Cisco has introduced its Silicon One G300 switching silicon and new Nexus 9000 and Cisco 8000 systems that it says deliver 102.4 Tbps switching capacity for AI data centers.
In its announcement, Cisco positioned the launch around scaling “agentic” AI-era infrastructure and said some new systems are available as a 100% liquid-cooled design.
Cisco is a networking vendor whose products are widely used in enterprise and service-provider networks. In its fiscal 2025 earnings release, Cisco reported $56.7 billion in revenue, underscoring the scale of its enterprise networking business as it pushes into AI data center fabrics.
“Silicon One” is Cisco’s in-house networking silicon family, meaning the custom chips that power routers and switches and move traffic across data center fabrics.
The G300 is the newest Silicon One switching chip Cisco is positioning for AI cluster networks and the silicon sits underneath the new systems Cisco announced this week.
Cisco’s Nexus 9000 line is its data center switching portfolio, typically deployed as leaf and spine switches that connect servers, storage and GPU clusters inside modern data centers.
That launch lands as HPE’s Juniper-branded QFX5250 is marketed as a liquid-cooled AI data center switch built on Broadcom Tomahawk 6 and rated at 102.4 Tbps.
Programmability and In-Place Upgrades
Cisco’s G300 data sheet describes the chip as a standalone switching processor aimed at fixed-form-factor data center spine and leaf designs spanning front-end, back-end, scale-out and scale-up roles.
The release also said the platform supports “in-place” programmability.
Cisco said the liquid-cooled design, paired with new optics, can improve energy efficiency by nearly 70%, and it positioned Nexus One as a unified management plane for faster fabric bring-up and operations.
AgenticOps expands from “AI for infrastructure” into multi-domain operations
Alongside the hardware, Cisco said it is extending AgenticOps across networking, security and observability, using unified telemetry and its Deep Network Model to support troubleshooting and automation.
Cisco and third-party coverage both highlighted “autonomous troubleshooting,” including multi-step investigations, root cause analysis and workflow-driven changes with validation before execution.
Cisco has also been building an observability data layer around Splunk. A Splunk press release tied Cisco Data Fabric to a planned AI Canvas integration and a Splunk Machine Data Lake in 2026, and Cisco’s time-series foundation model is published on Hugging Face as an open preview.
What the competitive timing signals for enterprise buyers
HPE is pitching a Broadcom-based, Juniper-branded path for high-bandwidth AI switching, while Cisco is pitching a vertically integrated alternative spanning silicon, systems, management and observability.
Cisco’s investor release pointed to second-half 2026 availability for the new systems, while its AgenticOps post described rollout beginning in February 2026.