Qualcomm is designing a new architecture for its smartphone processors, one built specifically for on-device AI inference, according to a Semafor exclusive report. The move would mark the most significant change to the company’s chip designs since it introduced custom CPU cores, and it could reshape competition in the mobile chip market.
The shift in thinking is simple to state: instead of running AI workloads on general-purpose processors with an add-on accelerator, Qualcomm wants processors where AI is the primary design consideration. Every part of the chip — the cores, the memory system, the interconnect — would be shaped around the demands of running models on the device.
The technical rationale is well understood. AI models make heavy use of matrix math and demand large amounts of memory bandwidth, and they run most efficiently when the hardware matches the workload. A chip designed around those needs can deliver more performance per watt than a general-purpose chip retrofitted for AI.
The commercial stakes are large. Smartphone makers are racing to add AI features — assistants, translation, image editing — that run locally, and on-device performance has become a selling point. Qualcomm supplies most of the Android world’s premium chips, and it wants to keep that position as AI reshapes what phones do.
The architecture transition is risky. Qualcomm’s current designs, built on cores developed through its Nuvia acquisition, are recent and well-regarded. A new architecture means new software tooling, new validation, and new risk of the kind of missteps that have derailed chip transitions before — Intel’s repeated delays and Apple’s early silicon struggles being the standard cautionary tales.
The competitive picture is pushing the pace. Apple designs its own chips with a decade of on-device AI work behind it, and MediaTek has gained share with competitive mid-range designs. Both are investing in AI-specific silicon. Qualcomm needs an answer that keeps flagship Android phones on its platform.
Edge AI is the broader story. Running models on the device rather than in the cloud cuts latency, protects privacy and reduces the cost of constant cloud calls. Analysts said the trend is accelerating as models shrink and phones gain memory and compute — and the hardware that makes edge AI practical is becoming a competitive battleground.
The implications for the AI industry are significant. If phones can run capable models locally, the balance of AI computing shifts — from centralized data centers toward billions of devices at the edge. Chip designers, cloud providers and model makers all have a stake in how that shift plays out.
Qualcomm’s approach would also change how it works with Android partners. A new architecture means new opportunities for differentiation — and new demands on phone makers to tune software for the hardware. The relationship between Qualcomm and its customers could become closer, and more demanding.
Timing is the open question. Semafor’s report did not specify a release date, and people familiar with the company’s roadmap said the new architecture is still years from shipping. In the meantime, Qualcomm will keep improving its current designs, layering AI features onto the existing foundation.
Investors have been patient with Qualcomm’s AI story, but the payoff is not yet visible in results. The company’s AI revenue remains small relative to its core licensing and chip businesses, and the architecture bet is a bet on where the market is heading, not where it is.
For developers, a new architecture could mean new ways to build. On-device AI models are currently tuned to run on whatever hardware exists; a chip designed for AI would invite models designed around the hardware. The two sides of the industry could start co-evolving, the way Nvidia’s hardware and software have co-evolved in the data center.
The stakes extend beyond phones. Qualcomm has pushed its chip designs into laptops, cars and internet-of-things devices, and an AI-first architecture developed for phones would likely appear across those categories. The mobile design could become the foundation of a broader edge-AI platform.
Analysts said the move, if executed well, could be the most important product decision Qualcomm makes this decade. If it stumbles, rivals with faster AI roadmaps could seize the initiative. Either way, the direction is clear: the phone chip of the future is being designed around AI, and Qualcomm intends to be the company that defines it.
The report lands amid broader movement in the mobile industry. On-device AI has gone from demo to default feature in two years, and the hardware race has intensified accordingly. Qualcomm’s answer to that race — a clean-sheet architecture — is the most aggressive response yet from the Android ecosystem’s dominant chip supplier.
What the architecture looks like in practice will matter as much as its existence. The company’s engineers will decide how much of the design is new silicon and how much is software orchestration of existing components. Either way, the signal to the industry is unmistakable: bolting AI onto general-purpose chips is no longer enough.
