COMPUTEX 2026 CEO Keynote: Qualcomm

Author: Cristiano Amon, President & CEO, Qualcomm Collection: Computex 2026

Summary

Qualcomm CEO Cristiano Amon delivered the opening keynote at Computex 2026 in Taipei, presenting his thesis that 2026 is "the year of the agents" — a structural inflection where AI evolves from conversational tool to autonomous actor. Amon argued that the transition to agentic AI (AI that takes autonomous, multi-step actions without continuous human intervention) fundamentally changes the architecture, economics, and device categories of computing. Because agents operate continuously, carry persistent context, coordinate across multiple data sources, and generate tokens at machine speed rather than human speed, every device in the edge ecosystem must be redesigned. Amon scoped the addressable surface area at 6 billion smartphones, 2 billion personal AI devices (wearables and glasses), 2 billion PCs, and 500 million connected cars — all becoming "endpoints for agents" rather than standalone platforms. (Image 345, 43:15)

A central analytical thread was the token-demand explosion driven by agentic workflows. Amon presented a three-level taxonomy: Level 1 conversational AI consumes roughly 10,000 tokens per single-turn task; Level 2 reasoning AI roughly 100,000 tokens per multi-turn task; Level 3 agentic AI 1,000,000+ tokens per autonomous multi-step task — a 100x increase across two generations. (Image 645, 74:25) In aggregate, global token demand in a 10-second window is estimated at 31.7 billion tokens in 2026, rising to 1.27 trillion by 2030 — a 40x increase driven almost entirely by agent-initiated workloads. (Images 656–657, 75:30–75:45) Amon concluded with the announcement of "Dragonfly," a new Qualcomm product brand for data center inference, extending the company's portfolio from sub-2mW earbuds to 200kW server racks. (Images 705–706, 81:35–81:40)

The keynote's practical demonstration cases included: distributed agentic AI on Snapdragon PCs saving 1.4 million tokens and achieving 60% lower cost on coding workloads (Image 667, 77:10); 30% fewer tokens and 4x lower cost on web-page creation tasks by routing work between on-device and cloud compute intelligently (Images 673/686, 78:10/79:20); a dual-layer automotive AI architecture separating cockpit personal intelligence from physical driving AI; a robotics platform encompassing autonomous mobile robots (AMRs), four-legged robots, humanoids, and drones; and a 6G framework described as having three new pillars — Connectivity, Distributed Computing, and Sensing — with sensing enabling city-scale digital twins via radio-frequency-based perception.