It's real. Matt Kimball, principal analyst at Moor Insights & Strategy, says that business and IT leaders want to talk to him about three things: “AI, AI, and AI.”
“I can’t have a conversation with a CIO, an IT leader, even business folks, without AI being the eventual topic of the discussion,” Kimball told The Forecast at the Nutanix .NEXT event in Chicago.
Invariably, people want to talk with Kimball about data readiness and governance, and then about how organizations can overcome supply chain constraints to access scarce compute resources like GPUs. But recently, more of these conversations veer into how organizations can best connect their data and compute environments for optimal AI performance.
AI bottlenecks follow a predictable path. Solve for compute, and storage steps forward. Solve for storage, and connectivity becomes the wall. Kimball says enterprises have to reckon with all of this.
“This is something not a lot of folks are focused on,” Kimball said. “But AI is like any other workload out there. As you progress through and evolve the workload into the enterprise, one bottleneck after another gets discovered and uncorked. And when you free up one bottleneck, you hit another.”
Kimball sees a clear signal in the shifting tenor of industry conversations.
“Last year, at .NEXT, we talked about how storage is ‘sexy’ again [storage reclaimed relevance as enterprises grappled with feeding GPU clusters]. I feel this year, and for the next couple of years, we’re going to be talking a lot about connectivity.”
Rethinking Connectivity
Implementing AI at scale, Kimball noted, isn’t like simply flipping a switch. Rather, it requires IT teams to address one chokepoint after another to keep performance in line with exploding demand.
“Connectivity is a huge source of performance constraint right now for a lot of AI workloads as they start to scale,” Kimball said.
Providing connectivity for AI is fundamentally different from traditional enterprise networking. Rather than simply deploying traditional switches and routers from large incumbent networking vendors, Kimball said, organizations will increasingly rely on gear outfitted with chips specifically designed for high-speed interconnects between critical AI infrastructure.
Kimball also pointed out that silicon photonics, a technology once largely confined to long-haul fiber networks, is now moving inside the data center itself, connecting chips and racks at the speeds demanded by AI clusters.
“When we think about connectivity, we think about traditional networking companies like Cisco and Juniper, who have been in the market for a long time and do enterprise networking,” Kimball said. “But let’s think about things a little bit differently with AI, because AI is not a few servers here and there, and connecting those to users.”
Instead, enterprises are building large compute clusters and trying to optimize the connectivity between GPUs, CPUs, racks and data centers.
“That interconnection that takes place at that level is something that enterprise organizations have never thought about before,” he said. “But they will think about how long it takes to get a response from ChatGPT for a user, which is directly impacted by that interconnectivity.”
The Whole Storage Equation Changes
AI also requires organizations to take a new approach to storage.
“Storage is no longer just a put-and-fetch kind of function,” Kimball dsaid. “Storage is being used to feed these GPU clusters that are training AI models or doing inference for the largest companies.”
That shift is also changing how storage vendors position themselves. Rather than emphasizing support for enterprise apps and databases, new players, particularly, are highlighting benefits such as data readiness and their ability to help companies make their data “AI-able.”
“With AI, the whole storage equation changes,” Kimball said. “It’s having a pull effect for traditional storage companies, too. They have all made major announcements about how they’re evolving their portfolio to be more focused on the data aspect, and not just storing bits and bytes.”
This messaging shift isn’t just marketing. Kimball said it reflects a real evolution of the role that storage plays in enterprise IT strategy.
“Storage is no longer just some passive place where you store your data and go get it,” he said. “It is an active part of the AI equation.”
Don’t Overlook the Control Layer
While high-profile players like NVIDIA and OpenAI tend to dominate the AI conversation, not enough attention is being paid to the IT realities of implementing the technology within large enterprises, Kimball said.
“We don’t think about how that enterprise company with 5,000 servers and 10,000 employees, how they’re going to activate AI across the organization,” he said. “You need that control plane or that operational model that allows an IT organization to take all that data and use it within their AI frameworks. And by the way, while you’re doing all that, you have to be able to support all these legacy apps that have been around forever and are still going to exist.”
For those enterprises, the complexity of AI adoption demands abstraction at scale, Kimball said.
"That's where Nutanix plays very, very well," he said.
His advice to IT leaders trying to optimize their infrastructure for AI is direct: resist the pull of legacy assumptions.
"Don't get lost in what you have," he said. "AI is wholly disruptive to your enterprise. Because of that, your infrastructure, your operating model, and the way you use and manage your data is also going to be wholly disruptive."
"Take a clean-sheet approach. Start with what that operating model and control plane looks like, and work out from there."
Related:
Can IT Infrastructures Meet AI Data Demands?
Memflation Wakes the Mother of Invention for IT Teams
AI Boom Jolts Pace of Business Amid Supply Crunch
The Fundamentals of Software-Defined Networking
From Cloud Native to AI-Native: The Evolution of Enterprise Infrastructure
AI Trends in 2026: Finding the Right Compute Platform for Every Workload
Editor's note: Learn more about Nutanix Enterprise AI capabilities, which helps IT teams connect, govern, and run AI across hybrid IT environments, with a centralized Agent Gateway and inferencing for agents and models at scale.
Calvin Hennick is a contributing writer. His work appears in BizTech, Engineering Inc., The Boston Globe Magazine and elsewhere. He is also the author of Once More to the Rodeo: A Memoir. Find him on LinkedIn.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.
Related Articles
