Open Source Takes on AI Clusters: SONiC Welcomes Lenovo and NEXAI

SAN FRANCISCO – October 05, 2026 – The artificial intelligence boom has exposed a critical vulnerability in modern data centers: the network. While the industry has historically fixated on the raw compute power of graphics processing units, the fabric connecting these processors has quietly become the ultimate bottleneck in large-scale model training. Today, the Software for Open Networking in the Cloud (SONiC) Foundation, an open-source network operating system hosted under the Linux Foundation, announced a significant maneuver to address this constraint. Lenovo, NADDOD, and Zhanwang (NEXAI) have officially joined as General Members, signaling a major industry push to mature open Ethernet fabrics for the AI era.

Originally developed for hyperscale cloud environments, SONiC has evolved into a linchpin for production networks worldwide. Its vendor-neutral architecture allows organizations to deploy and manage infrastructure across multiple hardware platforms, sidestepping the rigid confines of proprietary lock-in. Now, as AI and high-performance computing (HPC) place unprecedented demands on bandwidth, latency, and observability, the addition of these three hardware and silicon heavyweights underscores a broader market transition.

Open Source Takes on the AI Cluster

For years, proprietary high-speed interconnects like NVIDIA's InfiniBand have dominated the upper echelons of AI and HPC networking. They offered the lossless, ultra-low-latency performance required to keep massive GPU clusters fed with data, utilizing native credit-based flow control to prevent congestion. However, as enterprise AI adoption scales, operators are increasingly demanding the flexibility, cost-efficiency, and broad ecosystem support inherent to open Ethernet architectures.

Market projections reflect this shifting tide. Industry analysts estimate that the market for SONiC in data center switching will exceed $5 billion in revenue by 2026. This financial backdrop explains the urgency behind the foundation's latest expansion. Enterprise IT decision-makers are actively seeking ways to avoid the premium pricing and supply chain constraints of proprietary fabrics, making open-source alternatives highly attractive.

"AI infrastructure is raising the bar for networking, and the additions of Lenovo, NADDOD and NEXAI bring valuable knowledge from across the networking stack into the SONiC community," said Arpit Joshipura, general manager of networking, edge and IoT at the Linux Foundation. "Their participation will help advance an open, collaborative networking ecosystem that can evolve alongside the rapidly changing requirements of AI, cloud and data center infrastructure."

The challenge for open Ethernet has always been matching the lossless performance of proprietary fabrics. AI workloads are notoriously "bursty," generating massive, simultaneous data transfers that can easily overwhelm traditional network buffers. Solving this requires deep coordination between the network operating system and the underlying hardware—a synergy that the new members are specifically tasked with accelerating.

Lenovo's Enterprise Push and the 'Token Factory'

Lenovo's entry into the foundation represents a strategic expansion from enterprise server and AI compute dominance into hyperscale-grade open networking. The company is actively building automated, AI-driven operations capabilities anchored by SONiC, focusing on critical components such as Redis, the Switch Abstraction Interface (SAI), telemetry, BGP EVPN, and RDMA over Converged Ethernet (RoCE).

This initiative is encapsulated in a highly ambitious framework the company refers to as the "Token Factory." The concept envisions a deeply converged infrastructure where AI computing, storage, and networking operate seamlessly to power future AI generation. It represents a paradigm shift from selling isolated hardware boxes to delivering integrated, software-defined AI fabrics.

"Positioned at the forefront of the AI era, we are pioneering AI networking innovations anchored by SONiC to build Lenovo China's 'Token Factory': AI-powered AI empowers AI," said Billy Chen, ISG, VP, China Geo at Lenovo. "Guided by our vision of 'Smarter AI for All,' Lenovo China is partnering with the SONiC community to foster open innovation. By driving the deep convergence of AI compute, storage, and networking, we are building a next-generation Token-Driven Network to power future Token Factories."

By integrating open-source software with its vast hardware portfolio, Lenovo aims to challenge incumbent networking giants. The company plans to contribute deployment experiences and platform enhancements back to the community as its capabilities mature, effectively bridging the gap between hyperscale innovation and enterprise accessibility.

Silicon to Software: Taming AI Network Congestion

While Lenovo focuses on enterprise-scale orchestration, NADDOD and NEXAI are tackling the physical physics of data movement. Advancing open networking for AI requires intense innovation at the silicon and interconnect levels, particularly regarding RoCE and Priority Flow Control (PFC).

In distributed AI training, a single dropped packet can force an entire GPU cluster to pause and request retransmission, wasting millions of dollars in compute time. NEXAI is actively porting the open-source operating system to its high-performance switch silicon to prevent exactly this scenario. Their planned upstream contributions span RDMA, RoCE, PFC, SRv6, telemetry, and AI network observability. These protocols are the bedrock of lossless Ethernet, ensuring that data flows continuously without overwhelming the receiving nodes.

"Joining SONiC is a key milestone for NEXAI," said Jeff Zhang, Senior Director of Technical Marketing at NEXAI. "We believe open-source networking is foundational for next-gen AI data center infrastructure. We look forward to contributing our expertise in high-performance switch silicon and working hand-in-hand with the SONiC community to advance cloud networking innovation."

Complementing this silicon-level focus, NADDOD is deploying the operating system across its AI data center switch portfolio. The company specializes in high-speed optical interconnects and is contributing SAI enhancements, interoperability testing, and reference architectures derived from real-world deployments. Validating interoperability across different optical transceivers and switch ASICs is notoriously complex, making NADDOD's field-tested engineering perspective invaluable to the open-source community.

"Advancing open networking for AI requires innovation across network software, switching and interconnects," said Chang Liu, CTO at NADDOD. "NADDOD brings these together through our expertise in SONiC-based platforms and high-speed interconnects. We contribute an engineering perspective that connects hardware enablement, interoperability validation and RoCE optimization with the demands of today's AI infrastructure, supporting collective progress across the open AI networking ecosystem."

The Road to OCP Global Summit 2026

The timing of this ecosystem expansion is highly deliberate. The new members join just days before the 2026 Open Compute Project (OCP) Global Summit, scheduled for October 12-15 in San Jose. The summit serves as the premier gathering for open hardware and software innovators, and the foundation is poised to make a significant impact on the showroom floor.

During the event, the community will host a series of foundation-led activities, including live demonstrations at their booth and a co-located technical workshop on October 14. Furthermore, an extended offsite workshop will be hosted by Premier Member Nokia at its Sunnyvale office on October 16, underscoring the collaborative momentum driving the project forward.

As AI models continue to grow in parameter size and complexity, the network will remain the defining frontier of infrastructure performance. The integration of Lenovo's enterprise scale, NEXAI's silicon expertise, and NADDOD's interconnect architectures into a unified, open-source framework suggests that the industry is no longer willing to rely solely on proprietary solutions. By democratizing the tools required to build ultra-high-performance fabrics, this growing coalition is ensuring that the future of artificial intelligence remains open, interoperable, and fundamentally collaborative.