Rethinking virtualization in the age of AI

While traditional VM hypervisors are CPU centric and assume steady demand, AI is memory hungry, GPU bound, and latency sensitive, he adds. “The abstraction layers add cost and drag, so there are big problems when models hit real scale,” Balabanskyy says. “This is a weird new world for folks whose job it is to plan for future usage.”

Beyond cost issues, virtualization and AI technologies intersect at the need for quality data, says Sune Baastrup, CIO at Danfoss, a heating, cooling, and data center component manufacturing firm. The company has recently reevaluated its virtualization technologies with a focus on data portability and flexibility, he adds.

“We see virtualization as much more than the traditional just getting more out of one physical box of hardware,” he says. “It’s actually preparing for a future where you have much more dynamic use cases, where we’ll see a lot of workloads being moved from either central places out into more decentral locations, or all the way out to the edge.”