TECHnalysis Research Blogs
TECHnalysis Research president Bob O'Donnell publishes commentary on current tech industry trends every week at LinkedIn.com in the TECHnalysis Research Insights Newsletter and those blog entries are reposted here as well. In addition, those columns are also reprinted on Techspot and SeekingAlpha.

He also writes a regular column in the Tech section of USAToday.com and those columns are posted here. Some of the USAToday columns are also published on partner sites, such as MSN.

He also writes a 5G-focused column for Forbes that can be found here and that is archived here.

In addition, he also occasionally writes guest columns in various publications, including RCR Wireless, Fast Company and engadget. Those columns are reprinted here.

September 24, 2025
Qualcomm Focuses on Agentic AI with Latest Chips

By Bob O'Donnell

If you’re paying close attention, you can’t help but notice a key theme that’s started to weave its way into and through the messaging from today’s top tech companies. That theme? Hybrid AI—or a combination of running AI models, agents, and applications both in the cloud and on the edge.

At the company’s annual Snapdragon Summit, hosted in the amazing environs of Maui (to which they graciously covered my travel expenses), Qualcomm made it clear that they are doubling down on this strategy. The key product introductions focused on their newest Snapdragon 8 Elite Gen 5 mobile SOCs for smartphones and the Snapdragon X2 Elite and X2 Elite Extreme for PCs, but the overall story encompassed a much broader scope. From cars and smart glasses to headphones and IoT devices, the story is about driving an interconnected set of devices that leverage AI to deliver a highly personalized set of information and services to each individual.

Admittedly, this kind of big picture vision of multiple connected devices working together has been talked about a lot before, both by Qualcomm and other major tech players. But what was interesting this year, as CEO Cristiano Amon pointed out in his opening keynote, is that the arrival of AI-powered agents is starting to bring this somewhat fuzzy future perspective into much sharper focus. We’re not quite there yet, but the days of AI-powered agents running on devices that can quickly, seamlessly, and intelligently tap into a wide number of MCP (Model Context Protocol)-enabled models that exist on the device, in the cloud, and even in corporate data centers is getting tantalizingly close.

In this type of ideal scenario, the hardware (and software) requirements of devices start to shift, and Qualcomm reflected some of these changes in its latest offerings, particularly on the PC side. Most importantly, in the hybrid AI world, connectivity becomes as important as compute. Yes, of course, we will always need faster processors and a wider variety of compute options (I’m looking at you, NPUs) to do things like increase the rate that tokens can be created for future AI applications. However, the role of things like 5G connectivity becomes essential when running agents that at any point in time may need to access models that continue to run outside the device.

With that context in mind, it’s worth digging into some of the key takeaways from this year’s announcements. First, at a top level, it’s important to note that, for the first time, Qualcomm is bringing its latest generation Oryon CPU cores (now in Gen 3) to both the mobile and PC platforms. For smartphones, this is simply the next step, but last year’s X Elite PC processors used Gen 1 of Oryon, so it marks essentially a jump forward for next generation PC parts. Not surprisingly, that two generation leap is reflected in some of the early performance claims that Qualcomm is making for the X2 Elite platform. According to the company, when PCs using the new chips are released in early 2026, they will have an overall 31% improvement in performance and a 43% decrease in power demands.

More than just simply improving the speed (up to 5 GHz on the X2 Elite Extreme) and number of cores (up to 18 on the X2 Elite), the company increased the size of the cache memory for the CPU clusters on its PC chips to 53 MB. In addition, it raised the bandwidth of the memory connection, enabling up to 228 GBps on the X2 Elite Extreme. This improved bandwidth makes a huge difference when running many AI models, which need extremely fast access memory to perform well.

The company also made important improvements in its Adreno GPU, enabling enhanced gaming performance for both mobile and PC. Qualcomm now offers a sliced architecture that enables up to three simultaneous graphics threads to be executed at the same time. It also added support for what the company is calling HPM (High Performance Memory), which is a dedicated high-speed cache that reduces the latency that can occur when GPUs access memory. The net result of all these changes Qualcomm claims is a 23% increase in graphics performance.

For NPUs, the improvements also vary by platform. For mobile, Qualcomm disclosed a 37% improvement in performance but doesn’t offer a spec for TOPs. On the X2 Elite PC chips, however, the performance levels have now been raised to 80 TOPs, a near doubling of last year’s 45 TOPs. Of course, the challenge with NPUs, particularly on the PC side, is that very few software vendors have chosen to use them up until now because of the difficulties in supporting multiple different NP architectures from different vendors. As a result, much of the NPU performance capabilities have essentially gone to waste.

However, in a major step forward, Microsoft announced the general availability of its Windows ML framework earlier this week. Essentially what Windows ML does is provide a standardized way for software developers to leverage the capabilities of NPUs, without having to worry about the underlying differences in NPU architecture. By leveraging what are called Execution Providers (EPs), Windows ML abstracts away the hardware differences not only of different NPUs but even GPUs and CPUs, allowing software to get the most out of any different configuration of PC hardware. In simple terms, it’s similar to providing the equivalent of DirectX for NPUs. (DirectX allows software programmers to write to GPUs in general and then “translates” any differences between the GPU architectures of companies such as AMD, Intel, and Nvidia.) Windows ML is literally game changing in terms of the potential possibilities that it should enable, though realistically, it will be a while before we see a reasonable number of applications that take advantage of this architecture.

On the mobile side, the ISP (Image Signal Processor) also plays a critical role for photo and video applications. The latest version offers 20-bit operation, which enables better low-light photography. Interestingly, software developer ArcSoft created Dragon Fusion, which combines the ISP and the NPU on the 8 Elite Gen 5 for a computational video pipeline on smartphones, allowing every frame in a video to be AI processed in real time.

Finally, on the connectivity side, Qualcomm upgraded the mobile SOCs to support its MobileConnect 7900 WiFi and Bluetooth chips and the new X85 5G modem architecture that supports both GSMA Release 17 and 18 capabilities. For the PC SOCs, the company added support for the X75 modem architecture. While that doesn’t sound like a big deal, explicitly calling out 5G support for this next generation PC architecture should hopefully encourage the creation of more 5G-equipped PCs, all of which will be better equipped for a world of hybrid AI.

In addition, for the entire X2 Elite family of chips, Qualcomm unveiled what it’s calling Snapdragon Guardian technology. What it does is leverage a built-in low-power modem that can be used to automatically locate, update, manage or wipe a PC in the event it’s lost or stolen. More than just a “Find My” feature, the Snapdragon Guardian technology features a dedicated processor that can be remotely turned on, even if the PC is off or won’t boot. For enterprises, this is a hugely important new capability, and it could even prove interesting to consumers who, for example, want to track or manage a child’s PC or for other similar applications.

Beyond the new hardware capabilities, Qualcomm also highlighted its role in helping to enable new agentic AI experiences in combination with software vendors and OS providers, most notably Microsoft and Google. Ultimately, it’s the combination of these new chip-based advances from Qualcomm along with the critical software development from these big players that will lead to the kind of futuristic agent-driven interactions that Qualcomm expects to be part of its (and our) future. When that future vision will actually arrive still isn’t entirely clear, but it’s definitely approaching sooner than many initially expected.

Here's a link to the original column: https://www.linkedin.com/pulse/qualcomm-focuses-agentic-ai-latest-chips-bob-o-donnell-tqkfc

Bob O’Donnell is the president and chief analyst of TECHnalysis Research, LLC a market research firm that provides strategic consulting and market research services to the technology industry and professional financial community. You can follow him on LinkedIn at Bob O’Donnell or on Twitter @bobodtech.