The Hidden Problem With Cloud AI | Cisco Data Center Networking

Why It Matters

This matters because cost and data-security pressures could slow innovation if researchers remain cloud-only, and will push universities, enterprises, and service providers to invest in on-prem or hybrid infrastructure—reshaping AI deployment, vendor strategies, and data governance.

Summary

Speakers argue that relying solely on rented cloud GPU resources is discouraging AI research because usage-based billing forces researchers to limit experimentation. They advocate for on-premises GPU infrastructure—capitalized once and reused over long lifecycles—to enable sustained exploration, hand down hardware to students, and reduce per-token cost pressure. The conversation also highlights a shift away from hyperscale AI clouds for many users, driven by rising cost sensitivity, data confidentiality concerns, and challenges around moving petabytes of edge or distributed data to a central cloud. These factors are prompting institutions to reconsider decentralized, on-prem or hybrid models for AI workloads.

Original Description

Data security and massive file sizes are making the public cloud difficult for many researchers. It is time to move away from the giants and get your compute closer to your data. Explore the benefits of edge data and on-prem security with Cisco Data Center Networking's presentation at AI Infrastructure Field Day. #AIIFD4