The rapid expansion of AI is leading to a supply-constrained environment for network bandwidth and infrastructure, rather than an overbuild, explains Cisco President and Chief Product Officer Jeetu Patel. He asserts that the current situation is not comparable to the dot-com boom, where infrastructure was built speculatively. In contrast, all AI infrastructure being built today is immediately consumed.
Patel highlights that even with less than 2% of people currently using AI agents, the existing infrastructure is already facing supply constraints. This is largely due to the high consumption rate of AI agents; an agent uses 450% more network bandwidth than a human performing the same task. The exponential growth of AI, with the potential for trillions of agents performing tasks and constantly enriching context, will demand significantly more infrastructure and generate a massive volume of tokens.
The sheer amount of tokens required to process and manage large datasets, such as 140,000 emails, underscores why the current pace of data center buildout will likely remain insufficient. Patel emphasizes that this secular shift will fundamentally change how people live, necessitating substantial infrastructure investment, and that any perceived ‘overbuild’ is actually a struggle to keep up with escalating demand.