The race to build AI infrastructure systems has moved beyond chip specifications into a battle over entire rack-scale platforms, as inference and agentic workloads redefine what counts as a computer.

That shift is forcing challengers, once judged purely on GPU benchmarks, to prove they can ship complete, integrated systems spanning compute, memory, networking, and software. Advanced Micro Devices Inc.’s five-year sprint to compress what took Nvidia Corp. 15 years to build shows how urgent that transformation has become, according to Dave Vellante (pictured, right), co-CEO of SiliconANGLE Media Inc. and co-host of theCUBE.

“They spent $60 billion in M&A, $49 billion of which was on Xilinx, and they seeded the ecosystem with software like ROCm,” Vellante said. “They are now in the systems business. They’re no longer in just the chips business, and it’s a complete transformation and a reinvention. My premise is that they don’t have to take out Nvidia — they just have to become the essential second player, the second source in this AI wave.”

Vellante and John Furrier (left), co-founder and co-CEO of SiliconANGLE Media Inc. and co-host of theCUBE, held this conversation during theCUBE’s live coverage of the AMD Advancing AI event, an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They unpacked AMD’s shift from chipmaker to systems provider and how inference, agentic workloads and enterprise economics are reordering AI infrastructure priorities. (* Disclosure below.)

AI infrastructure systems shift from chips to rack-scale platforms

Beyond hardware, the conversation increasingly centers on routing workloads to the right resource rather than defaulting to the most expensive silicon available. Enterprises are learning that not every task needs a frontier model or a top-tier GPU, Furrier noted.

“You start to see the software intelligence coming in to support the end-user, bottoms-up experience,” Furrier said. “This model goes to that CPU. It’s not that expensive, it’s not a high priority, but we need to have a good enough answer. Don’t use the expensive GPU for that — go to the right AI resource.”

AMD’s answer to that demand has been consolidating disparate acquisitions into a single systems narrative. Xilinx brought adaptive computing via field-programmable gate arrays, or FPGAs, while Pensando added DPUs and networking expertise as a critical part of the portfolio. That buildout has coincided with a broader climb in the data center business, as AMD’s rapid climb has written a new chapter of success that tech companies with its kind of longevity seldom see.

Still, Furrier cautioned that AMD and Nvidia are pursuing different philosophies even as both race toward rack-scale integration — one favoring tight, proprietary co-design and the other relying on openness and chiplets to win customers wary of lock-in.

“The winner will be whoever can, in the fastest way possible, redesign the entire computing experience around how humans actually think and work,” Furrier said. “Whoever gets there first, no one will care how it’s done.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of the AMD Advancing AI event:

(* Disclosure: TheCUBE is a paid media partner for the AMD Advancing AI event. Neither AMD, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE