At a time when AI is moving rapidly from centralized systems into devices, networks, vehicles, factories and the physical world, semiconductor strategy is taking on a much broader role. It is no longer only about increasing performance or shrinking power consumption in isolation. It is about enabling a new architecture of intelligence that stretches from the edge to the data center and back again. 

Speaking with Meenakshi Benjwal, Global Industry Marketing Head for TMT Industries at HCLTech, Dr. Durga Malladi, EVP & GM, Technology Planning, Edge Solutions and Data Center at Qualcomm Technologies, described how this shift is changing the way leaders need to think about AI, connectivity and the future of computing. 

AI is now mainstream across the technology stack 

For Malladi, the first point is that AI has already crossed into the mainstream. “AI is now a pretty mainstream technology that’s permeating every sector of the industry,” he said.

That changes the strategic conversation. AI is no longer confined to a single layer of the technology ecosystem, nor is it shaped by a single market. The demands of AI now stretch across devices, industrial systems, automotive platforms and data centers, each with very different requirements. 

Malladi explained that Qualcomm’s move into the data center market last year completed a broader arc “which takes us all the way from devices to the data center.” But the way AI works across those domains is far from uniform. On edge devices such as smartphones, PCs and XR systems, the emphasis is on inference using sophisticated but relatively compact models. In industrial and automotive settings, model sizes and use cases are becoming increasingly demanding. At the data center end of the spectrum, the industry is dealing with foundational models running into the trillion-parameter range, with very different power and infrastructure profiles. 

That is why a new operating model for AI is beginning to take shape. “The main conversation for us, from Qualcomm’s perspective, and I think it’s resonating with the rest of the industry, is how do you tap into that, upgrade that network and bring in a true hybrid AI model, wherein we have inference that goes everywhere, from the edge to the data center.” 

Why hybrid AI changes the consumer experience 

For end users, this is not primarily a story about architectures or model sizes. It is about outcomes.

As Benjwal pointed out during the conversation, consumers care far more about what intelligence enables than how it is delivered.  

Malladi agreed: “An average consumer probably doesn’t know and doesn’t care that much as to how AI inference is occurring, as long as they see the benefits of it.” 

That is where hybrid AI becomes important. In high-quality connectivity environments, some processing can happen close to the device and some deeper in the network or data center, without the user noticing the handoff. The result is an experience that feels fast, responsive and seamless, even when compute is dynamically distributed. 

This is especially relevant for emerging use cases such as Physical AI, low-latency edge applications and AI-enabled glasses. In these environments, some inference can happen on the device itself, while other parts may need to rely on systems at the edge of the network. Making that work smoothly depends on more than compute alone. It requires a connectivity fabric and a set of protocols that can decide what runs where in real time, without creating visible friction for the user. 

Why 6G is about more than faster connectivity 

This is also why Malladi sees the next network era differently. “This is the first time that as Qualcomm, we talked about 6G,” he said. But, unlike earlier generational shifts, “this time, it’s very different.” 

He outlined three foundational pillars for 6G.  

1. The first is still connectivity, but with new attributes shaped by AI-driven workloads.  
2. The second is wide-area sensing, where the network itself develops a broader view of its environment and can support entirely new capabilities, from drone detection to traffic and infrastructure intelligence.  
3. The third is compute inside the network itself. 

In other words, 6G is not just another upgrade in speed. It is part of a wider shift from connectivity to distributed intelligence. Radios, network cards, servers and racks begin to form part of a much broader AI continuum. 

It is “a journey from connectivity to intelligence,” said Benjwal. 

What boards and CEOs should be doing now 

For enterprise leaders, Malladi said that over the next 12 to 24 months, the right move depends heavily on an organization’s existing position in the value chain. 

“If you’re already a sovereign data center provider, for instance, and you’re also a network provider, it’s a natural stepping stone to start thinking of the assets that go into the data center” and how those capabilities can move closer to the user through the network, he said. 

Infrastructure vendors need to think differently too. The right processor architecture for network workloads may not be the same one used in the data center. Device manufacturers face another challenge: identifying the new product categories that AI may unlock.  

For Malladi, XR glasses, Physical AI and robotics all point to the emergence of new classes of connected devices that will complement rather than replace smartphones and PCs. 

And for ecosystem players such as memory providers, the shift is equally significant. Malladi noted that AI inference increasingly depends on different stages of compute and memory performance. Some workloads are dominated by raw compute, while others become bottlenecked by memory bandwidth. That means the memory architecture itself must evolve alongside the rest of the AI stack. 

His broader point was that every part of the ecosystem now must rethink its role in the next three to four years. 

The next era of tech and semis 

When asked to summarize where the next era of technology and semiconductors is heading, Malladi was clear on the direction of travel. 

“Energy efficiency and high-performance computing is here to stay. There is no going back,” he said. 

At the same time, he sees a more visible shift in how people will interact with technology itself.  

“Voice and AI [are] becoming the more natural user interface to all the devices around us, with Agentic AI that's permeating everything,” he said. In practical terms, that means a world in which people increasingly talk to their devices rather than relying on touchscreens and keyboards as the default interface. 

He also pointed to the rapid rise of Agentic AI as an immediate reality, not a future concept. “AI agents, they’re real now,” he said. Combined with high-performance computing, new memory architectures, stronger connectivity and wider sensing, they are helping define the next wave of intelligent systems. 

From performance metrics to intelligent systems 

The discussion made one broader point especially clear. Semiconductor innovation is no longer being measured only by traditional hardware metrics. Performance and power still matter, but they now sit inside a much larger question: how do you build the infrastructure for AI that is distributed, efficient and always connected? 

That is what makes this moment significant. As AI expands from the cloud into devices, networks and the physical world, semiconductor strategy becomes inseparable from business strategy, platform design and the user experience itself. 

For leaders across telecom, computing, infrastructure and manufacturing, the next era will not be defined by raw speed alone. It will be shaped by how well they connect intelligence across the whole stack and how effectively they turn that into real-world value.