AI is Calling in 20 Years of Technical Debt Presented by Broadcom

Adding generative AI to enterprise applications is exposing up to 20 years of technical debt in enterprise IT infrastructure and applications. This episode of the Tech Field Day podcast features Sabina Anja from Broadcom’s VMware Cloud Foundation team discussing the impact of AI with Andy Banta, Ron Pagani Jr., and Alastair Cooke. The panel argues that the public cloud never eliminated technical debt, merely hid it behind a bill, and that AI workloads are now surfacing the same question of how quickly organizations can move from idea to production. Part of the problem is an “immortal VM” culture where decades-old virtual machines persist unchanged, creating a mismatch with ephemeral agentic AI workloads that must coexist with legacy systems on the same infrastructure. Compounding drivers include AI itself generating more code and debt through agentic coding, the disconnect between developers who build infrastructure informally and finance/ops teams locked into three-to-five-year hardware cycles even as GPU, RAM, and SSD economics shift within a year, and the “best of breed” fallacy of buying excellent components without solving integration and orchestration. The panel’s proposed remedy centres on treating infrastructure and process as inseparable, starting from clear business goals, building elastic software-defined resource pools that can be dynamically allocated via automation and CI/CD, and giving developers more direct, API-driven access to infrastructure. Sabina likened today’s GPU underutilization to mainframe time-sharing, and Alastair using a “Toyota Corolla vs. tuned AMG” analogy to argue for differentiating generic from purpose-built AI infrastructure.