Cisco took action at Cisco Live Amsterdam to bring hyperscale to a wider range of users, introducing new chips and systems for hyperscalers, AI-savvy service providers, and enterprises.
At the heart of this release is Cisco's new Silicon One G300 switching chip, designed to provide the powerful scale-out networking capabilities needed to connect gigawatt-scale AI clusters. The chip notably uses a technology that Cisco calls an "intelligent collection network" that optimizes path selection and includes a buffering mechanism to absorb bursts of heavy AI traffic without packet loss. Cisco says this technology can help achieve up to 33% increase in network utilization and shorten task completion time.
Sameh Boujelbene, Group Vice President of Dell'Oro Group, told Fierce: "Cisco is trying to solve a real pain point that comes with AI and large-scale cluster workloads: unpredictable traffic patterns, huge east-west data flows, and the scaling limitations of traditional architectures. All network providers are trying to solve these problems and enhance the Ethernet architecture to reduce task completion time and optimize the number of tokens generated per GPU hour, which will directly impact profitability. ”
The G300 chip is the basis for two new Ethernet switching systems: Cisco's 102.4T N9100 and Cisco 8000 Series, both available in air- and liquid-cooled versions. Cisco noted that the liquid-cooled version improves energy efficiency by nearly 70%. Alan Weckel, founder and principal analyst at 650 Group, told Fierce that energy efficiency is becoming an increasingly important differentiator in the market. "The power efficiency of systems and solutions is what customers value because it allows them to run more GPUs on the same power budget, which can be a matter of whether a new cloud provider is losing money or making a profit," he explained. ”
Fierce competition in a $100 billion market
Cisco's new products are entering an extremely competitive AI back-end networking market. The company is competing with Arista, Celestica, Accton, Juniper, Nokia and NVIDIA, and the Dell'Oro Group expects to spend more than $100 billion in this market by 2030. Boujelbene told Fierce that Celestica and Nvidia will collectively hold 50% of the market share in the space for most of 2025, but she noted that Cisco will start gaining share in the second half of 2025.
"It's a highly competitive market. Arista and Celestica have built strong established strengths in the front-end networking space of hyperscalers. "At the same time, there is a clear need for vendor diversity and a strong desire for innovation in AI back-end networks." Cisco is approaching this problem with a different perspective. Both Boujelbene and Weckel point to the breadth of Cisco's portfolio — from purpose-built chips (ASICS) and optical transceivers to systems and software — as a key advantage for the company's future growth.
Cisco expands into software and services
In fact, Cisco's willingness to expand beyond its traditional networking hardware vendor role is clear. The company is heavily promoting updates to its Nexus One software platform, which now updates unified dashboards, operating system interoperability, integrated Splunk visibility, and Cisco Live Protect capabilities – all designed to support AI deployments. The vendor has also expanded its layout in the AI services space with a new AIOps tool built on the Cisco Deep Networking model, which has been specifically trained for network troubleshooting.
Kevin Wollenweber, senior vice president and general manager of data centers and internet infrastructure at Cisco, told Fierce that expanding its existing Nexus One platform to accommodate the needs of AI deployments is a deliberate move. "When we talk to customers about whether we can evolve the Nexus data center stack and network stack that they already have deployed in enterprise data centers to be able to run AI workloads on it, it's very attractive because they don't have to learn new tools," he said. He concluded: "Learning a new operating system just to deploy AI doesn't feel like something many of our enterprise customers want to do. ”
