Modern artificial intelligence’s requirements highlight a significant gap between the locations where intelligence must be applied and the places where existing infrastructure was originally designed to support it. As inference workloads multiply and agentic systems require tighter real-time controls, edge AI is emerging not as a niche application but as a foundational architectural shift — one that is simultaneously rewriting the economics of infrastructure investment, network design and enterprise security.

Those forces were front and center at MWC Barcelona 2026, where theCUBE’s coverage captured exclusive conversations with executives, operators and analysts navigating the collision of AI ambition and infrastructure reality. As inference demands grow and agentic AI frameworks push the boundaries of what networks weren’t designed to support, the architectural gap between today’s infrastructure and tomorrow’s edge AI requirements is widening, according to Sarbjeet Johal, founder and chief executive officer of Stackpane Ltd.

“The inference is compute-hungry; there’s no doubt about that,” he told theCUBE. “It will be even more compute-hungry going forward because we are throwing policy at these agentic frameworks. We have to reduce the hallucinations. We have to program our systems of record using the AI constructs, and for that, we need more control over the flow of logic.”

During the event, Johal and other industry experts joined theCUBE Research’s John Furrier and Dave Vellante to provide exclusive commentary about the barriers to edge AI adoption, the market opportunity in enterprise wireless, the transition from centralized data center AI to distributed real-time deployments, the convergence of networking and data infrastructure reshaping enterprise architecture, and the push toward Open Radio Access Network as a path to sovereign digital strategies and edge modernization. (* Disclosure below.)

Here’s theCUBE’s complete video interview with Sarbjeet Johal:

Insight #1: Executive leaders see AI pressure building at every layer of infrastructure, from power-constrained data centers to real-time edge AI environments.

The buildout of AI infrastructure is entering a phase that looks less like an upgrade cycle and more like a foundational rearchitecting of how compute, networking and power intersect. With data center investment projections approaching $5 trillion and agentic AI placing continuous demands on global networks, the constraints shaping this transformation are no longer theoretical, according to Jeetu Patel, president and chief product officer of Cisco Systems Inc.

“My take on this is in the long term. If you think about this as a seven-to-10-year window, we are grossly underestimating the capacity required to fulfill the needs of AI — underestimating, not overestimating,” Patel told theCUBE. “We are still trailing the demand with supply. We are supply-constrained. Every single bit of token-generation capacity that’s being sold is getting consumed right away. This is not something that you have to grow into the demand side. The demand is already there.”

The enterprise wireless edge is becoming a battleground as 5G fixed wireless access shifts from a niche connectivity option to primary business infrastructure. That transition is accelerating across industries — from office environments to factories and field operations — creating pressure on vendors to serve a much wider range of enterprise deployments than early fixed wireless access solutions were designed to reach, according to Juho Sarvikas, CEO of Inseego Corp.

“One thing that we take great pride in is that we have SMB or micro-enterprise ease of use, but then the power of a full enterprise solution,” Sarvikas told theCUBE. “I see a lot of deployments, not only in the carpet environments as we call them, but also in industrial verticals. Definitely the uptake of 5G is advancing — even more so with [reduced capability].”

Telecom operators are navigating a structural tension between the urgency of AI adoption and the weight of installed infrastructure built for a different era. Legacy server sprawl — much of it two or three generations behind current technology — is compounding power and space pressures just as AI workloads accelerate, according to Derek Dicker, corporate vice president of the Enterprise Business Group at Advanced Micro Devices Inc.

“I think that the single biggest thing that they’re focusing on right now is how do they take their existing infrastructure … and make sure that they modernize it predominantly for the purposes of preserving the performance and then lowering the power,” Dicker said during the event. “Everything that I hear when we talk to customers is [around], ‘How can we actually take the power structure of these solutions and make them much more efficient?’”

Edge AI is no longer a data center concept stretched thin across a network — it’s a fundamentally different computational problem. Where cloud AI generates responses, edge AI must act within fixed time windows, and in safety-critical environments, missing that window isn’t a performance issue — it’s a failure, according to Paul Miller, chief technology officer of Wind River Systems Inc. The company’s real-time OS is already operating at scale across major carrier networks globally, a footprint that reflects how quickly that demand is materializing.

“I need to perform some computational problem within a fixed period of time, or it’s not going to be sufficient for the application,” Miller told theCUBE. “That requires determination and high performance. That determinism and high performance can only be realized if running on a real-time operating system.”

Here’s theCUBE’s complete video interview with Jeetu Patel:

Insight #2: Analysts see a revenue opportunity in telco AI — and a set of structural barriers that could keep most operators from reaching it.

Telecom operators have weathered major technology transitions before — from circuit switching to broadband, from 3G to 5G — and historically the pattern has been the same: Infrastructure investment rises, connectivity improves and the value accrues elsewhere. The shift to software-defined, AI-enabled networks has revived the question of whether this cycle will finally break that pattern, according to Zeus Kerravala, founder and principal analyst at ZK Research LLC.

“I don’t think there’s any question that AI is going to help [telco companies] lower their operational costs,” he said during the event. “I think agentic ops is coming, and that’s going to make them a lot more efficient. But what I’ve been looking for is can somebody answer the question — once AI is in the network, do the telcos actually have an opportunity to raise their value [proposition] and create new sources of revenue?”

Even for operators motivated to pursue that opportunity, the path is narrowed by structural forces that predate the current AI wave, according to Johal. AT&T Inc.’s $14 billion Open RAN commitment with Ericsson illustrates that reshaping infrastructure at scale requires capital and geographic reach that most operators don’t have. 

“Most of the telcos serve locally; only [a] few telcos go across borders. For that reason, their scale is small,” Johal told theCUBE. “They’re highly regulated by the government, local governments … [on] what you can do and what you can’t do. That keeps them at bay from innovation. Scale is the best friend of innovation. If you have scale, you can do economies of scale, but also division of labor — you can divide and conquer.”

Here’s theCUBE’s complete video interview with Zeus Kerravala:

Insight #3: For IBM, AI-era security has two fronts — sovereign infrastructure for telcos and quantum-safe foundations for every enterprise.

Telecom operators occupy a structural position that hyperscalers can’t replicate — physical proximity to end users, years of experience managing sovereign networks and an established presence at the network edge. As AI shifts from internal cost optimization toward revenue-generating services, that position is opening a new competitive lane: becoming the trusted infrastructure partner for enterprises that need governed, sovereign-grade AI capabilities but lack the resources to build and manage them independently, according to Eoin Coughlan, global CTO and industry lead for telecommunications, media and entertainment at IBM Corp., and Fran Heeran, VP of global telecommunications at Red Hat Inc.

“[Telcos] should provide the AI factory and they should give them the tools to govern, to build their own capabilities, to bring in their own models,” Coughlan said during the event. “That’s what we’re talking to customers about right now is, ‘How does the telco manage that for all of the enterprise?'”

While telcos weigh their sovereign infrastructure opportunity, a different but equally pressing security challenge is already building across the enterprise. The threat of “store now, decrypt later” attacks — where encrypted data harvested today is held until quantum computers can break it — means that organizations waiting for quantum computing to mature before acting may already be behind, according to Mark Hughes (pictured), global managing partner of cybersecurity services at IBM.

“The first thing to do is establish, as an executive, what cryptography you actually have running in your organization,” Hughes told theCUBE. “You have to go through the environment and understand where it is, whether it’s embedded in hardware or software. The initial phase, fairly straightforward, is just a straight discovery exercise.”

Here’s theCUBE’s complete video interview with Eoin Coughlan and Fran Heeran:

Insight #4: Real-world AI deployments show that outcomes depend as much on infrastructure decisions as on the intelligence itself.

A city once ranked among the least connected in the country, Brownsville, Texas, has built its own in-city AI factory — converting a dense network of sensors into a real-time operational platform for public safety and city services. The transformation illustrates a broader economic reality facing edge AI deployments: Actionable intelligence requires bringing compute to where data is generated, not routing it back to infrastructure that wasn’t designed for distributed, low-cost environments, according to Shahid Ahmed, global head of edge services at NTT Data Inc., and Jorge Cardenas, chief information officer of the city of Brownsville, Texas.

“We have sensors in the area that … will predict how and when something’s going to happen, and [the police] can react to it,” Cardenas said during the event. “That’s the level of how beneficial this technology is for us.”

Quantum-safe networking is emerging as the next architectural dividing line for enterprises scaling AI into core infrastructure, where governance, orchestration and data sovereignty are becoming competitive differentiators rather than compliance checkboxes. Telefónica S.A., one of the world’s largest telecommunications operators, works with IBM to advance both AI deployment and quantum-safe technology as part of a broader push to modernize its network while maintaining service integrity. For operators already running AI agents on live infrastructure, the most mature deployments share a common trait: Human oversight remains in the loop while systems prove their reliability, according to Gustavo Carvalho Domingos, director of customer operation and technology at Telefónica, and Luq Niazi, global managing partner of industries at IBM.

“We have the operation agent already working with our technicians,” Domingos told theCUBE. “We already have … [the] knowledge database developed that helps … make the troubleshooting that we need before implementing any solution to recover the service.”

Here’s theCUBE’s complete video interview with Gustavo Carvalho Domingos and Luq Niazi:

For more of theCUBE’s coverage of MWC Barcelona, check out these segments:

• Jeff Aaron, VP of product and solutions marketing at Hewlett Packard Enterprise Co.
• Jason Kelley, global head and managing partner of core business applications at IBM Consulting, a division of IBM
• Chris Lewis, managing director of Lewis Insight Ltd.
• Mark Austin, VP of data science and AI at AT&T, and Molham Aref, CEO of RelationalAI Inc.

To watch more of theCUBE’s coverage of MWC Barcelona, here’s our complete video playlist:

(* Disclosure: TheCUBE is a paid media partner for MWC Barcelona. Sponsors of theCUBE’s event coverage do not have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE