Cisco Boosts AI Networking Portfolio

Cisco unveiled a new Silicon One networking chip and associated products this week in a move that signals a major push to own its own destiny in the AI networking market.

The G300 chip, introduced at the Cisco LIVE EMEA conference in Amsterdam, brings 102.4 Tbit/s of Ethernet capacity to optimize the performance of networks linking large AI clusters. It can run at per-port rates of 1.6 Tbit/s and it supports to 512 Ethernet ports per chip, linking more GPUs at the network edge. The G300 is debuting in new N9000 and 8000 switches, which are liquid cooled for power efficiency.

The new chip is part of a large-scale redesign of Cisco's AI networking offerings, including an overhaul of the Nexus switch with the launch of the N9000, which is anchored by Cisco's streamlined management platform, Nexus Dashboard. Cisco has been busy in the past year, cutting deals with NVIDIA, updating its portfolio, and integrating its datacenter and AI networking operating systems. Cisco in November launched Cisco Nexus One Fabric, converging Cisco ACI and NX-OS.

The networking giant is also introducing new 1.6T OSFP (Octal Small Form-factor Pluggable Optics) that will offer customers high performance and reliability, 800G Linear Pluggable Optics (LPO), and an expanded portfolio of Silicon One P200-powered systems that build on the introduction of 51.2T systems for hyperscale deployments. The new P200-powered N9000 systems and expanded OS support on 8223 systems offer datacenter interconnect, universal spine, and core and peer routing capabilities to neoclouds, service providers, and enterprises, Cisco said.

According to Cisco, the network is critical to AI training and inference to get the most out of GPU investments. Silicon One G300 and the Cisco switches it powers are likely to compete with NVIDIA's Mellanox InfiniBand and Spectrum-X Ethernet Photonics switch systems as well as Broadcom's offerings such as its XPU and Tomahawk switches.

According to Jeetu Patel, Cisco President and Chief Product Officer, the chip delivers significant performance benefits. In a blog he states:

“The impact [of the G300] is measurable: 33% higher network utilization and 28% faster job completion times compared to non-optimized infrastructure. To give some perspective, this capability is in the order of five million simultaneous 4K Netflix streams.”

What Does It Mean for the Competition?

The onslaught of new products from Cisco shows how far it has come in a year. A little more than a year ago, the company launched a major reorg, including a change in leadership of the networking group, responding to fears it was losing ground in the AI infrastructure race.

This reorg led to the elevation of Patel, who is now widely seen as the heir-apparent to Cisco CEO Chuck Robbins. Patel has revitalized Cisco's networking portfolio and made it a major part of the AI infrastructure conversation. Cisco shares are hovering near an all-time high, adding hundreds of billions of dollars in value to the company in the space of a year.

The new announcements show that Cisco's in a competitive mood and it might now raise questions about Cisco's relationship with NVIDIA. In February of 2025, Cisco announced a deal with NVIDIA, packaging Cisco networking products with NVIDIA Spectrum-X offerings. Cisco has a parallel strategy of shipping its own networking products based on Silicon One but also another set of products with NVIDIA using Spectrum-X switches, which use NVIDIA's own silicon.

Interestingly, that arrangement isn’t highlighted in Cisco’s product blitz this week. An NVIDIA spokesperson says the G300 will be offered alongside other Cisco integrations for the Spectrum-X. Another major difference is that much of NVIDIA's portfolio uses NVIDIA's Cumulus operating system, whereas the products it ships with Cisco use Cisco's NX-OS.

The release of the G300 does appear to increase Cisco's competition with NVIDIA on the hardware front. NVIDIA's BlueField DPUs are playing an expanded role in optimizing the performance of NVIDIA-based AI networks.

The Cisco announcements increase Cisco's position against the rest of the competition. The race to sell networking gear for the largest AI projects has been primarily a battle among three giants: NVIDIA, Cisco, and Arista Networks.

Cisco is looking to bring its offering up to par with the strong portfolio offered by Arista, which is based on Broadcom silicon. Cisco's differentiation of the G300 won't be defined by its throughput credentials. Arista Networks, for instance, offers faster data rates with its 7800 chassis equipped with Broadcom Jericho3 silicon. That switch supports up to four 800-Gbit/s ports combinable into a single 3.2-Tbit/s Hyperport. Fewer links facilitates faster load balancing, Arista says.

Still, the G300 has several compelling capabilities, collectively dubbed Intelligent Collective Networking. These include an enlarged packet buffer embedded in the chip, which Cisco claims offers 2.5 times the absorption of traffic bursts compared to other solutions on the market. This reduces the potential for bursts to slow up traffic to GPUs.

The G300 also provides hardware-based autonomous load balancing to avoid congestion and delays. And network telemetry helps to head off problems and improve network optimization, Cisco says.

New Optical Modules Too

Another competitive differentiator will be Cisco's optical capabilities. Cisco is unique in owning a large optical components business, a path it went down in purchasing Acacia Communications in 2021.

Cisco broadened its component arsenal with new optical modules, including a 1.6-Tbit/s OSFP (Octal Small Form-factor Pluggable) connector for 1.6-Tbit/s switch-to-NIC and 1.6-Tbit/s, 800-Gbit/s, 400-Gbit/s, and 200-Gbit/s switch-to-server links. Notably, Arista also offers OSFP pluggables, as do a range of component suppliers.

Cisco also announced 800-Gbit/s Linear Pluggable Optics (LPO) modules to reduce power consumption in tandem with the N9000 and 8000 systems, which support LPO, which is also a technology widely adopted by component manufacturers. Cisco claims power consumption can be halved with the new modules and switches.

With the G300 and the new optical modules, Cisco is taking further steps to owning the entirety of the AI networking supply chain based on its products. The vendor now competes not only with suppliers such as Arista and Juniper but with their component suppliers, especially Broadcom.

Cisco Builds an Integrated Platform

The G300, Patel stresses, is part of an overall move to support agentic AI workloads in a platform that unifies silicon, switches, software, and security for hyperscalers, neoclouds, sovereign cloud providers, and enterprises.

In that vein, Cisco has augmented its AgenticOps management capabilities, which are based on a Deep Network Model LLM created by Cisco to provide a natural language interface for identifying and solving network issues automatically. Among new agentic functions are autonomous troubleshooting and continuous optimization for campus, branch, and industrial networks; improvements in observability and intelligent event correlation and performance optimization recommendations in Cisco Nexus One, the vendor’s datacenter fabric platform; and enhanced automation for service provider networks via the Cisco Crosswork automation platform.

Security is part of the raft of announcements, featured in updates to the Cisco AI Defense solution. Included are centralized visibility and governance for AI software assets; an MCP catalog; red-teaming enhancements for AI models and agents; and guardrails to detect threats.

Mary Jander contributed to the writing and reporting in this story.

Futuriom Take: Cisco’s impressive blast of networking products, particularly the new Silicon One G300, indicate the vendor’s ambition to dominate the AI networking market. Cisco isn’t waiting on any merchant silicon provider or third party to expand its reach across the entire gamut of customer demand from hyperscalers to enterprises of all sizes.