Research Finder
Find by Keyword
Cisco Launches Next-Gen Silicon and Systems to Power the Agentic AI Transformation
Cisco is redefining AI infrastructure by integrating its 102.4 Tbps Silicon One G300 with liquid-cooled systems and unified management software, transforming the network into a high-efficiency control plane that maximizes GPU performance while slashing energy consumption and operational complexity.
2/11/2026
Key Highlights
Through Intelligent Collective Networking, Cisco achieves a 28% reduction in job completion times and a 33% increase in network utilization, enablng enterprises to generate more tokens per GPU-hour.
The Silicon One G300 chip delivers a massive 102.4 Tbps of switching capacity, closing the performance gap between standard Ethernet and proprietary InfiniBand to challenge NVIDIA’s dominance in AI training.
By integrating 100% liquid-cooled designs and 1.6T optics, Cisco's new N9000 and 8000 systems improve energy efficiency by nearly 70%, enabling one unit to provide the bandwidth previously requiring six systems.
Unlike competitors using multi-die chiplets, Cisco’s Unified Architecture maintains a single silicon design from the edge to the core, significantly simplifying the software stack and reducing operational friction.
Cisco distinguishes itself as a one-stop shop by providing a fully integrated stack, including the chassis, liquid cooling, high-density optics, and the Nexus One management layer, rather than just standalone silicon.
Enhanced with native Splunk integration, the Nexus One platform provides job-aware observability from the network down to the GPU, allowing for real-time bottleneck resolution and secure data locality.
The News
Cisco unveiled the Silicon One G300, a 102.4 Tbps switching silicon designed for massive scale out AI cluster buildouts. The Cisco Silicon One G300 will power new Cisco N9000 and Cisco 8000 systems that can push the frontier of AI networking. The systems feature liquid cooling and support high-density optics to achieve new efficiency benchmarks and ensure customers get the most out of their GPU investments. In addition, the company enhances Nexus One to make it easier than ever for enterprises to operate their AI networks, on-premises or in the cloud, aiming to remove a key barrier holding organizations back from scaling AI data centers. For more information, read the Cisco press release.
Analyst Take
Cisco has set a new benchmark for AI infrastructure with the introduction of the Silicon One G300, a 102.4 Tbps switching chip specifically engineered for massive scale-out AI clusters. This new silicon will anchor the next generation of Cisco N9000 and 8000 systems, seeking to push the boundaries of network efficiency through integrated liquid cooling and high-density optics.
By optimizing these systems for maximum GPU use, Cisco is directly addressing the hardware demands of the Agentic Era, ensuring that enterprise investments in high-end compute deliver peak performance. Additionally, enhancements to the Nexus One platform simplify the management of these networks across hybrid environments, eliminating the operational friction that often prevents organizations from successfully scaling their AI data centers.
We see the G300 distinguishing itself through Intelligent Collective Networking, a hardware-centric innovation that significantly boosts data center profitability. By combining a fully shared packet buffer with proactive telemetry and intelligent path-based load balancing, the G300 can absorb the bursty traffic patterns typical of AI workloads and prevent packet drops that would otherwise stall critical jobs. This approach results in a 33% increase in network utilization and a 28% reduction in job completion times compared to non-optimized systems. For enterprises, this translates into a higher volume of tokens generated per GPU-hour, making their AI infrastructure more productive and cost-efficient.
Designed for long-term viability, the Silicon One G300 architecture is highly programmable and natively secure. This flexibility enables organizations to upgrade network functionality and adapt to emerging AI use cases without the need for expensive hardware overhauls. Furthermore, with security features fused directly into the silicon, clusters can maintain holistic, at-speed protection to ensure constant uptime. From our viewpoint, the Silicon One unified architecture has evolved into the industry's most versatile portfolio, providing a consistent networking foundation that spans from hyperscalers and service providers to the modern AI-driven enterprise.
Beyond the Passive Pipe: Engineering the AI Control Plane with Cisco Silicon One
We see the Cisco announcement as signaling a major architectural pivot in the AI networking industry, steering the sector away from proprietary black box systems toward a unified, Ethernet-based standard for massive GPU clusters. By reaching the 102.4 Tbps threshold with its Silicon One G300, Cisco has closed the performance gap between Ethernet and InfiniBand, directly challenging NVIDIA’s historical dominance in high-end AI training.
This evolution, supported by 100% liquid-cooled designs and 1.6T optics, sets a new industry benchmark for sustainability and forces competitors to address the power constraints currently limiting AI data center scale. Furthermore, by integrating deep telemetry and security directly into the silicon, Cisco is transforming the network from a passive pipe into an active AI control plane that directly accelerates GPU job completion times.
In the 102.4 Tbps generation, the competition between the Cisco Silicon One G300 and the Broadcom Tomahawk 6 has emerged as the primary battleground for the future of the AI data center. While both chips achieve the same massive throughput milestone, they represent distinct architectural philosophies. Cisco’s primary advantage lies in its integrated systems approach; rather than selling a standalone merchant chip, Cisco provides the entire cooled chassis, optics, and the Nexus One management layer. This positioning makes Cisco a highly attractive one-stop shop for enterprises and neoclouds that prefer a turnkey solution over the complexity of designing custom hardware around third-party silicon.
The fundamental distinction in architectural efficiency is found in the physical construction of the silicon itself. Broadcom’s Tomahawk 6 uses a multi-die chiplet architecture, which strategically balances high-speed I/O and processing power by splitting components across multiple dies. In contrast, Cisco champions a Unified Architecture approach, maintaining a single, consistent silicon design across every network role from the edge to the core. This consistency simplifies the software stack and eliminates the operational friction typically associated with managing disparate hardware types within the same fabric.
Next-Gen AI Infrastructure: Cisco Debuts 102.4T Liquid-Cooled Systems and 1.6T Optics
To support AI network builders ranging from hyperscale giants to enterprise players, Cisco is launching the next generation of its N9000 and 8000 series Ethernet systems. These fixed and modular platforms are powered by the Silicon One architecture and specifically engineered to withstand the extreme thermal and power requirements of modern AI workloads. Central to this launch are the 102.4T systems featuring the Silicon One G300, which is set to establish a new industry standard for performance. By offering 100% liquid-cooled designs, these systems provide significantly higher bandwidth density and achieve a nearly 70% improvement in energy efficiency, allowing a single unit to deliver the same bandwidth that previously required six separate systems.
The hardware evolution is complemented by a new suite of innovative optics designed to maximize reliability and efficiency. The 1.6T OSFP optics provide ultra-high bandwidth connectivity for AI scale-out solutions, targeting everything from switch-to-NIC links to various server connections. Additionally, Cisco is introducing 800G Linear Pluggable Optics (LPO), which reduces optical module power consumption by 50%. When paired with the new N9000 and 8000 systems, LPO technology allows customers to cut overall switch power by 30%, fostering more sustainable and reliable data center operations.
Cisco is also broadening its portfolio of Silicon One P200-powered scale across systems to serve neoclouds, service providers, and enterprises. Building on existing 51.2T hyperscale deployments, the expansion includes new P200-powered N9000 systems and modular 28.8T line cards, along with expanded OS support for 8223 platforms. When combined with Cisco’s 800G ZR/ZR+ coherent pluggable optics, these offerings enable a wide range of customers to implement a common architecture across diverse network roles, including data center interconnects, universal spines, and core routing.
To manage these evolving infrastructures, Cisco is advancing Nexus One with a unified operating model that combines silicon, platforms, optics, and programmable software into a single solution. The new Unified Fabric approach enables Nexus One to converge ACI, NX-OS, SONiC, and the new Nexus Hyperfabric onto common N9000 hardware. This enables organizations to manage multiple fabric types through a consistent operational model, preventing team fragmentation and eliminating the need for extensive retraining while providing built-in, API-driven automation.
Nexus One is being enhanced with AI Job Observability and native Splunk integration to provide deep, job-aware visibility from the network down to the GPU. This feature correlates network telemetry directly with AI workload behavior, allowing customers to analyze data where it resides without the need for external transfers. This capability is particularly critical for sovereign cloud deployments and compliance-sensitive environments where data locality and security are paramount, ensuring that network bottlenecks are identified and resolved within the local infrastructure.
Looking Ahead
We believe that by innovating across the entire stack, from high-performance silicon to integrated software, Cisco is transforming the network from a passive connection into an active component of the AI compute engine. The introduction of the Silicon One G300 and new N9000/8000 systems ensures that organizations can move past the limitations of raw GPU speed to achieve scalable, congestion-free data movement. This full-stack approach enables every customer, from global hyperscalers to private enterprises, to fully maximize their hardware investments and deploy secure, deterministic AI infrastructure at massive scale.
As such, Cisco is positioned to succeed by delivering a 28% reduction in AI job completion times and a 33% increase in network utilization, directly maximizing the profitability of massive GPU investments. By leveraging its breakthrough Silicon One G300 architecture, the company's unique one-stop shop approach, combining 102.4 Tbps capacity with 100% liquid-cooled designs and high-density optics, sets a new industry benchmark for sustainability by improving energy efficiency by nearly 70%. Furthermore, by integrating deep telemetry and AI job observability into the Nexus One platform, Cisco removes critical operational friction, enabling enterprises to scale complex AI data centers across hybrid environments with ease.
To further advance the Silicon One strategy through 2026, Cisco should prioritize transitioning from high-performance components toward an integrated, scale-across ecosystem that seamlessly connects disparate AI clusters. This strategy can evolve through deeper integration of Ultra Ethernet Consortium (UEC) standards into the G300, ensuring its Intelligent Collective Networking remains the industry benchmark for multi-vendor, low-loss fabrics. Additionally, Cisco can capitalize on its thermal innovations by offering pre-validated "AI Factories" that bundle Silicon One with NVIDIA GPUs and Acacia coherent optics, significantly simplifying the transition for organizations currently struggling with power and cooling limitations.
Finally, the strategy would benefit from expanding the Nexus One AI Job Observability suite to provide predictive guided remediation, leveraging native Splunk analytics to resolve network bottlenecks before they impact GPU utilization. By aggressively pushing Silicon One into Unified Edge appliances, Cisco can refine its competitive edge, enabling the same high-performance architecture to manage real-time agentic AI workloads at the edge as effectively as it does in the core data center.
Ron Westfall | VP and Practice Leader for Infrastructure and Networking
Ron Westfall is a prominent analyst figure in technology and business transformation. Recognized as a Top 20 Analyst by AR Insights and a Tech Target contributor, his insights are featured in major media such as CNBC, Schwab Network, and NMG Media.
His expertise covers transformative fields such as Hybrid Cloud, AI Networking, Security Infrastructure, Edge Cloud Computing, Wireline/Wireless Connectivity, and 5G-IoT. Ron bridges the gap between C-suite strategic goals and the practical needs of end users and partners, driving technology ROI for leading organizations.
Share
Stephen Sopko | Analyst-in-Residence – Semiconductors & Deep Tech
Stephen Sopko is an Analyst-in-Residence specializing in semiconductors and the deep technologies powering today’s innovation ecosystem. With decades of executive experience spanning Fortune 100, government, and startups, he provides actionable insights by connecting market trends and cutting-edge technologies to business outcomes.
Stephen’s expertise in analyzing the entire buyer’s journey, from technology acquisition to implementation, was refined during his tenure as co-founder and COO of Palisade Compliance, where he helped Fortune 500 clients optimize technology investments. His ability to identify opportunities at the intersection of semiconductors, emerging technologies, and enterprise needs makes him a sought-after advisor to stakeholders navigating complex decisions.