Snapdragon Insider bias right on the table: I have been tracking Qualcomm’s AI chip strategy for years, and the Snapdragon 8 Gen 4 is not a refresh. It is a strategic pivot toward becoming a full-stack AI platform rather than just a mobile SoC. The on-device AI performance, the custom Oryon CPU cores, and the integration of the Hexagon NPU into every layer of the processing pipeline signal that Qualcomm is building a chip that competes as much with Apple’s Neural Engine as it does with MediaTek’s Dimensity 9400.
The Oryon CPU and what it means
Qualcomm’s custom Oryon cores debuted in the Snapdragon X Elite for laptops earlier this year. The Snapdragon 8 Gen 4 brings that same custom architecture to phones, replacing the licensed Arm Cortex cores that powered every previous generation. This is Qualcomm designing its own CPU from scratch, the same strategic move Apple made with the A-series and Google made with Tensor. The performance gains are meaningful, but the real value is differentiation: Qualcomm no longer ships the same CPU cores as every other Android chipmaker.
Snapdragon Summit 2024 confirms what Qualcomm has been building toward: a unified silicon architecture across form factors. Phones get the 8 Gen 4. Laptops already run the X Elite and X Plus, which I covered when Computex launched the Copilot+ PC era in May. XR headsets run the XR2+ Gen 2. AR glasses like the Ray-Ban Meta carry the AR1 Gen 1. Industrial and edge deployments are covered by the DragonWing platform announced alongside today’s Summit. One company, one architecture, five form factors. That kind of platform breadth is something only Apple can match, and Apple does not license to other hardware makers.
On-device AI as a moat
The Hexagon NPU in the 8 Gen 4 delivers a significant jump in on-device AI performance: faster image processing, better natural language understanding running locally, and multimodal capabilities that do not require a cloud round-trip. This matters for privacy, latency, and cost. Every inference that runs on the phone instead of the cloud saves the user data, saves the developer money, and reduces the dependency on connectivity. The structural argument for on-device inference is not about convenience. It is about control over where your AI compute happens and who can reach it.
The European Commission’s move to block Apple’s on-device Siri AI features in July makes this point sharper than any benchmark. When regulators scrutinize even a local-processing AI implementation for compliance, the question of who controls your inference layer is not theoretical. On-device inference is not just a performance feature. It is an independence feature. The AI running on your Snapdragon chip cannot be revoked by an API policy change or an export control directive. Meta’s Llama 3.2 11B, released last month, runs on exactly this hardware and puts a real open-source multimodal model in the hands of anyone with a device carrying a sufficiently capable NPU.
The competitive picture
The 8 Gen 4 competes directly with Apple’s A18 Pro and Google’s Tensor G4. Apple has the vertical integration advantage: it controls the chip, the OS, and the frameworks. Google has the model integration advantage: Gemini is embedded in Android at the system level. Qualcomm’s advantage is reach: its chips power Samsung, OnePlus, Xiaomi, OPPO, and the vast majority of the premium Android ecosystem. No single Android OEM can match Apple’s integration. Qualcomm is the layer beneath all of them, which is a different kind of moat entirely.
On the laptop side, the X Elite competes against Apple M4, Intel Core Ultra 200, and AMD Ryzen AI 300 in a Copilot+ PC market that Qualcomm effectively created. The 8 Gen 4 is Qualcomm’s phone-side stronghold, and defending it requires the Hexagon NPU to keep pace with Apple’s Neural Engine across successive generations. The 45 TOPS figure announced today is competitive. It is not class-leading against the A18 Pro. The gap to close is the software stack: Core ML integration is still tighter than what Android OEMs deliver on top of the Hexagon SDK.
The Snapdragon 8 Gen 4 is the chip that ties Qualcomm’s entire 2024 strategy together. Custom Oryon CPU cores, a 45 TOPS Hexagon NPU that makes on-device AI a real product proposition, and a platform architecture extending from pocket to desk to headset to industrial edge. That breadth is the moat. Apple owns the premium. Google owns the model. Qualcomm owns the reach, and reach is what makes an AI chip ecosystem, not a single product.
Related reading: how Xiaomi factors into Qualcomm’s design-win roadmap.