Cisco Makes Its AI Case Around Agent-Driven Network Demand

Cisco supplies equipment that links GPUs inside large data-centre clusters, helping data flow while companies including OpenAI and Anthropic train models and deliver ChatGPT and Claude Code. [1] The FT’s interviewer said Cisco had signed deals worth more than $10bn with hyperscalers and start-up cloud providers during the current fiscal year. [1] Techmeme described the interview as covering Cisco’s multibillion-dollar AI networking deals and competition from partners such as Nvidia. [2]

Patel said an AI agent performing the same task as a human consumes 450 per cent more network bandwidth than that human. [1] He argued that GPUs do not provide value without networking and that growing capacity requirements give Cisco a natural advantage. [1] Cisco makes its own silicon for networking switches, routers and WiFi, along with photonics technology that uses light to transmit data. [1]

Cisco’s argument ties the transition from chatbots to agents directly to infrastructure demand: if agents require more bandwidth for equivalent work, networking becomes part of AI capacity planning rather than an auxiliary purchase. The competitive picture is more complex because Nvidia is described as both a partner and competitor, even as Patel argues that GPUs depend on networks to produce value. Cisco’s ownership of networking silicon and photonics gives it multiple components around which to build that case. What the sources don't address: whether Cisco’s bandwidth estimate applies consistently across different agent designs, tasks and deployment environments.