On theCUBE Pod: AMD nips at Nvidia’s heels, and IBM undergoes quiet transformation

Last week's AMD Advancing AI event in San Francisco wasn't really about a new chip. That's the take from John Furrier and Dave Vellante of theCUBE Research, who spent the week talking to executives and came away convinced the industry has stopped asking who has the fastest GPU. The real question, they argue, is who can assemble an entire AI system — chips, software, orchestration, the works — into something enterprises can actually run their businesses on.

That reframing matters for AMD specifically. The company has quietly moved from being a chip vendor chasing Nvidia's benchmarks to something closer to a systems integrator, stitching CPUs, GPUs and software agents into what Furrier calls an AI factory. Vellante thinks the bigger prize is what gets built on top of that factory: a software layer that actually monetizes all the capital expenditure pouring into data centers. Every infrastructure boom eventually needs an application layer to justify the spending, he points out, and this one won't be different.

Nvidia still owns the top spot, and nobody on the podcast pretended otherwise. But Vellante's argument is that AMD doesn't need to dethrone Nvidia to win. It needs to be the credible number two — the vendor enterprises turn to when they want pricing leverage, supply chain redundancy, or simply an alternative to putting all their AI infrastructure eggs in one basket. That's a lower bar than total dominance, and arguably a more durable business than chasing a lead that may never materialize.

Meanwhile IBM is getting hammered by Wall Street even as it makes what Furrier considers a defensible long-term bet. Instead of chasing frontier models, Big Blue is positioning itself as the operating layer that helps large enterprises bolt AI onto decades-old systems securely. It's less glamorous than building the next GPT, but it's also a market that isn't going anywhere — most companies aren't ripping out legacy infrastructure any time soon, they're trying to make it talk to AI without breaking everything else.