for metro internet and data center connections, while highlighting the importance of thorough testing and thermal management as AI applications drive cooling requirements.
Breaking Records: 448G/Lane Demo with PAM-4
- Keysight demonstrates 448 Gb/s optical transmission using 224 Gb PAM-4 signaling and an EML.
- The setup integrates 120 GHz bandwidth components.
- A Digital Communications Analyzer enables real-time monitoring of signals up to 480 Gb/s.
Full summary
Fabio Pittalà, Product Planner at Keysight Technologies, showcases a high-speed optical transmission system that achieves 448 Gb/s using 224 Gb PAM-4 signaling through specialized signal generation equipment and an electro-absorption modulated laser. The advanced setup integrates 120 GHz bandwidth components and a Digital Communications Analyzer to enable real-time monitoring of signals reaching up to 480 Gb/s.
Next-Gen Switching: Co-Packaged Optics vs. Co-Packaged Copper
- Marvell showcases a co-packaged copper switch-tray design using substrate-mounted flyover wires at 224G.
- The liquid-cooled solution enables direct XPU ASIC to CPO engine connections.
- Supports large-scale multi-rack clusters up to 2,000 nodes with a single switching layer.
Full summary
George Hervey, Principal Architect at Marvell, showcases an innovative co-packaged copper design for switch trays that enhances passive copper performance through substrate-mounted flyover wires and 224G compatibility. The liquid-cooled solution enables direct XPU ASIC to CPO engine connections and supports large-scale multi-rack compute clusters up to 2,000 nodes with a single switching layer.
Multi-Vendor Demo: 800ZR, ,400ZR and Multi-Span Optics
- OIF's Karl Gass leads a coherent pluggable optics demo across three simulated data centers.
- The demo combines multiple module vendors, traffic generators, and orchestrators.
- Shows interoperable 100G, 400G, and 800G Ethernet across form factors and line systems.
Full summary
Karl Gass, Physical and Link Layer Working Group Optical Vice Chair at OIF, leads a comprehensive demonstration of coherent pluggable optics featuring multiple Ethernet implementations across three simulated data centers. The demonstration incorporates various module vendors, traffic generators, and orchestrators, showcasing interoperable solutions for 100G, 400G, and 800G Ethernet applications using different form factors and line systems.
Next-Gen 224G & 448G/Lane Electrical Connectivity
- OIF's Michael Klempa demonstrates 112G and 224G connectivity while examining 448G requirements.
- Testing spans multi-vendor interoperability across linear and RTLR systems.
- Analyzes signal characteristics and performance metrics across connectivity configurations.
Full summary
Michael Klempa, Interoperability Working Group Chair at OIF, leads demonstrations of 112G and 224G connectivity solutions while examining requirements for future 448G implementations at OFC 2025. The comprehensive testing includes multi-vendor interoperability across linear and RTLR systems, with analysis of signal characteristics and performance metrics for various connectivity configurations.
Stress Testing 1.6T for AI/ML
- Spirent's Asim Rasheed demos 1.6T transmission and the upcoming Ultra Ethernet Transport standard.
- Highlights Spirent's role as an independent test vendor.
- Enables objective technology comparisons across the industry.
Full summary
Asim Rasheed, Product Manager for High Speed Ethernet at Spirent, demonstrates advanced Ethernet technologies including 1.6T transmission and the upcoming Ultra Ethernet Transport standard. The presentation showcases Spirent's latest solutions for high-performance testing while highlighting their role as an independent test vendor enabling objective technology comparisons across the industry.
448G Will Be The First AI-Native Network Standard
- 448G is the first speed standard designed specifically for the AI era and hyperscaler next-gen chips.
- AI has expanded the data center networking market from $10-15B to a $50-100B opportunity.
- Associated server markets could reach $300-500 billion.
Full summary
Alan Weckel, Founder and Technology Analyst at 650 Group, explains how the new 448 speed standard at OAF marks the first speed designed specifically for the AI era, addressing hyperscalers' next-generation chip requirements. His analysis shows how this advancement has expanded networking from a $10-15 billion market to a $50-100 billion opportunity in data centers, with server markets potentially reaching $300-500 billion.
448G Is Inevitable—But Far from Easy
- 448G PAM4 continues the networking speed progression from 50G to 400G and beyond.
- AI applications are accelerating demand for advanced data center technologies.
- Both coherent optics and PAM4 solutions are in play for next-gen interconnects.
Full summary
Scott Wilkinson, Lead Analyst, Networking Components at Cignal AI, examines the development of 448G PAM4 technology and its implications for networking speed transitions from 50G to 400G and beyond. His analysis combines technical expertise with market insights to evaluate how AI applications are accelerating the need for advanced data center technologies, including coherent light and PAM4 solutions.
Google's AI Challenge: Scaling Networks for 100K+ TPU Clusters
- Google projects it will need 448G signaling by 2028 for future TPU generations.
- The 7th-gen TPU Ironwood and successors drive the bandwidth roadmap.
- LLM training clusters exceeding 100,000 TPUs are pushing toward PAM4, PAM6 and PAM8 modulation.
Full summary
Tad Hofmeister, Optical Hardware Engineer for Machine Learning Systems at Google, outlines the bandwidth requirements and technical hurdles for the seventh-generation TPU Ironwood and future TPU generations, highlighting the need for 448G signaling by 2028. He examines the challenges of achieving higher data rates through PAM4, PAM6 and PAM8 modulations to support large language model training that requires over 100,000 TPUs in a single cluster.
448 Signaling Sets New Path for Ethernet's Future
- The IEEE P802.3dj Task Force is doing early planning on 448 signaling for future Ethernet standards.
- D'Ambrosia has chaired three previous Ethernet speed transitions.
- 448 work is starting while the current 200-to-400 gigabit standard is still being finalized.
Full summary
John D'Ambrosia, Chair of the IEEE P802.3dj Task Force at IEEE, leads industry collaboration on 448 signaling development at the OAF workshop, representing early planning efforts for future Ethernet standards. His experience guiding three previous Ethernet speed transitions positions him to address the technical challenges of advancing from 200 to 400 gigabit implementations while the current standard is still being finalized.
Rethinking Connectors with 2D Interfaces for the 448G Era
- Marvell is researching high-density 2D connector designs to support 448G signaling.
- The goal is lossless electrical channels for next-generation AI data centers.
- Design priorities center on signal integrity optimization and thermal efficiency.
Full summary
Matt Traverso, Distinguished Engineer at Marvell, leads research on high-density 2D connector designs to enhance 448G signaling capabilities for AI and large-scale computing. His work prioritizes signal integrity optimization and thermal efficiency to create lossless electrical channels for next-generation data centers.
ling Bandwidth from 228 to 448G
- Meta is targeting a scale from 228 to 448 Gbits per second for AI networks.
- Power and space constraints are key limits on next-gen bandwidth.
- Modulation formats and interconnect components need industry alignment to advance.
Full summary
Xu Wang, Hardware Engineer at Meta, examines the growing network bandwidth demands for AI systems, focusing on scaling from 228 to 448 Gbits per second while managing power and space constraints. He outlines technical hurdles in modulation formats and interconnect components, emphasizing the need for industry alignment to support next-generation AI infrastructure development.
448G Connectivity: Key Requirements & Solutions
- Microsoft frames the shift from 224G to 448G around power budget and retimer challenges.
- Requirements include roughly 40 dB reach specifications alongside power and cost targets.
- Both copper and optical implementation approaches are on the table.
Full summary
Ashwin Gumaste, Principal AI Systems Architect at Microsoft, examines the industry's shift from 224G to 448G connectivity, focusing on power budget constraints and retimer solution challenges. He explores technical requirements including 40 dB reach specifications, power optimization targets, and cost considerations while discussing both copper and optical implementation approaches.
Gearbox-Free Electrical-Optical Co-Design for 448G Era
- Nubis argues for tight electrical-to-optical integration in the 448G era.
- Its approach eliminates power-consuming gearboxes.
- Optics can be driven directly from SerDes.
Full summary
Peter Winzer, Founder & CTO of Nubis Communications, emphasizes the importance of electrical-to-optical integration at the OIF 448G workshop, focusing on solutions that operate without power-consuming gearboxes and can be directly driven from SerDes.
Breaking 100GHz Barriers for Higher Bandwidth
- OIF's 448G workshop targets next-gen deployment challenges through industry collaboration.
- Higher data rates are pursued via both module density and signal speed improvements.
- Work spans connector interfaces, system architectures, and new bandwidth standards.
Full summary
Sam Kocsis, Director of Standards and Technology at OIF, leads discussions at the organization's 448G AI workshop to address next-generation deployment challenges through industry collaboration. The workshop examines connector interfaces and system architectures, focusing on increasing data rates through module density and signal speed improvements, while developing new standards for enhanced bandwidth capabilities.
448G Building the Future of High-Speed Connectivity Standards
- OIF's Interoperability Working Group is building a framework document for 448G standards.
- It covers 448G applications from long-reach down to chip-to-chip connections.
- Input is drawn from SerDes vendors, connector manufacturers, and end users across multiple standards bodies.
Full summary
Michael Klempa, Interoperability Working Group Chair at OIF, leads discussions at the 448G AI workshop to establish industry standards for high-speed connectivity. The initiative brings together multiple standards organizations and industry partners to develop a comprehensive framework document covering various 448G applications, from long-reach to chip-to-chip connections, while incorporating feedback from SerDes vendors, connector manufacturers, and end users.
Coherent Optics: From 400G to 1.6T
- Coherent optics standards are progressing from 400G ZR to 800G ZR and upcoming 1.6T interfaces.
- AI/ML workloads are making coherent optics essential for both inter- and intra-data center links.
- Optical switching requirements further strengthen the case for coherent.
Full summary
Tom Issenhuth, Co-Chair of the MA&E Committee at OIF, outlines the progression of coherent optics standards from 400G ZR to 800G ZR and upcoming 1.6T interfaces. The increasing demands of AI/ML workloads and higher data rates are making coherent optics essential for both inter- and intra-data center connectivity, especially when accounting for optical switching requirements.
Why 400G Systems Demand a New Era of Industry-Wide Collaboration
- The shift to 400G systems demands unprecedented collaboration across equipment vendors, component makers, and chip designers.
- OIF is coordinating industry-wide evaluation of circuit designs and modulation approaches.
- Power and reliability hurdles remain the main obstacles to 400G deployment.
Full summary
Cathy Liu, SerDes Architect at OIF, explains how the shift to 400G systems demands unprecedented collaboration between equipment vendors, component makers, and chip designers to overcome bandwidth constraints. She describes how OIF is coordinating industry-wide efforts to evaluate circuit designs and modulation approaches that will enable successful 400G deployment despite power and reliability hurdles.
Energy-Efficient Solutions for AI Infrastructure
- OIF's Energy Efficient Interfaces group builds on its 2020 co-packaging work with hyperscalers.
- It advances transmit linear receiver solutions and high-density connectors.
- Power consumption and reliability are the critical challenges in AI training environments.
Full summary
Jeff Hutchins, Director of the CTO Office at OIF, leads the organization's Energy Efficient Interfaces group, which builds upon their 2020 co-packaging work with hyperscalers. The group advances multiple energy-saving technologies, including transmit linear receiver solutions and high-density connectors, while addressing critical power consumption and reliability challenges in AI training environments.
Shaping Next-Gen AI Infrastructure & Network Solutions
- OIF's workshop convenes industry experts, hyperscalers, and analysts around AI compute infrastructure.
- It addresses six key areas including electrical data rates, coherent technology, and network capacity management.
- Tracy highlights recent progress in 448 gigabit solutions.
Full summary
Nathan Tracy, President of OIF, leads a comprehensive 400 gigabit workshop focused on AI compute infrastructure requirements, bringing together industry experts, hyperscalers, and analysts. The workshop addresses six key areas including electrical data rates, coherent technology, and network capacity management, with Tracy emphasizing recent progress in 448 gigabit solutions while tackling technical hurdles in achieving higher data rates.
TE Connectivity Advances Channel Modeling for 448G
- TE Connectivity is advancing channel models and analysis for 448G next-gen interconnects.
- It is working with ecosystem partners to evaluate channel performance.
- The challenge is meeting rising data rates within existing equipment size limitations.
Full summary
Nathan Tracy, System Architect at TE Connectivity, led discussions at OIF's 448G networking workshop for AI applications, focusing on channel models and analysis for next-generation interconnects. TE Connectivity is working with ecosystem partners to evaluate channel performance within equipment size limitations while meeting increasing data transmission requirements.
GPU Networking Evolution: From 200G to 400G
- UALink Consortium has released a new specification amid the fast move from 200G to 400G.
- The standard enables multi-vendor GPU and switch compatibility.
- High-bandwidth, low-latency links let multiple GPUs operate as a unified system.
Full summary
Kurtis Bowman, Chairman of UALink Consortium, outlines the industry's quick advancement from 200G to 400G networking while highlighting the organization's new specification release. The standardization enables multi-vendor GPU and switch compatibility through high-bandwidth, low-latency connections that allow multiple GPUs to operate as a unified system.
AI Infrastructure Predictions That Will Shape 2025
- AI shifts from training to inference in 2025.
- AI models increasingly deploy at network edges.
- Arrcus embeds security into routing and switching while optimizing channels for power and cost efficiency.
Full summary
Shekar Ayyar, CEO of Arrcus, outlines three major AI infrastructure predictions for 2025, including a shift from training to inference and the deployment of AI models at network edges. His analysis highlights how Arrcus is developing intelligent networks with optimized channels for power and cost efficiency, while integrating security features directly into routing and switching infrastructure.
Bright Prospects for Ethernet in AI Infrastructure
- Ethernet dominates GPU clusters and scale-up networking.
- Efficient processors and networking make AI deployment more accessible.
- Organizations of all sizes gain the ability to implement large language models.
Full summary
Pete Del Vecchio, Data Center Switch Product Line Manager at Broadcom, shares his vision for 2025, highlighting Ethernet's dominance in GPU clusters and scale-up networking. His analysis points to a future where AI deployment becomes more accessible through efficient processors and networking solutions, enabling organizations of all sizes to implement large language models.
Ethernet, Standards & Copper's Comeback
- InfiniBand's role shrinks as Ethernet becomes the standard for scale-out networks.
- Proprietary interfaces align with 224G IEEE Ethernet standards.
- Liquid cooling revives copper for NIC-to-switch links, offering better power efficiency and reliability at lower cost.
Full summary
Don Barnetson, VP Product - HiWire AECs at Credo, forecasts key networking shifts for 2025, including InfiniBand's reduced role as Ethernet becomes the standard for scale-out networks and proprietary interfaces align with 224G IEEE Ethernet standards. His analysis highlights how liquid cooling's space advantages will drive renewed interest in copper connectivity for NIC-to-switch connections, offering better power efficiency and reliability at lower costs.
Data Center Tech Evolution in 2025
- Hybrid cloud adoption increases in 2025.
- AI deployments shift from InfiniBand to Ethernet.
- Energy optimization comes via efficient optical modules, liquid cooling, and AI-driven facilities management.
Full summary
Amit Sanyal, Senior Director of Data Center Product Marketing at Juniper, outlines key networking trends for 2025, including increased hybrid cloud adoption and a shift from InfiniBand to Ethernet for AI deployments. He emphasizes the importance of energy optimization through efficient optical modules, liquid cooling solutions, and AI-driven facilities management systems in data centers.
Accelerating the Ethernet Market
- AI infrastructure drives a $200 billion Ethernet switching market opportunity.
- Bandwidth growth is accelerating from 30-40% to 70-80% CAGR.
- Data centers projected to exceed one billion ports with 4x revenue growth across LPO, LRO, ACC, AEC, and CPO.
Full summary
Alan Weckel, Founder and Technology Analyst at 650 Group, presents from TEF 2025 on the $200 billion Ethernet switching market opportunity driven by AI infrastructure requiring multiple interconnected networks for GPUs and XPUs, with bandwidth growth accelerating from 30-40% to 70-80% CAGR. He projects data centers will exceed one billion ports within years, creating multi-billion dollar opportunities across LPO, LRO, ACC, AEC, and CPO transceiver technologies as the industry experiences 4x revenue increases alongside dramatic expansion in speeds and capacity.
Power, Cooling & the Future of Data Center Design
- Power, cooling, and space constraints increasingly drive data center network design.
- 1.6 terabit deployments with hybrid cooling anticipated by 2026.
- Scale-across connectivity between distributed data centers is a key trend for the coming year.
Full summary
Arihant Jain, Manager, Systems Engineering at Arista Networks, examines scale-out networking architectures and interconnect technologies while highlighting how power, cooling, and space constraints are increasingly driving data center network design decisions alongside traditional requirements. He emphasizes Ethernet's expanding role across scale-up and scale-out applications, anticipates 1.6 terabit deployments with hybrid cooling solutions by 2026, and identifies scale-across connectivity between geographically distributed data centers as a key deployment trend for the coming year.
Data Center Optimization Strategies
- Modern AI data centers are reaching 5 gigawatts and multi-million GPU clusters.
- Optimization spans every layer, from nanometer-scale silicon transistors to full racks.
- TEF convenes the entire AI infrastructure ecosystem to enable next-generation buildouts.
Full summary
Pete Del Vecchio, Data Center Switch Product Line Manager at Broadcom, presents at the Technology Exploration Forum for Internet Alliance 2025, highlighting how the event uniquely brings together the entire AI infrastructure ecosystem to address the massive scale of modern data centers, with some reaching 5 gigawatts and multi-million GPU clusters. He emphasizes the comprehensive optimization approach being applied across all components, from nanometer-scale silicon transistors to racks, as the industry collaborates to enable next-generation AI infrastructure capable of supporting systems the size of Manhattan.
Ethernet for AI: Defining Infrastructure Standards
- Ethernet Alliance conference convenes decision makers from Meta, Microsoft, and Hewlett Packard.
- Focus is on the convergence of power and data for AI and quantum computing infrastructure.
- Interconnect industry learns required innovations, codes, and standards directly from technology providers.
Full summary
David B. Kiddoo, Executive Director / CEO of CCCA and IWCS, explains his organizations' sponsorship of the Ethernet Alliance conference that brings together decision makers from Meta, Microsoft, and Hewlett Packard to address Ethernet for AI and the convergence of power and data supporting AI and quantum computing infrastructure. The conference enables the interconnect and network infrastructure industry to learn directly from technology providers about required innovations, necessary codes and standards, and how to allocate limited resources effectively to advance AI and quantum computing technologies.
AI Infrastructure Demands Are Accelerating 400G Per Lane Ethernet
- Hyperscaler XPU and GPU deployments require proportional networking growth across scale-up, scale-out, and scale-across.
- Industry is focused on achieving 400 gigabit speeds through ecosystem collaboration.
- Open standards like Ethernet are central to advancing AI infrastructure.
Full summary
Ravi Shah, Director, Corporate Strategy at Cisco Systems, discusses how AI use cases are driving unprecedented infrastructure adoption as hyperscalers deploy XPUs and GPUs requiring proportional networking growth across scale-up, scale-out, and scale-across architectures. He explains the industry's focus on achieving 400 gigabit speeds through ecosystem collaboration addressing engineering challenges, while emphasizing the forum's role in enabling thought leaders to leverage open standards like Ethernet and advance AI infrastructure development.
Building Next-Gen Ethernet: Collaboration, Innovation & Industry Challenges
- Next-gen optical modules need doubled improvements in signal integrity, power, cost, and manufacturing yield.
- UltraEthernet standards work demands cross-industry collaboration.
- The timeline for these next-generation innovations remains uncertain.
Full summary
Adee Ran, Architect of Ethernet at Cisco Systems, discusses his standards work with IT and UltraEthernet, emphasizing the collaborative effort required across industries to develop next-generation networking technologies. He explains that achieving mature next-generation technology requires inventing numerous components, such as optical modules with doubled improvements in signal integrity, power consumption, cost, and manufacturing yield, though the timeline for these innovations remains uncertain.
Faster and Faster Networking Upgrade Cycles
- 112 gigabit SerDes (400G ports) giving way to 224 gigabit SerDes enabling 800G for GPU platforms.
- Accelerator platforms now advance every 18 months or less.
- Ecosystem moving toward 440 gigabit SerDes networks for next-generation AI.
Full summary
Baron Fung, Senior Research Director at Dell'Oro Group, examines data center infrastructure markets with emphasis on servers and connectivity, tracking the evolution from current 112 gigabit SerDes supporting 400 gigabit ports to emerging 224 gigabit SerDes enabling 800 gigabit speeds for GPU platforms. He highlights that accelerator platforms now advance every 18 months or less, driving the ecosystem toward Ethernet-based solutions and the upcoming transition to 440 gigabit SerDes networks to support next-generation AI and accelerated computing requirements.
Why Betting Against Ethernet Fails: Meeting AI's Speed Demands
- 400 gigabit Ethernet is emerging before 200 gigabit is fully deployed.
- Industry should focus on 'just enough' rather than attempting everything at once.
- Ethernet has consistently succeeded over competing technologies.
Full summary
Peter Jones, Chair of the Ethernet Alliance and Cisco Systems employee, addresses the accelerating pace of Ethernet speed transitions at the second TF Ethernet for AI event, noting that 400 gigabit Ethernet is emerging before 200 gigabit is fully deployed. He emphasizes that the industry must focus on "just enough" rather than attempting everything at once, while highlighting Ethernet's consistent success over competing technologies.
Industry-Wide Collaboration as Ethernet Speeds Ahead
- The next 400 Gbit/s Ethernet upgrade requires collaboration from all ecosystem stakeholders.
- Ethernet Alliance facilitates industry-wide collaboration through TEF events and partnerships.
- Faster consensus among manufacturers, material providers, and component makers accelerates solutions.
Full summary
John D'Ambrosia, Chair, IEEE P802.3dj Task Force at the Ethernet Alliance, discusses the evolution of 400 Gbit per second signaling for Ethernet in AI applications, noting that the next upgrade requires collaboration from all stakeholders across the industry ecosystem. He explains that the Ethernet Alliance facilitates this industry-wide collaboration through TEF events and partnerships, recognizing that faster consensus-building among manufacturers, material providers, and component makers accelerates progress toward solutions.
Network Interoperability: Hyperscalers and Ethernet Standards
- Hyperscalers like AWS optimize AI-focused ecosystems while Ethernet standards prioritize cross-vendor interoperability.
- Ethernet Alliance roadmap, first released in 2015, is now updated quarterly online.
- The roadmap tracks industry technologies, speeds, and interface nomenclatures.
Full summary
David Rodgers, Sr. Business development manager- EXFO and President & Events chair - Ethernet Alliance, examines the bifurcated state of network interoperability, contrasting hyperscalers like AWS optimizing AI-focused ecosystems with Ethernet standards prioritizing cross-vendor interoperability for broader markets. He highlights the Ethernet Alliance's evolving roadmap, first released in 2015 and now updated quarterly online, which tracks industry technologies, speeds, and interface nomenclatures to guide the industry through rapid changes in Ethernet technology.
Advanced Modulation, FEC & Next-Gen Fiber Tech
- IEEE P802.3dj is advancing 400 Gbit/s Ethernet signaling, modulation, and forward error correction.
- Fiber options under study include hollow core and multi-core fiber.
- An IEEE workshop on fiber technologies is announced for late February.
Full summary
John D'Ambrosia, Chair, IEEE P802.3dj Task Force at IEEE, discusses the working group's progress on 400 Gbit per second Ethernet signaling, covering modulation schemes, forward error correction, channel characteristics, and various fiber options including hollow core and multi-core fiber. He announces an IEEE workshop in late February to examine fiber technologies, emphasizing the goal of effective information transmission across different channel types as high-speed Ethernet development continues to advance.
448 Gig Ethernet: Deployment Challenges & Engineering Choices
- 448 gig deployment barriers are driven by hyperscaler growth and data center demand.
- Key challenges include higher-order modulation formats, next-generation connectors, and channel considerations.
- Engineering choices for the next speed grade remain under discussion.
Full summary
John Calvin, Senior Strategic Planner at Keysight Technologies, presents from the 2025 Ethernet Alliance hosted TEF event in Mountain View, discussing 448 gig technology and deployment barriers driven by hyperscaler growth and data center demands. He addresses challenges including higher order modulation formats, next generation connectors, and channel considerations, noting that engineering choices for the next speed grade remain under discussion at the event.
AI Networking at 1.6T & 3.2T: Performance vs Power Trade-offs
- 1.6T Ethernet deployment has begun while hyperscalers plan for 3.2T Ethernet.
- 400G signaling is advancing from lab to product amid performance, power, reliability, and supply chain trade-offs.
- Keysight works with customers on testing requirements at these speeds.
Full summary
Hadrien Louchet, Product Planner at Keysight Technologies, presents at a technology exploration forum on AI networking's future, noting that while 1.6T Ethernet deployment has begun and hyperscalers are planning for 3.2T Ethernet, the industry must navigate critical trade-offs between performance, power efficiency, reliability, and supply chain security as 400G signaling advances from lab to product. Louchet explains that Keysight, as a test and measurement expert, works with customers to understand testing requirements at these speeds and prepares products to support large-scale deployment of next-generation AI networking infrastructure.
Overcoming 400G Technology Challenges
- 400G advancement requires collaboration across connector, cable, silicon, optical, and system vendors.
- Challenging 400G conditions demand precise link training protocols and standardized error correction.
- Robust Ethernet standards are needed to ensure 400G interoperability.
Full summary
Jim Hsieh, Senior Technical Manager at Mediatek, presents on the critical need for industry-wide collaboration to advance 400G technology, emphasizing cooperation among connector manufacturers, cable vendors, silicon vendors, optical vendors, and system providers. He explains that 400G's challenging initial conditions require precise link training protocols, standardized error correction methodologies, and robust Ethernet standards to ensure interoperability and reliable connections across all industry participants.
Ethernet Scaling Challenges & Solutions
- Ethernet remains the preferred choice for scale-up and scale-out AI applications.
- Unprecedented demand spans network connectivity, semiconductors, and testing.
- TEF is a critical venue for identifying technologies that must integrate to solve interoperability challenges.
Full summary
Hani Daou, Business Development Manager at Multilane, presents on behalf of the Ethernet Alliance at the Technology Exploration Forum, discussing how Ethernet remains the preferred choice for scale-up and scale-out AI applications amid unprecedented demand for network connectivity, semiconductors, and testing. He emphasizes the forum's role as a critical venue for identifying technologies that must integrate seamlessly to address interoperability challenges and meet the demanding requirements of cloud service providers and end users with their growing appetite for AI-based applications.
OIF's 448G Framework: AI Fabric Challenges & Standards
- OIF publishes its 448 gigabit framework, the first deliverable from a project launched in August 2024.
- The framework creates a common industry language for 448 Gbps data rate challenges.
- It covers modulation selection, reach, AI fabric reliability, and electrical/optical integration.
Full summary
Nathan Tracy, President of the OIF and representative from TE Connectivity, announces the publication of OIF's 448 gigabit framework document at the Ethernet Alliance's TEF 2025, marking the first deliverable from the project launched in August 2024. The framework establishes a common industry language for addressing challenges at 448 gigabit per second data rates, including modulation selection, reach requirements, AI fabric reliability demands, and the integration of electrical and optical modulation approaches.
CEI-448G Framework: New Standards AI Interconnect Networks
- OIF publishes the CEI-448G framework for next-generation AI interconnect networks.
- It identifies future CEI-448G XSR, VSR, and LR variants as follow-on projects.
- Calls for collaboration across IEEE 802.3 Ethernet, OCP, SNIA, and UEC.
Full summary
Cathy Liu, SerDes Architect at OIF, presents the newly published CEI-448G framework, a collaborative achievement by OIF members across AI hyperscalers, system vendors, and component manufacturers that establishes common guidelines for next-generation AI interconnect networks. The framework serves as foundational work identifying future projects including CEI-448G XSR, VSR, and LR variants, while emphasizing the need for collaboration among standards bodies such as IEEE 802.3 Ethernet, OCP, SNIA, and UEC.
Ethernet at 400G per Lane: Standards & Challenges
- Champions an IEEE 802.3 project for 400 gig electrical and optical signaling optimized for radix scale-out.
- Requires collaboration among hyperscalers, OIF, UEC, SNIA, and component suppliers.
- Three 448/400 gig challenges: faster time to market, optimal use of materials, and AI's real-world impact.
Full summary
Kent Lusted, Distinguished Architect at Synopsys, discusses next generation Ethernet for AI at TEF, highlighting his role as champion for an IEEE 802.3 project focused on 400 gig electrical and optical signaling for AI networks optimized for radix scale-out use cases. He emphasizes the need for collaboration among hyperscalers, standards organizations like OIF, UEC, and SNIA, and component suppliers to address three major challenges for 448 and 400 gig Ethernet: faster time to market, optimal use of available information and materials, and recognition of AI's impact on daily life through search, recommendations, pattern recognition, and infrastructure.
The Race to 400 Gig Ethernet
- AI bandwidth demands push Ethernet to 400 Gig with shrinking channel budgets.
- Higher sensitivity to impairments is addressed via advanced modulation and improved connector designs.
- TE Connectivity relies on industry collaboration to solve 400G engineering challenges.
Full summary
Ashika Pandankeril Shaji, Manager System Architect at TE Connectivity, discusses the rapid evolution of Ethernet technology to 400 Gig speeds driven by AI network bandwidth demands and the significant technical challenges this transition presents, including shrinking channel budgets and increased sensitivity to impairments. She highlights how TE Connectivity addresses these challenges through advanced modulation techniques and improved connector designs, while emphasizing the value of industry collaboration at the TIP conference to solve complex engineering problems shaping the future of Ethernet and AI networks.
Reinventing Commercial Networking: Meter's Integrated Infrastructure Solution
- Meter's Anil Varanasi describes a vertically integrated approach to modernizing enterprise networking.
- Meter offers a full-stack hardware and software solution combining routing, switching, and wireless.
Full summary
Anil Varanasi, Co-founder and CEO of Meter, discusses the company’s vertically integrated approach to modernizing enterprise networking infrastructure. Meter aims to simplify network deployment by offering a full-stack hardware and software solution that combines routing, switching, and wireless capabilities.
Tracking NaaS as a Consumption Model
- Vertical Systems Group examines the complexity of monitoring NaaS and reports significant SD-WAN services growth.
- Broadband and Ethernet Dedicated Internet Access are key drivers of network services expansion.
Full summary
Erin Dunne, Director of Research Services at Vertical Systems Group, examines the complexities of monitoring NaaS and reports significant growth in related SD-WAN services. Her analysis highlights the importance of broadband and Ethernet Dedicated Internet Access as key factors driving the expansion of network services.
The Future of Co-Packaged Optics Looks Bright
- Marvell's Nick Kucharewski positions co-packaged optics for scale-up networking with lower power and latency.
- Marvell's light engine provides up to 51 Tbps of bandwidth for scale-up switches.
Full summary
Nick Kucharewski, SVP and GM of Network Switching, highlights co-packaged optics (CPO) as a promising technology for scale-up networking, integrating fiber links directly into product packages for higher speed signals with lower power consumption and latency. Marvell's light engine provides to 51 Tbps bandwidth for scale-up switches and supports reliable large-scale deployment.
Scaling Networks Beyond the Rack: Choosing the Right Interconnect Technology
- Marvell's Nick Kucharewski examines expanding networks from rack to row level.
- Key deployment factors include capacity, fault tolerance, and field replaceability.
Full summary
Nick Kucharewski, SVP and GM of Network Switching, explores the challenges of expanding networks from rack to row level and the critical role of interconnect technology selection. He outlines key factors for network deployment, including capacity, fault tolerance, and field replaceability.
Never Bet Against Ethernet
- ACG Research's Ray Mota flags rising interest in quantum computing for security and segment routing.
- AI infrastructure and the Ethernet versus InfiniBand debate are drawing growing attention.
Full summary
Ray Mota, CEO and Principal Analyst at ACG Research, discussed the growing interest in quantum computing for security, increasing focus on segment routing, AI infrastructure, and Ethernet vs Infiniband.
Network Evolution with 400G Ethernet & Network Slicing
- Spirent's Steve Douglas discusses the evolution toward 400G Ethernet and network slicing for deterministic services.
- Data center fabrics are evolving to support AI processing, including use of RoCEv2.
Full summary
Steve Douglas, Head of Market Strategy at Spirent, speaks about the evolution towards 400G Ethernet, network slicing for deterministic services, and evolution of data center fabrics to support AI processing, including use of ROCEv2.
Next Gen Networks for AI
- Arrcus is building a platform that connects AI components across the network for operations.
- New support includes SmartNIC and DPU platforms like Nvidia BlueField, plus the AR IQ observability platform.
Full summary
Sanjay Kumar, VP of Products and Marketing at Arrcus, highlighted the company's strategy to leverage AI in network operations, including the introduction of a platform that connects various AI components across the network. Arrcus has also announced support for new platforms with enhanced ethernet connectivity, SmartNIC and DPU platforms like Nvidia Bluefield, and AR IQ, an observability platform providing a comprehensive network view for operators.
Impactful AI Connectivity
- Astera Labs' Ahmad Danesh names three data center challenges: AI workloads, real-time reasoning, and rapid deployment of new accelerators.
- Its Ethernet retimers, Smart Fabric switch, and PCIe 6 retimers pair with the Cosmos software stack to improve connectivity and observability.
Full summary
Ahmad Danesh, Sr. Director of Product Management at Astera Labs, outlines three key challenges in data center infrastructure: AI workloads, real-time reasoning, and rapid deployment of new accelerators. Astera Labs addresses these issues through their Ethernet retimers, Smart Fabric switch, and PCIe 6 retimers, combined with their Cosmos software stack to enhance connectivity and observability.
The Ethernet Path to Scale-Up Networks
- Broadcom's Ram Velaga notes Ethernet dominates backend networks for GPU clusters of 10,000 to over 100,000 GPUs.
- He anticipates Ethernet winning scale-up networking within racks on both technical and economic merits.
Full summary
Ram Velaga, GM and SVP of the Core Switching Group at Broadcom, highlights Ethernet's dominance in large GPU clusters for scale-out networking, noting its prevalence in backend networks for clusters of 10,000 to over 100,000 GPUs. Ram anticipates Ethernet's success in scale-up networking within racks, citing its ability to meet requirements and economic advantages.
All-Optical Circuit Switching for Scalable Data Centers
- Coherent's Liquid Crystal-based optical switch offers low drive voltage, no moving parts, and data-rate agnosticism.
- The platform cuts power 30-40%, with sampling in coming months and commercialization planned next year.
- Coherent cited strong interest from data center operators, hyperscalers, and OEMs for the switch.
Full summary
Sanjai Parthasarathi, CMO of Coherent, presented the company's Liquid Crystal-based Optical switch platform, emphasizing its low drive voltage, absence of moving parts, data rate agnosticism, and power reduction capabilities of 30-40%, making it an attractive solution for power-hungry data centers. Parthasarathi also revealed plans to sample the platform in the coming months, commercialize it next year, and noted strong interest from data center operators, hyperscalers, and OEMs.
800G Linear Receive Optics Connected to 51.2T Switch
- Credo demoed its Dove 850 Linear Receive Optics (LRO), where only the transmit path uses a DSP.
- Placing a DSP solely on the transmit path for retiming and equalization unlocks dramatic power savings.
Full summary
Easwar Sankar, Optical FAE Director at Credo, demos the company's new Linear Receive Optics (LRO) product, Dove 850. In an LRO transceiver, or Active Optical Cable (AOC), only the transmit path from the electrical input to the optical line side output includes a DSP for signal retiming and equalization. This unlocks dramatic power savings.
Coherent Pluggables for 25GE to 400GE Metro Aggregation
- DZS unveiled the Saber 2200, a high-capacity layer 2 Ethernet aggregation switch built with Infera.
- The Saber 2200 supports XR Optics with 24 ports of varying capacity, part of DZS's XRS and Edge RMS work.
Full summary
Jason Lauzon, Strategic Sales Director at DZS, unveiled the Saber 2200, a high-capacity layer 2 Ethernet aggregation switch developed in collaboration with Infera. The new product supports XR Optics, features 24 ports of varying gig capacity, and is part of DZS's exploration of XRS with their Edge RMS, a hardened RM system that utilizes XR as a spectrum as a service.
800 Gigabit Ethernet Interoperability Demo
- The Ethernet Alliance booth demonstrated interoperability at speeds up to 800 Gigabit Ethernet.
- The demo spanned switches, routers, and interconnects across OSFP, QSFP-DD, QSFP, and SFP form factors.
Full summary
The Ethernet Alliance's OFC 2024 booth showcases Ethernet’s versatility and reliability at speeds of up to 800 Gigabit Ethernet (GbE). The booth demonstrated a wide range of switches, routers, interconnects and interfaces including OSFP, QSFP-DD, QSFP, and SFP pluggable form factors. Dave Estes shares an overview.
Demonstrating Marvell's Latest Portfolio Solutions for AI
- Marvell showcased a 51 Tbps switch, 800G AEC cables, and 800G optics for low-latency AI switching.
- The demo included a liquid-cooled 51 Tbps platform and another using 64 800G OSFP modules.
Full summary
George Hervey, Principal Architect at Marvell, showcased their AI infrastructure solutions: a 51 Tbps switch, 800G AEC cables, and 800G optics for low-latency switching. The demonstration also highlighted a 51 Tbps platform with a liquid cooling option and another 51 Tbps platform using 64 800G OSFP modules.
Demonstrating Marvell's Latest Portfolio Solutions for AI
- Marvell showcased a 51 Tbps switch, 800G AEC cables, and 800G optics for low-latency AI switching.
- The demo included a liquid-cooled 51 Tbps platform and another using 64 800G OSFP modules.
Full summary
George Hervey, Principal Architect at Marvell, showcased their AI infrastructure solutions: a 51 Tbps switch, 800G AEC cables, and 800G optics for low-latency switching. The demonstration also highlighted a 51 Tbps platform with a liquid cooling option and another 51 Tbps platform using 64 800G OSFP modules.
New Optical Capabilities for the Metro Edge
- Nokia launched the 1830 PSS, a Metro-optimized optical platform for diverse Metro Edge requirements.
- The 1830 PSS targets 5G applications, higher-capacity enterprise access, and data center connectivity.
Full summary
Hector Menendez, Product Marketing Manager at Nokia, has announced the launch of the 1830 PSS (photonic service switch), a Metro-optimized optical platform designed to support diverse requirements at the Metro Edge, particularly beneficial for 5G applications, enterprises needing higher capacity access, and data centers requiring high capacity connectivity.
Spirent's 800G 802.3df Appliance & LPO optics support
- Spirent unveiled the B3 800G Appliance supporting IEEE 802.3df and linear pluggable optics (LPO).
- The B3 showcased 800G interconnects, including DACs, AOCs, ACC, and ZR transceivers, from 11 partners.
Full summary
Steve Rumsby, Senior Director of Architecture & Platform at Spirent, unveiled the Spirent B3 800G Appliance supporting IEEE 802.3df. The B3 Appliance also supports linear pluggable optics (LPO), an interconnect that uses lower power and latency. He also showcased 800G interconnects (DACs, AOCs, ACC and ZR transceivers) from 11 different partners.
Validating 800G links and AI/ML Workloads
- Spirent showcased products from 100G to 800G, including an A1 Appliance emulating hyperscaler network flows.
- Spirent demonstrated sending 1.6 Tbps of live traffic through two 800G ports on a Juniper switch.
Full summary
Asim Rasheed, Product Manager, High Speed Ethernet, at Spirent, showcased a range of products from 100G to 800G, including the A1 Appliance with 16 ports of 400G, capable of emulating hyperscaler network flows. Rasheed also demonstrated sending 1.6 Tbps of live traffic through 2 800G ports on a Juniper switch.
AI Reshaping the Ethernet Landscape
- 650 Group's Alan Weckel says AI is spawning three new networks with faster speeds and a vendor product race.
- High-speed Ethernet port adoption is growing rapidly, with Nvidia entering Ethernet semiconductors and systems.
Full summary
Alan Weckel, Founder and Technology Analyst at 650 Group, examines AI's influence on the Ethernet market, highlighting the creation of three networks with faster speeds and a vendor race for new products. He discusses the rapid growth in high-speed Ethernet port adoption and the expanding data center switching market, noting Nvidia's entry into Ethernet semiconductors and systems as a key development.
Optimizing Networking for AI Scale and Performance
- Astera Labs highlights the diverse requirements of different AI network types as networks expand.
- Astera's PCIe and Ethernet retimers provide intelligent connectivity to address channel challenges in AI architectures.
Full summary
Christopher Blackburn, System Architect & Director of Field Applications Engineering at Astera Labs, explores the challenges and future of AI networks, focusing on the diverse requirements of different network types and the need for optimization as networks expand. He discusses how Astera Labs' product line, including PCIe and Ethernet retimers, offers intelligent connectivity solutions to address channel challenges in various AI architectures.
AI-Driven Networks for the Future
- Cisco's Jiri Chaloupka argues AI networks should build on existing Ethernet standards while solving new challenges.
- He stresses standardization for AI needs like high bandwidth, low latency, and large-scale GPU interconnectivity.
Full summary
Jiri Chaloupka, Principal Engineer Technical Marketing at Cisco Systems, explores the progression of internet technology towards AI-driven networks, emphasizing the need to build upon existing Ethernet standards while addressing new AI-specific challenges. He underscores the importance of industry collaboration and standardization to support the unique requirements of AI networks, including high bandwidth, low latency, and large-scale GPU interconnectivity.
Analyzing the Market Impact of AI Networks
- Dell'Oro's Sameh Boujelbene notes AI workload networks drive accelerated refresh cycles and rising bandwidth demand.
- Ethernet is gaining on InfiniBand in AI clusters on technological advances and hyperscaler preferences.
Full summary
Sameh Boujelbene, Vice President at Dell'Oro Group, examines the rapid evolution of AI workload networks, highlighting accelerated refresh cycles and increasing bandwidth demands. She discusses the competition between InfiniBand and Ethernet in the AI cluster market, noting Ethernet's growing adoption due to technological advancements and hyperscaler preferences.
Ethernet's Future: Technical Collaboration for Scaling to the AI Challenge
- Ethernet Alliance chair Peter Jones emphasizes industry collaboration to meet AI networking requirements.
- Jones expects the event to drive greater industry alignment on overcoming AI networking challenges.
Full summary
Peter Jones, Chair of the Ethernet Alliance, discusses the Ethernet in the Age of AI Networking event, emphasizing industry collaboration to address challenges in meeting requirements. Jones expresses confidence in Ethernet's ability to overcome obstacles and anticipates that the event will lead to greater industry alignment in addressing AI networking challenges.
Ethernet Evolution: Powering High-Speed AI Connections
- John D'Ambrosia of the IEEE P802.3dj Task Force explains how Ethernet's growth supports the AI ecosystem.
- The two-day Technology Exploration Forum focuses on 200 Gbps signaling to enable Ethernet up to 1.6 Tbps.
Full summary
John D'Ambrosia, Chair of the IEEE P802.3dj Task Force and adviser to the Ethernet Alliance board of directors, discusses how Ethernet's growth supports the AI ecosystem. He introduces the two-day Technology Exploration Forum covering issues impacting faster data rates, focusing on 200 Gbps signaling to enable higher-speed Ethernet connections up to 1.6 Tbps.
Ethernet's AI Era: Navigating 400G Per Lane Challenges
- EXFO's David Rodgers explores the complexity of achieving 400 Gig per lane Ethernet speeds in the AI era.
- He points to a potential need for specialized testing and industry collaboration to clear these hurdles.
Full summary
David Rodgers, Technical Business Development Manager at EXFO, explores the complexities and opportunities of Ethernet in the AI era, focusing on the challenges of achieving 400 Gig per lane speeds. He discusses the potential need for specialized testing and emphasizes the importance of collaboration within the industry to overcome these hurdles.
Ethernet's Future in AI Networks
- Google's Moray McLaren notes Google uses an Ethernet-based proprietary network in its TPU systems.
- He sees Ethernet potentially dominating scale-out and host networks, but alternatives for latency-sensitive cases.
Full summary
Moray McLaren, Principle Engineer at Google, explores Ethernet's capabilities in addressing interconnect challenges for machine learning networks, highlighting Google's use of an Ethernet-based proprietary network in their TPU systems. He suggests Ethernet's potential dominance in scale-out and host networks for ML applications, while acknowledging the need for alternative solutions in more latency-sensitive architectures.
Adapting Ethernet for AI, ML, and HPC Networking Challenges
- IEEE's Kent Lusted describes ongoing 802.3dj work adapting Ethernet for AI, ML, and HPC applications.
- He stresses addressing both scale-out and scale-up networks while balancing component requirements.
Full summary
Kent Lusted, 802.3dj Electrical Track Chair at IEEE, explores the ongoing development of Ethernet to address market demands and customer applications in AI, ML, and HPC. He examines the challenges of balancing component requirements and adapting Ethernet capabilities for future needs, emphasizing the importance of addressing both scale-out and scale-up networks in AI.
Powering AI's Future with Advanced Ethernet
- Intel's Mike Li discusses 200 Gbps per lane Ethernet and development of 800 Gbps and 1.6 Tbps Ethernet.
- Work is underway toward 400 Gbps per lane, with future generations reaching up to 6.4 Tbps.
Full summary
Mike Li, a Fellow at Intel, discusses high-speed networking advancements for AI applications, including 200 Gbps per lane Ethernet specifications and development of 800 Gbps/1.6 Tbps Ethernet. Li emphasizes Ethernet's importance for GPU acceleration and clustering architectures, while also highlighting ongoing work towards 400 Gbps per lane speeds and future generations reaching up to 6.4 Tbps.
Ethernet Interoperability in the AI Age
- Intel's Sam Johnson emphasizes Ethernet interoperability as critical for networking and AI applications.
- As chair of the Ethernet Alliance's high-speed subcommittee, he promotes vendor collaboration and testing.
Full summary
Sam Johnson, Link Applications Engineering Manager at Intel, emphasizes the critical role of Ethernet interoperability in networking and AI applications. As chair of the Ethernet Alliance's highspeed networking subcommittee, Johnson highlights the organization's efforts to promote collaboration and testing among vendors to advance the Ethernet ecosystem.
Building Ultra Ethernet for AI Data Center Evolution
- Intel's Uri Elzur outlines the Ultra Ethernet Consortium's approach to AI data center networking.
- Optional features like link-level retry and credit-based flow control enhance Ethernet for AI applications.
Full summary
Uri Elzur, GPU Networks and System Architecture at Intel, outlines the Ultra Ethernet Consortium's approach to addressing AI-related challenges in data center architecture and networking. He describes the consortium's collaboration with the Ethernet Alliance to enhance Ethernet's capabilities for AI applications through optional features like link-level retry and credit-based flow control.
Achieving 400G Ethernet: Industry Challenges in the AI Era
- OIF president Nathan Tracy highlights the OIF's 448 Gbps project aimed at unifying the industry.
- Companies like TE Connectivity are already investigating 400G solutions amid ongoing 200 Gbps work.
Full summary
Nathan Tracy, President of OIF, outlines industry efforts to achieve 400 Gbps Ethernet for AI applications at the Ethernet Alliance's Technology Exploration Forum. Tracy highlights OIF's 448 Gbps project, which aims to address challenges and unify the industry, while noting that companies like TE Connectivity are already investigating solutions despite ongoing work on 200 Gbps specifications.
Ethernet Insights from OCP Global Summit
- OCP CTO Bijan Nowroozi stresses Ethernet's technical and economic advantages for AI's future.
- He notes the recent OCP Global Summit showed Ethernet is having its moment in the industry.
Full summary
Bijan Nowroozi, Chief Technology Officer of Open Compute Project (OCP), discusses the significance of Ethernet in AI's future, emphasizing its technical and economic advantages. The recent OCP Global Summit featured numerous speakers highlighting Ethernet's benefits, with Nowroozi noting that Ethernet is currently experiencing its moment in the industry.
Small Form Factors for High-Speed Ethernet
- Samtec's Thomas Palkert represents the SNEA small form factor group defining connectors and transceivers.
- He stresses balancing faster Ethernet speeds with higher density, lower power, and reduced cost for AI.
Full summary
Thomas Palkert, System Architect at Samtec, represents the small form factor group of SNEA, which defines connectors and transceivers for optical and storage industries. Palkert emphasizes the challenges of density, power, and cost for Ethernet in AI applications, highlighting the need for faster speeds while achieving higher density, lower power consumption, and reduced costs.
Unleashing CXL Tech's Potential in AI & Data Center Evolution
- 650 Group's Alan Weckel positions CXL to compete with NVLink for enhancing AI and machine learning servers.
- Growing bandwidth demand requires new and multiple switches per rack, with spending shifting from cloud to AI.
Full summary
Alan Weckel, Founder and Technology Analyst at 650 Group, discussed the potential of CXL technology to compete with NVLink, emphasizing its importance in enhancing servers for AI and machine learning. Weckel also noted the rapidly growing demand for bandwidth, necessitating new high switches and multiple switches per rack, and highlighted the shift in spending from cloud to AI, likening it to the significant architectural shift brought about by cloud technology 20 years ago.
Democratizing Networking with SONiC
- Broadcom democratizes networking with open-source SONiC, bringing hyperscaler features to all users.
- Enhanced enterprise-class SONiC and silicon telemetry help customers quickly root-cause network and application issues.
Full summary
Kamran Naqvi, Principal Network Architect at Broadcom, discussed the company's efforts to democratize networking by making features previously exclusive to hyperscalers available to all users, including the Sonic software, an open-source network operating system. Broadcom has enhanced Sonic to be more enterprise-class, leading to increased adoption among enterprise customers, and has enabled silicon telemetry features in its silicon, which can help customers quickly identify the root cause of network or application performance issues.
CPO Switches are Here!
- Broadcom's Near Margalit says the company has tackled the main technical challenges of deploying CPO switches.
- Focus now shifts to proving reliability and cost-effectiveness of silicon photonics and shortening time to market.
Full summary
Broadcom's GM and VP of Optical Systems Division, Near Margalit, announced that the company has successfully tackled the main technical challenges of deploying CPO switches and is now focusing on demonstrating the reliability and cost-effectiveness of the core silicon photonics technology. Margalit also addressed concerns about the reliability of the laser component and the time to market, stating that Broadcom is working to shorten the time between the availability of CPO technology for integration with the switch and its market launch.
The Arms Race for AI Clusters
- Independent advisor Brad Booth describes AI data centers expanding rapidly as hyperscale and cloud players adopt AI.
- High GPU and optical module demand is spurring exploration of new materials for optics and memory to meet bandwidth needs.
Full summary
Brad Booth, an independent Ethernet and Optical Technology Advisor, discussed the rapid expansion of AI data centers, driven by AI's increasing capabilities and its adoption by hyperscale and cloud market players. He highlighted the high demand for GPUs and optical modules, the impact on memory devices, and the surge in innovation, including the exploration of new materials for optics and memory elements to meet bandwidth demands.
Moving Up the Stack with NaaS
- Vertical Systems Group's Erin Dunne sees layer 1 and 2 services like wavelengths and ethernet growing crucial.
- NaaS delivers the simplicity enterprises want, enabled by APIs and standards across vendors.
Full summary
Erin Dunne, Director of Research Services at Vertical Systems Group, emphasized the growing significance of network connectivity in facilitating automated services, with layer one and two services like wavelengths and ethernet becoming increasingly crucial. She also highlighted the role of Network as a Service (NaaS) in providing simplicity desired by enterprises, the importance of APIs and standards for service providers and equipment vendors, and the efforts of MEF in enabling cost-efficient and simple automated services among multiple vendors.
Transforming Data Centers with PCIe, CXL and Ethernet
- Astera Labs enables PCI Express infrastructure to scale with growing GPUs and AI accelerators.
- It partnered with Lenovo and AMD on CXL memory expanders and showcased an ethernet retimer for high-speed rack connectivity.
Full summary
At the OCP 2023 event,Thad Omura, SVP of Business and Corporate Development at Astera Labs, discussed the company's advancements in scaling AI platforms and cloud infrastructure, including enabling PCI Express infrastructure to scale with increasing GPUs and AI accelerators. Astera Labs has also partnered with Lenovo and AMD to offer unprecedented memory capacity using their CXL memory expanders, and showcased their industry-leading ethernet retim used for active electrical cases, allowing high-speed connectivity in the rack and from switch to switch.
800ZR/ZR+ OSFP for Regional Data Center Interconnects
- Marvell's 800 Gbps ZR/ZR+ coherent pluggable module converts a switch or router into a transport platform.
- Leveraging Marvell's DSP, TIAs, drivers, and silicon photonics, it enables 25.6 terabits single-fiber capacity between data centers.
Full summary
Radha Nagarajan, SVP and CTO of Optical and Cloud at Marvell, highlighted the company's new 800 Gbps ZR/ZR+ coherent pluggable module designed to enhance data center connectivity by converting a switch or router into a transport platform. The module, which leverages Marvell's capabilities at 800 gig, including the DSP, high-speed TIAs, drivers, and silicon photonics, enables a single fiber capacity of 25.6 terabits between data centers, facilitating larger data aggregation.
Power Efficiency of Next Gen CPO Data Center Switches
- Micas Networks presented switches co-packaged with Broadcom optics delivering up to 30% lower power consumption.
- Its co-packaged optics technology interests large hyperscalers seeking data center power savings.
Full summary
Max Simmons, Chief Marketing Officer at Micas Networks, presented at the OCP Global Summit 2023 about the company's new switches co-packaged with optics from Broadcom, which deliver up to 30% lower power consumption. Micas Networks showcased their C-package optics technology, which is of interest to large hyperscalers for power savings in data centers, and is eager to collaborate with these companies for its implementation in the future.
What's next for multi-cloud routing and switching in 2023?
- Arrcus's Shekar Ayyar expects telcos and cloud operators to further integrate their routing and switching infrastructure.
- The open 2023 question is whether multi-cloud networking converges into one big mash-up.
Full summary
As we look ahead to the future of routing and switching in 2023, how will telcos and cloud operators integrate their infrastructure? Are we headed to a big mash-up? Here are 3 predictions from Arrcus’s Shekar Ayyar .
What's next for data center technology in 2023?
- Credo's Scott Feller predicts whether 800G pluggables arrive for data center switches in 2023.
Full summary
Will 800G pluggables arrive for data center switches in 2023? Credo's Scott Feller makes his predictions for data center technology in 2023.
What's next for data center switching spending in 2023?
- Dell'Oro Group's Sameh Boujelbene shares three predictions on data center switching in 2023.
Full summary
Here are 3 predictions from Dell’Oro Group’s Sameh Boujelbene on data center switching in 2023.
What's next at OCP in 2023?
- Open Compute Project's Cliff Grossner spotlights immersion cooling, optical switching, and disaggregated systems.
- He weighs which of these new OCP technologies grab the limelight in 2023.
Full summary
From immersion cooling, to optical switching, and disaggregated systems there are a lot of new technologies arriving at the Open Compute Project Project. Which ones will be in the limelight in 2023? Open Compute Project's Cliff Grossner makes his predictions for OCP in 2023.
Private 5G is about Business Outcomes
- NTT's Parm Sandhu says customers want business outcomes like automation, worker safety, and productivity, not just networks.
- NTT offers private 5G as a service on a monthly subscription, starting from the desired outcome and financials.
Full summary
Parm Sandhu, VP of Enterprise 5G Products & Services at NTT, highlighted that customers are more interested in achieving business outcomes such as automation, worker safety, and productivity, rather than just purchasing a network or edge compute. NTT has been successful in meeting these needs by starting with the desired business outcome, working out the financials, and offering private 5G as a service, which allows customers to switch to a monthly subscription model, with a focus on providing secure, robust, and reliable communication for mission-critical use cases.
1.6Tbps OSFP-XD AECsfor hyperscale spines
- Credo unveils a 1.6Tbps OSFP-XD HiWire CLOS active electrical cable supporting lengths up to 2.75m.
- The thin, compact cable targets dense spine switching for next-generation hyperscale data centers.
Full summary
Don Barnetson, Vice President of AEC Product, Credo discusses their recently unveileed 1.6Tbps OSFP-XD HiWire CLOS Active Electrical Cable (AEC) supporting lengths up to 2.75m in a thin, compact, cable form factor to address the dense spine switching requirements for next generation hyperscale data centers.
Market drivers for Active Electrical Cables
- Marvell's AECs, powered by Alaska PAM4 DSPs, overcome the distance limits of passive direct attach cables.
- They enable 400G, 800G, and 1.6T server-to-switch and switch-to-switch interconnects using 100G serial I/Os.
Full summary
Venu Balasubramonian, VP of product marketing, High Speed Connectivity and PHY Business Unit, Marvell, explains how AECs (such as those powered by Marvell Alaska’s PAM4 DSPs) overcome distance constraints of passive direct attach cables and enable next-generation 400G, 800G and 1.6T server-to-switch and switch-to-switch interconnects that require 100G serial I/Os.
OCP's Future Technology Symposium
- OCP's Lesya Dymyd shares highlights from the Future Technology Symposium spanning academia, startups, and industry.
- Submitted papers covered five tracks: Network, HPC, Data Center Efficiency, Systems, and Power and Sustainability.
Full summary
OCP’s Lesya Dymyd shares highlights from the Future Technology Symposium at the OCP Global Summit 2022 which invited authors from academia, startup, industry and venture communities to submit papers across five technological tracks: Network, High Performance Computing (HPC), Data Center Efficiency, Systems (server, storage, switches) and Power and Other Data Center Sustainable Solutions.
Automotive Ethernet, Car-to-Cloud, Custom Silicon
- Marvell's Will Chu points to automotive Ethernet, car-to-cloud, and custom silicon in 2022.
Full summary
Will Chu, SVP & General Manager, Automotive at Marvell, shares predictions for 2022. Download AvidThink's 2021 Select Report Collection to explore strategies and opportunities in the Telco, Private Mobile Networks, Service Assurance, SD-WAN and Open RAN space.
Can SONiC become the Linux of Network Operating Systems?
- Microsoft's David Maltz details SONiC, an open-source Linux-based network operating system.
- SONiC runs on switches from multiple vendors and ASICs.
Full summary
In this video, David Maltz, Technical Fellow and Corporate Vice President at Azure Networking, details the features and benefits of SONiC, an open source network operating system based on Linux that runs on switches from multiple vendors and ASICs. Download AvidThink's 2022 SmartNICs and Infrastructure Acceleration Report to explore developments in network acceleration.
SD-WAN and SASE need a Broadband Boost
- Masergy's Ajay Pandya says broadband adoption is becoming a regulation-level expectation in the enterprise.
- Masergy Performance Edge boosts an SD-WAN connection to perform more like a private line or ethernet circuit.
Full summary
Broadband adoption is becoming a regulation in the enterprise. In this video, Ajay Pandya, Director of Product Management at Masergy, discusses why a broadband boost is needed for SD-WAN and how Masergy Performance Edge boosts the performance of a SD-WAN connection, making it perform more like a private line or an ethernet circuit.
NGCO Network Infrastructure
- Broadcom's Rochan Sankar calls the NGCO a confluence of vision and product development.
- The NGCO requires virtualization and cloudification, with Ethernet-based interconnects opening room for innovation.
Full summary
The NGCO is an amazing opportunity and a confluence of vision and product development, says Rochan Sankar, Senior Director, Switch Products, Broadcom. NGCO requires virtualization and cloudification of the underlying network. The Ethernet-based interconnects for this infrastructure presents many opportunities for innovation.
Stop Burning CPU Cores on Networking and Security
- Ethernity offloads the data path on its FPGA SmartNIC where general-purpose CPUs fall short.
- It accelerates mobile and fixed networks and offers complete switch-router capability on a NIC.
Full summary
Network functions in the NGCO require high throughput, low-latency, and deterministic behavior which general-purpose CPUs cannot always deliver. Ethernity provides data path offload on its FPGA SmartNIC, accelerating mobile and fixed networks to meet next-generation requirements and offering complete switch-router capability on a NIC.