Open Networking & Edge Summit 2025#

Core#

SONiC: The Global Networking Operating System for All#

Abstract Recording

Michael Ma, Microsoft; Peter Plak, Dell

• Why: SONiC is the closest thing the industry has to a standard open NOS for datacenter switches, and it is the layer where the PM's physical-fabric and cloud-ops interests converge — disaggregated switch silicon, automated fleet operations, and Azure-scale production validation. Whatever SONiC's community roadmap is this cycle directly affects build-vs-buy for DC fabrics.
• What: A status-and-roadmap session on SONiC's evolution as a modular, hardware-agnostic open source network OS, plus the speaker's own work at Microsoft Azure on accelerating network device updates using container technologies (from abstract only).
• How: SONiC decouples the NOS from switch hardware via containerized subsystems and SAI as the switch-abstraction layer; the Azure-side contribution described here is using container tooling to make switch software updates faster and less disruptive — a direct operational-velocity win, though specifics (mechanism, measured update times) are not in the abstract.
• Where applicable: Broadly applicable across merchant-silicon datacenter fleets (broad ecosystem of ODM/merchant switch platforms via SAI). The Azure device-update experience is Microsoft-specific in scale but the technique is general. Caveat: this is a co-presented community/vendor session, so expect roadmap framing rather than empirical evaluation.

IOWN Global Forum and Open Source - Key Enablers for Next Decade#

Abstract Recording

Masahisa Kawashima, NTT Corporation

• Why: IOWN is one of the few serious large-scale programs pushing integrated photonics into compute and network infrastructure — photonics-based DC interconnect, high-capacity optical I/O, and energy-efficient transport. For a PM whose core is optics and physical DC/WAN systems, this is a rare look at where the optical roadmaps of a major operator actually intersect open source.
• What: A roadmap talk on NTT's Innovative Optical and Wireless Network (IOWN) architecture and, importantly, how IOWN is now partnering with open source communities — LFN, LF Edge, and OPI — to integrate open source software into the IOWN platform (from abstract only).
• How: Not a technique talk: the mechanism is cross-organizational — the IOWN Global Forum defines the photonic architecture requirements, and the partnerships with LFN/LF Edge/OPI are intended to supply the open software layer (orchestration, provisioning, DPU-adjacent compute interfaces) around it.
• Where applicable: NTT/IOWN-specific architecture, but the open-source integration point matters broadly — OPI in particular covers DPU/accelerator provisioning standards, which has cross-industry applicability. Expect vision and roadmap depth, not deployment results.

AI - Lessons Learnt from World's Largest Network#

Abstract Recording

Dr. Junlan Feng, China Mobile Research Institute

• Why: This is the ml-for-networking slot — a decade-plus of applied AI on the world's largest operator network, from intent-based operations to autonomous networks. For a PM interested in ML for infrastructure operations, hearing what actually worked (and didn't) at China Mobile's scale is a higher-value signal than most vendor AIOps pitches.
• What: Lessons from China Mobile's AI journey operating the largest service-provider network: the progression from intent-based networking to automated/autonomous networks, and the role of open technology adoption in that progression (from abstract only).
• How: Not specified in the abstract — expect a retrospective of AI/autonomous-networking deployments (fault prediction, closed-loop operations, intent translation) rather than a single novel technique.
• Where applicable: Operator-scale and telecom-specific; the lessons about staging intent → automation → autonomy likely transfer, but the data and network scale are China Mobile's. Treat reported results with the usual caution until specifics appear.

Adjacent#

State of Open Source Networking - Arpit Joshipura & Hilary Carter, The Linux Foundation#

Abstract Recording

Arpit Joshipura & Hilary Carter, The Linux Foundation

• Why: The ecosystem-setting keynote, and the one place in this program where survey data (LF Networking's global research survey) about real open source networking adoption and deployment will be presented — useful calibration for where open networking actually is versus the hype.
• What: Vision, predictions, real-world deployment examples, and results of the LF Networking global research survey (from abstract only).
• How: Survey methodology and sample are not described in the abstract; treat the numbers accordingly.
• Where applicable: Cross-industry view rather than a specific environment.

An Industry Analyst's Perspective on Telco Cloud#

Abstract Recording

Christoforos Sarantopoulos, James Crawshaw & Inderpreet Kaur, Omdia

• Why: A condensed analyst view of telco cloud trends — operational complexity, resource-utilization architecture choices, CI/CD, and AI in cloud ops. The reskilling/data-governance/CI-CD recommendations are the transferable part for anyone running private cloud or NFV estate.
• What: An overview of telco cloud evolution from NFV through cloudified 5G core, with recommendations for service providers (reskilling, data governance, CI/CD) and advice for vendors on supporting open-source adoption and a common automation framework (from abstract only).
• How: Analyst synthesis across operators and vendors; no single mechanism.
• Where applicable: Telco/operator context. Analyst talks of this kind are useful for pattern-spotting but carry no empirical evaluation — and vendor-recommendation content can shade into pitch.

Sponsored Session: Ready for Impact: EVE Steps from 10k Kubernetes Clusters to Edge AI#

Abstract Recording

Erik Nordmark, ZEDEDA

• Why: Flagged as a sponsored session, but with a concrete production number attached: EVE (LF Edge) is claimed to run 10,000+ single-node Kubernetes clusters and VMs in production. Fleet-style Kubernetes at the edge — one-cluster-per-site, upgrade, and resiliency mechanics — is directly relevant to the PM's K8s-operations interests, and edge AI inference workflow integration is the forward-looking part.
• What: A status talk on EVE's graduation to LF Edge's Impact stage, the 10k-cluster production footprint, current work to enable AI inference at the edge (workflows and integrations), and new resiliency efforts (from abstract only).
• How: EVE is an open source edge OS providing virtualization and container runtime isolation on distributed edge nodes, orchestrated centrally (ZEDEDA's commercial orchestration sits on top). The AI-inference enablement is described as workflow/integration work rather than a new mechanism.
• Where applicable: Edge deployments with single-node-per-site topology — quite different constraints from a central DC cluster. The 10k figure is ecosystem-reported, not independently audited; keep in mind the speaker is ZEDEDA's CTO.

Network Infrastructure to APIs - How Open Source Enables Speed of Innovation and Monetization#

Abstract Recording

Nathan Rader & Sebastian Zechlin, Deutsche Telekom

• Why: The CAMARA-style network-API movement is where telco network capability is being productized; understanding DT's open-source-based approach is useful context for anyone weighing API exposure as the interface layer between infrastructure and applications.
• What: DT's vision and journey of a fully open network spanning infrastructure through APIs (including CAMARA), framed around innovation speed, lower cost, and asset monetization, plus a roadmap for community contribution (from abstract only).
• How: Layered open source adoption across infrastructure, OSS, and the CAMARA API framework.
• Where applicable: Large-operator context; the CAMARA framework itself is multi-operator. Expect strategy and journey narrative rather than technical depth.

Standards and Open Source Harmonization - Silvia Almagia, ETSI; Hui Deng, Huawei; Raúl Álvarez, Open Compute Project Foundation; Moderated by Arpit Joshipura, The Linux Foundation#

Abstract Recording

Silvia Almagia (ETSI), Hui Deng (Huawei), Raúl Álvarez (OCP Foundation), moderated by Arpit Joshipura (The Linux Foundation)

• Why: The standards↔open-source convergence question (ETSI/NGMN/OCP vs. LFN projects) affects interoperability roadmaps in exactly the PM's domains — cloud-native networking, green/efficiency targets, and AI infrastructure. Knowing where this harmonization effort stands after five years is cheap but valuable context.
• What: A moderated panel reviewing progress and upcoming requirements for harmonizing standards and open source across cloud-native, green, and AI themes (from abstract only).
• How: Inter-organizational alignment between standards bodies (ETSI, NGMN) and open communities (OCP, LFN); no technical mechanism.
• Where applicable: Ecosystem-wide but abstract-level; expect directional signals, not results.

Wildcard#

AI-RAN and Open Source: Building Scalable Dataset Pipelines for AI Training#

Abstract Recording

Alex Jinsung Choi, SoftBank Corp. (Chair, AI-RAN Alliance)

• Why: Not about the PM's domains — it's telecom RAN — but the core problem is one they know intimately from ML systems: high-quality training data is the bottleneck, and open, collaborative dataset pipelines (aggregation, curation, annotation, synthetic data, with security/compliance guardrails) are an emerging pattern for any domain where proprietary data can't be shared. Watching how the AI-RAN Alliance solves data sharing under governance constraints is a preview of how other infrastructure-ML domains will.
• What: How open-source tooling can build collaborative dataset pipelines for AI-driven network optimization in RAN, including AI-RAN's role in synthetic data generation and its connection to global AI governance (from abstract only).
• How: Open-source pipelines for data aggregation, curation, and annotation with security/compliance built in, plus synthetic data generation to compensate for scarce labeled telecom data.
• Where applicable: Telecom RAN domain (SoftBank/AI-RAN Alliance-specific datasets), but the dataset-governance pattern is general.

InstantX: Vodafone's Edge Innovation for the Future of Telco#

Abstract Recording

Sampada Basarkar, Vodafone

• Why: Outside the PM's core, but notable as a signal: a major operator insourcing a production edge platform (V2X/A2X communication) into the open source community. When telcos start donating runtime platforms rather than just joining foundations, it says something about where edge infrastructure economics are heading.
• What: A keynote on Vodafone's InstantX edge platform — powering scalable V2X and A2X communication — and Vodafone's decision to insource the technology into open source, including the challenges overcome (from abstract only).
• How: Platform details are not given in the abstract; the mechanism of interest is the open-sourcing process itself (moving a proprietary telco platform into a community).
• Where applicable: Vodafone-specific production history; generalizable mainly as a precedent for operator platform contributions.

Skipped#

AI 101 for Networking & Edge (explicit beginner/buzzword tutorial), the Infosys sponsored session on AI frameworks and Responsible AI (vendor framework pitch with no validated results), Edge AI: The $100B+ Opportunity (consulting market forecast), the two AMA/panel community sessions, and the closing remarks — none offer transferable technical content.

Themes#

• Open source networking is consolidating around SONiC + LFN as the de facto stack, with SONiC now positioned as the "global NOS"; the interesting shift is operational maturity (faster device updates, container tooling) rather than new architecture.
• AI is the headline everywhere, but almost always as aspiration — only China Mobile's session promises deployment-scale lessons; most AI content here is framework, governance, or forecast talk.
• Telco cloud's center of gravity has moved from NFV to automation and CI/CD — the recurring threads across Omdia, Turkcell, and the DT sessions are operational complexity, resource utilization, and reskilling, not virtualization itself.
• Edge is being redefined by AI inference and by Kubernetes-at-the-edge scale (EVE's 10k single-node clusters being the concrete number in the program), moving edge from "IoT gateway" to "distributed K8s + inference."
• Optics is represented by IOWN's photonics roadmap, and its notable move is toward open source integration (LFN/LF Edge/OPI) — a signal that even hardware-centric programs now need an open software layer to be credible.
• Standards bodies and open source communities are formally converging (ETSI/NGMN/OCP with LFN), which will increasingly determine interop requirements in cloud-native, green, and AI infrastructure.