I came back after spending a couple of interesting days at the Broadcom Mainframe Analyst event in Boston. I would be lying if I said I went in not expecting another AI platform story. Instead, Broadcom Mainframe Software presented a different and measured narrative.
It is not trying to create a loud Artificial Intelligence (AI) platform story or a one-size-fits-all AI answer. It seems to be doing something more consistent with its operating model: serve a select set of high-value customers deeply, understand their real operational demands, and apply AI where it improves outcomes.
That may sound less exciting than the AI platform stories we are hearing everywhere else, but in the mainframe world, it is pragmatic. The estate is complex, mission-critical, and surrounded by decades of business logic and seasoned practitioners. Broadcom’s focus is not about chasing the broadest market. It is about creating value for enterprises that already depend on its mainframe software every day. So, the decision not to force-fit a generic AI platform narrative is fair. Maybe even smart.
Reach out to discuss this topic in depth.
Broadcom’s AI story: Less theater, more relevance
What stood out was that Broadcom is not ignoring AI. It is just not presenting AI as a separate island. The approach is more practical: embed AI wherever it is relevant for enterprise mainframe customers.
Its security portfolio, including ACF2 and Top Secret, becomes especially relevant in an AI-enabled enterprise. If AI agents are going to observe, recommend, summarize, or take low-risk actions, the control layer around who or what gets access to the mainframe becomes critical.
Then there is WatchTower, Broadcom’s observability and AIOps play for the mainframe. AI in mainframe operations does not have to start with “rewrite my COBOL.” It can start with better signals, stronger correlation, and faster root cause analysis.
Broadcom also spoke about its involvement in Anthropic’s Project Glasswing and its experimentation with Mythos from the beginning. Despite multiple questions, no details were shared due to confidentiality. The responses came with smiles. I will leave that for the reader to interpret.
The “What if?” question that kept coming back
Throughout the event, my mind kept going back to a broader question: what does Infrastructure for AI really mean when applied to different technology environments, including mainframes?
I started brainstorming with several seasoned Broadcom mainframe experts around a simple question: “What if?”
What if Broadcom, as a company, brought capabilities from different parts of its organization together to create an Infrastructure for AI platform? Not just for generic enterprise AI, but for complex, secure, hybrid, regulated, high-availability environments.
That is where the conversation became interesting.
Broadcom’s theoretical Infrastructure for AI stack
To be clear, this is not the story Broadcom Mainframe Software was explicitly telling at the event. But if one steps back and looks at Broadcom as a company, it has one of the most interesting Infrastructure for AI portfolios in the market because it spans from silicon to private cloud to security to observability to mission-critical systems.
| Infrastructure for AI layer | Broadcom company-wide capability |
| 1. AI connectivity and silicon | Semiconductors, Ethernet switching, optical connectivity, PCIe, custom silicon |
| 2. Private / hybrid cloud control plane | VMware Cloud Foundation, vSphere, vSAN, NSX |
| 3. Network and security virtualization | VMware NSX, segmentation, security controls |
| 4. Enterprise security layer | Symantec, Carbon Black, ACF2, Top Secret |
| 5. Observability and operations | WatchTower, AIOps, VMware operations capabilities |
| 6. Automation and modernization | Mainframe DevOps, automation, modernization tooling |
| 7. Mission-critical systems | Broadcom Mainframe Software |
Broadcom’s potential story could start from silicon and connectivity at one end, all the way to the mission-critical mainframe estate at the other. That is a powerful arc, and a unique position to potentially build a mission-critical Infrastructure for AI platform.
The mainframe is not sitting outside the AI conversation anymore
For years, mainframe modernization was framed as migrate, rehost, refactor, retire, or keep the lights on. AI is changing that framing. Suddenly, the mainframe is not just a legacy estate to be dealt with. It is a source of trusted data, embedded business logic, operational resilience, and modernization opportunity.
IBM is pushing its “AI on Z” story through Watsonx Code Assistant for Z. AWS Transform for mainframe signals how hyperscalers see AI accelerating code analysis, business rule extraction, refactoring, and application reimagining. SIs are also positioning generative AI and agentic AI as part of legacy and mainframe modernization.
However, mainframe transformation will not be solved by AI agents alone. It will need practitioners, domain knowledge, governance, validation, and operating discipline. AI can accelerate the journey, but it cannot replace institutional context overnight.
What enterprises should take away
First, mainframe data readiness is now non-negotiable. The mainframe is unlikely to become the dominant infrastructure for running AI workloads, and I do not expect AI to autonomously touch the “mystical” mainframe code without strict human oversight. But mainframe data integration is critical. The mainframe must be able to speak to the rest of the enterprise infrastructure.
Second, do not shy away from experimenting with AI and agents in the mainframe estate. Status quo is not a strategy, even if the estate is 40 years old. Start with AI consumption: code understanding, dependency documentation, root cause analysis, and upgrade recommendations. Then move to AI configuration: observability reporting, alert creation, operational summaries, security patch recommendations, and compliance reporting. Over time, enterprises can explore orchestration, live risk assessment, modernization planning, and automated validation.
Third, the mainframe may not be the impenetrable fortress anymore. Mythos-class models are here and ready to ask tough questions. The smiles at the Broadcom event gave away nothing, but the signal is clear: enterprises need to understand what frontier AI means for mainframe security, vulnerability discovery, and remediation. Talk to partners involved in initiatives such as Project Glasswing.
The boring market just became interesting again
AI possibilities in mainframe are still undefined and emerging. I would not be surprised if we see digital twin environments for mainframes to safely experiment with AI-driven modernization, performance tuning, security testing, and operational simulation.
At the same time, mainframe is not just about tech. It is “legacy” attached with deep emotions among practitioners, and many believe it will continue to matter for at least another decade.
But what was different at the Broadcom analyst day was the openness. Almost everyone I spoke with is openly embracing AI and AI-enabled mainframe modernization. There was caution, but not defiance.
Next year might be a different story. Until then, there is a lot to understand, debate, question, and innovate.
If you enjoyed this blog, check out, Cisco Live 2026: The infrastructure for AI decision point is here – Everest Group Research Portal, which delves deeper into another topic relating to Broadcom.
To discuss more on this topic, contact Mukesh Ranjan ([email protected]).