AI infrastructure strategy moves beyond VMs to distributed data and inference, vendors say

At VMware Explore, Cody Hosterman, Everpure Inc.’s senior director of product management, said conversations with customers had shifted away from VMs. “The conversations I’ve been having in our meeting room and at our booth and in the hallways have not once been about VMs,” Hosterman said. “It’s actually been talking about the datasets and the applications that are consuming them with people who are traditionally infrastructure administrators.” He added: “So the conversation has fully changed.”

Everpure is aligning its data platform with VMware Cloud Foundation 9.1 to support VMs, Kubernetes and AI workloads, Hosterman said. Customers are using AWS and Microsoft Azure public-cloud resources to test configurations before right-sizing infrastructure in their data centers, he said, adding: “We can right-size this with best-of-breed infrastructure in the data center.” The costliest error is treating AI enablement as an end in itself; the correct starting point, he said, is a business destination with AI as the tool to reach it.

At Horizon, Equinix announced Equinix Fabric One, an intent-driven, managed any-to-any connectivity service, and Equinix Inference Exchange, a distributed inference offering built with Nvidia Corp. and Together AI Inc. Chief product officer Chris Audie said enterprise networks cannot be built one connection at a time in a distributed-AI world. “That approach doesn’t scale in a world of distributed AI that demands dynamic, flexible and real-time connectivity,” Audie said. Fabric One will use open connectivity specifications developed with Amazon Web Services Inc. and Google Cloud, its lead integration partners; beta is expected later this year and general availability in 2027, initially in North America.

Cisco Systems Inc. also featured the edge as an AI compute platform. Its Unified Edge, which won the 2026 Tech Innovation CUBEd Award for most innovative IoT or edge platform, is designed for real-time AI inference at edge locations and supports CPUs, GPUs, up to 120 TB of storage, redundant power and cooling and integrated 25-gigabit networking. James Leach, Cisco’s director of product management, said workloads are moving from centralized clouds toward where data is generated. “If you think of data as the new oil, as they say, we have to go and process it and we have to refine it as close to the mining of it as possible,” Leach said. theCUBE Research chief analyst Dave Vellante added that if AI agents increase workflow traffic and tool calls, “the network becomes a multiplier, not just background plumbing.”

Jim McGregor, founder and principal analyst at Tirias Research, said inference workloads are far from uniform. “We tend to think of AI as a single workload, and it’s not. It’s thousands, it’s millions, it’s billions of different workloads,” he said. McGregor said organizations must optimize the entire network, including memory and storage, around actual workloads; AI inference depends on data movement, caching, storage proximity and low-latency retrieval.

Hybrid cloud decisions also reflected workload-specific thinking at VMware Explore. Rob Carter, chief executive of Lightedge Solutions Inc., said many customers are in different phases because of cloud-first decisions and acquisitions, making placement choices about the right fit rather than a universal model. “Either you do it yourself and you have to build that bench. Storage engineers, virtualization engineers, network — there’s so much that goes into it,” Carter said. “We will handle that layer for you.” Lightedge is one of 14 U.S.-based VMware Cloud Service Provider Pinnacle partners; Carter stressed the provider should not push one platform when another is better for a business. “Where do you need to be, and [how do we] put you in the right place versus jamming you into something just because that’s what I have to sell?” he said. “That’s where companies get in a bad spot.”