Qualcomm's stock price fell over 7% in late trading on Monday, but the AI growth outlook presented by CEO Cristiano Amon in his Computex 2026 opening keynote offered medium to long-term optimism for the market. Amon forecast that by 2030, AI token generation will surge from the current 31.7 billion per 10 seconds to approximately 12.7 trillion per 10 seconds, representing a nearly 40-fold increase.
Amon clearly designated 2026 as the "Year of the Agent" in his speech, marking AI's evolution from a tool that passively responds to commands into an intelligent partner capable of autonomous planning, reasoning, and task execution. Amon noted: "Today's devices are not designed for agents. Existing devices are mostly designed around 'active user operation.' When a user and an AI agent operate the same device simultaneously, power consumption and latency become greater engineering challenges."
He emphasized that agents need to run continuously in the background, retain context, coordinate tasks across multiple systems, and interact with software at machine speed. This will impose new requirements for the co-design of CPUs, NPUs, and GPUs. Qualcomm's accumulated low-power computing capabilities in the mobile sector are becoming its core advantage in the agent era.
Facing the exponential growth in AI token demand, Amon proposed a "compute continuum" strategy, which involves building a complete spectrum of computing power covering everything from in-ear connections using less than 2 milliwatts to kilowatt-scale systems in data centers. The core logic of this architecture is that agents will dynamically choose to execute tasks on the device, at the edge, or in the cloud based on cost, power, latency, and privacy needs, rather than solely relying on centralized cloud computing power.
Towards the end of his speech, Amon announced the launch of a new data center product brand, "Dragonfly," and previewed that a detailed roadmap would be revealed at an investor day on June 24. Qualcomm had previously confirmed that its custom AI inference chip collaboration projects with leading hyperscale cloud vendors are progressing smoothly, with the first shipments expected within the year. These custom chips focus on data center AI inference scenarios, aiming to compete differently in a market dominated by Nvidia's training chips. Qualcomm believes that as agents proliferate, the demand for inference computing power will far exceed that for training, and its technical expertise in energy efficiency and total cost of ownership will form a key barrier.
In his presentation, Amon also unveiled the Dragonwing IQ10 reference design for industrial AMRs and humanoid robots, featuring up to 700 TOPS of AI performance and utilizing an 18-core Oryon CPU. Furthermore, Qualcomm recently launched the Snapdragon C platform, aiming to bring AI PCs to the entry-level market above $300 and expand the Windows on Arm ecosystem. In automotive electronics, Qualcomm has maintained year-over-year growth exceeding 38% for multiple consecutive quarters, with its cockpit and driver-assistance platforms securing multiple mass production projects from automakers like Li Auto, Leapmotor, and Nio.
From mobile phones to cars, and from robots to data centers, Qualcomm is attempting to permeate its performance-per-watt advantage from the mobile era into every corner of the computing world. Amon concluded: "In the future, AI will run everywhere, and all of this will happen across the entire compute continuum."
Comments