On theCUBE Pod: AI doubters get loud, and IBM seeks competitive advantage

Americans are getting louder about AI’s downside, and the market felt it. Nvidia, CoreWeave, Oracle and Nebius all slipped last week after reports that OpenAI’s annualized net revenue was about $50 billion, which was roughly $20 billion lower than earlier numbers had suggested. But on theCUBE Pod, John Furrier called the backlash temporary. His view is that the real action is happening inside the data centers, where the stack itself is being rebuilt around AI.

That matters for IBM, because the company is not trying to win a beauty contest with flashy models. Dave Vellante said customers are spending on the infrastructure and models that create intelligence, then coming to IBM to make that intelligence useful. In his telling, the value sits in the data, the workflows and the operating model. IBM has been pitching a four-part setup built around agents, data, automation and hybrid infrastructure, with an orchestration layer that can pull in agents from different vendors.

This is IBM leaning hard on Red Hat and on the idea that models will become commodities. Furrier put it bluntly: if that’s the bet, IBM had better be right. The company does not need to invent its own proprietary operating system, he argued, because Red Hat already gives it an open abstraction layer that enterprises will pay to manage. The opportunity, at least in theory, is to become the place that runs all the agents while everyone else fights over the models.

Vellante’s caution was more pointed. Customers trust IBM, he said, but they do not currently see it as an AI innovator. So the company has to live in both worlds at once: trusted enterprise vendor and AI infrastructure layer. That is a narrow path, but a real one. And if IBM gets it right, it could own the boring but valuable part of AI that most of the market keeps pretending doesn’t matter.

The other big theme on the podcast was the shift from training to inference. Marvell Technology came up as a company with a long-range plan, a $70 billion to $90 billion revenue target for fiscal 2031, and a recent partnership with Nvidia’s NVLink Fusion. Furrier argued that Marvell saw early that AI was becoming a connectivity problem as much as a compute problem. Vellante added that workloads have already started tilting: a year ago, he said, 90% were training; now it is closer to 60-40, and the expectation is that it flips to 10-90 as inference becomes the money maker.