Qualcomm’s New Chip Promises AI Agents in Your PC

Qualcomm’s Snapdragon Summit, its annual conference held in Maui, Hawaii, took the wraps off the new generation of processors for PCs, Snapdragon X2.

First introduced last year as part of Microsoft’s high-profile Copilot AI PC push, the processor lineup was notable because its CPU cores use the Arm instruction set instead of x86. That initial Snapdragon X rollout was targeted at smaller and more portable Windows PCs. But for its next move, Qualcomm is looking to expand its reach with a third-generation CPU and an upgraded neural processing unit (NPU), a one-two punch meant to accelerate AI workloads in mini PCs, all-in-one desktops, and high-performance laptops.

“We’re going to bring AI everywhere,” says Qualcomm CEO Cristiano Amon. “Everything we’ve been talking about is starting to happen, and this dream of what’s going to happen with AI, it’s getting to the point where we see what’s going to happen at scale.”

Keeping up with the Apples

Oryon, Qualcomm’s CPU core in Snapdragon X, repeated a trick Apple pulled off with its introduction of Arm-based silicon in 2020. (The team that designed Oryon included a number of CPU architects who worked on Apple’s chips.) The first generation of the architecture placed an emphasis on efficiency and multicore performance. It also integrated an NPU, which is why it led Microsoft’s AI-focused Copilot Plus PCs. The third-gen Oryon architecture inside the Qualcomm Snapdragon X2 upped the core count from 12 to 18 for the top-end version, the Snapdragon X2 Elite Extreme. (The second generation was released only for smartphones.) But in a move that mirrors Intel and Apple’s architectures, it now combines two different kinds of CPU cores. (Qualcomm’s mobile CPUs also have two types of cores.)

Unlike Apple and Intel, none of the cores in Snapdragon X2 are designed for low-performance workloads, Qualcomm stressed. Snapdragon X2 will pair what it calls “Performance” cores having multicore clock speeds up to 3.6 gigahertz (similar to those of the prior Snapdragon X) with new “Prime” cores that can reach multicore clock speeds up to 4.4 GHz, or up to 5 GHz in dual-core workloads.

Why? Once again, it comes down to Apple.

The world’s other major player in Arm chips for personal computing outpaced Qualcomm with the Apple M4, which achieves clock speeds up to 4.5 GHz and, in its 16-core top-tier M4 Max configuration. That left Snapdragon X a step behind in both single-core and multicore performance.

Qualcomm believes that Snapdragon X2’s high clock speeds will give it the lead, though whether that’s the case remains to be seen. The first PCs with Snapdragon X2 aren’t expected until the first half of 2026, setting the stage for a showdown with Apple’s M5, which is expected to arrive in late 2025 or early 2026.

More memory for more AI

While Qualcomm and Apple butt heads on CPU performance, there’s another aspect of chip architecture where Qualcomm is ahead—the NPU.

Competitors also have NPUs, which accelerate AI workloads, in their chips. But Qualcomm has staked the most on NPU performance. The first-gen Snapdragon X was quoted a NPU performance at 45 trillion operations per second (TOPS), and it claimed that for all chips in the product stack, not just the most expensive silicon. Snapdragon X2 will boost that metric to 80 TOPS.

Snapdragon X2 also provides memory upgrades to improve AI workloads. The line of processors can support up to 128 gigabytes of advanced, low-power DRAM memory, LPDDR5x RAM. This is an upgrade from Snapdragon X, which topped out at 64 GB, and similar to AI-focused chips like AMD’s Ryzen AI Max. Memory bandwidth is also boosted from 135 GB/s to a maximum of 228 GB/s in the Snapdragon X2 Elite Extreme.

Both metrics are important to generative-AI workloads. More memory makes it possible to load larger and more intelligent models, while added memory bandwidth improves the speed at which models can generate a response to a user’s prompt.

A lot to offer, a lot to prove

Qualcomm’s CEO speculates that personal computers will see a grand shift toward AI workloads. “As the AI can understand what we say, what we see, what we write…that becomes the new [user interface] of computers. The UI is human-centric and it gets processed where you are,” says Amon.

Qualcomm’s NPU theoretically makes Snapdragon X2 a great choice for AI software. But it still has a lot to prove.

While Amon’s keynote address in Maui imagined a future with numerous AI agents working on-device, the current reality is more modest. Qualcomm has Microsoft’s support and has a few software wins in specific features, like the AI-powered “magic mask” that removes unwanted objects from video in DaVinci Resolve Pro, a popular professional video editor.

But such features remain exceptions to the rule, and Amon’s examples of how agentic AI might work were mostly hypothetical. An agentic AI assistant running on Qualcomm’s Snapdragon might one day be able to handle your calendar or pay your bills—but such an on-device agent doesn’t exist yet.

The idea, then, appears to be “If you build it, they will come.” Qualcomm’s Oryon third-gen architecture and Snapdragon X2 chip imagines a future of personal computing with a focus on CPU and AI performance. But the company is still waiting to see if developers want to play ball.