Next-gen connectivity for new form factors: Moving beyond smartphones to uplink-heavy use-cases in devices like AR glasses that rely on "see what I see" for real-time streaming. Support AI-driven interfaces, like voice-first interactions, while enabling seamless collaboration between devices for edge and cloud processing.

Wide area sensing: Networks to evolve and move beyond data pipes, to intelligent sensing platforms that can offer a "bird’s eye" view for industrial and defence sectors to manage traffic, detect drones and more.  

Distributed compute: Creating a continuum of AI processing, from 10B parameter models on device to 10T parameter models in data centers, with localized "AI accelerator cards" in the middle of the network. Opens doors for mobile operators to evolve into becoming token providers, not just data providers.

In the end, its blending the Wireless and AI worlds to revolutionize with intelligent connectivity.

Building an AI-native Network

Durga emphasized that 6G must be "AI-native," where every node from the base station to the core should have AI inferencing capabilities.  

• Qualcomm's approach: To have heterogeneous compute architecture CPUs for RAN workloads, NPUs/XPUs for AI inference, all running together on a single server.
• The hardware shift: 

o   Devices → ~10B parameter models
o   Edge accelerator cards → 200B-300B parameters
o   Air-cooled servers → ~1 trillion
o   Liquid-cooled data center racks → ~10 trillion 

• 58 ecosystem partners have already signed on, spanning operators, infrastructure players, device makers, and server builders.
• 3GPP specs targeted for Q1 2029; commercial device launches expected from late 2029 onwards.

Democratizing AI Development – One Software Stack to Rule Them All

• Qualcomm's AI stack is built on openness and ease of use, shifting priority from serving OEMs to serving a broad developer community.
• Supports any framework: PyTorch, LiteRT, WinML, and more.
• Any model from Hugging Face can be tested on Qualcomm hardware with a single click.
• Hyperscalers and ISVs can plug into whichever layers they need, including full stack or selective.

Qualcomm is moving 6G from theoretical research into a concrete commercial path centred on energy-efficient, AI-driven infrastructure, unlocking monetisation beyond connectivity for operators, whereas developers benefit from an open, scalable AI ecosystem.

Analyst Take

• With 5G, Qualcomm took a leadership role in bringing out premium as well as affordable 5G and 5G-A capabilities at the edge with the ecosystem partners from smartphones to cars.
• However, with 6G, Qualcomm is taking an even wider and deeper approach to offer an “AI native” framework from cloud to edge more than a radio or compute design.
• The impact is going to be significant, redefining cellular-led AI inferencing at scale with maximum performance and high efficiency. 
• Building a strong but power-optimised foundation which is flexible, heterogeneous and distributed, will be critical.  
• Personally, I was fascinated with the wireless sensing demos and scenarios on unlocking network-scale use-cases and opportunities for the ecosystem.
• Qualcomm has taken the lead with 3GPP to define a 6G vision with a platform approach enable new classes of devices, services, and deployments to enable intelligent compute and connectivity.

It feels the 6G hype cycle is starting earlier but it is very pertinent for the ecosystem to get involved early to build an AI-native network architecture which is sensing, scalable, seamless and sustainable.