LIVE: Qualcomm CEO Cristiano Amon speaks at Computex

Key Concepts

• Agentic AI: Autonomous AI systems capable of performing multi-step tasks, interacting with software, and operating across multiple services.
• Compute Continuum: The seamless integration of cloud and edge computing, where AI workloads are distributed based on efficiency and necessity.
• Tokens: The fundamental unit of AI processing; the "currency" of the AI era, with demand projected to grow exponentially.
• Orchestrators: Software layers (e.g., Open Interpreter, Claude Desktop) that manage and route AI tasks between local device compute and cloud resources.
• Physical AI: AI integrated into physical systems (cars, robots, industrial machinery) that perceives, plans, and acts in the real world.
• 6G Wireless: A next-generation connectivity standard designed for the AI age, focusing on distributed computing and RF-based sensing.
• Dragonfly: Qualcomm’s new product brand for data center AI solutions.

---

1. The Evolution of Agentic AI

The presentation argues that we are moving beyond the distinction between "cloud" and "edge" toward a unified system. Agentic AI is defined by its autonomy—unlike traditional software designed for human operation, agents interact with applications at machine speed.

• The Upgrade Cycle: Qualcomm posits that the shift to agentic AI will trigger one of the largest hardware upgrade cycles in industry history, as devices require dedicated NPUs (Neural Processing Units) and GPUs to handle local inference.
• Contextual Intelligence: Agents require sensor data to be proactive. By understanding the user's context (location, environment, habits), agents become useful assistants rather than just reactive tools.

2. Distributed Computing and Token Economics

A central argument is that AI workloads must be intelligently routed to optimize for cost and performance.

• Token Demand: Token consumption is scaling rapidly:
  • Level 1 (Conversational): ~10,000 tokens per task.
  • Level 2 (Reasoning): ~100,000 tokens per task.
  • Level 3 (Agentic/Autonomous): ~1,000,000 tokens per task.
• Efficiency: By using an orchestrator to keep specific tasks on the device (edge) and offloading complex reasoning to the cloud, users can achieve significant cost savings (e.g., 60% lower cost in coding tasks) and reduced token consumption.

3. Physical AI: Cars, Robots, and Industry

Qualcomm emphasizes that AI is moving into the physical world, requiring a "hierarchical compute system."

• Automotive: Vehicles are evolving into "AI-defined vehicles." They utilize two layers of intelligence:
  1. Cockpit AI: Personalizes the user experience and interacts with the user's agent.
  2. Physical AI: Uses cameras, radar, and sensors for navigation and safety.
• Robotics: Robots require a blend of consumer electronics integration (for battery life and cost) and industrial-grade reliability. They operate on three layers: Instant Execution (reflexes/balance), Action/Grounding, and Reasoning.

4. The Role of 6G and Sensing

Qualcomm introduces 6G as the first wireless generation designed specifically for AI. It rests on three pillars:

1. Connectivity: Enabling high-definition, continuous data exchange (e.g., "see what I see" capabilities).
2. Distributed Computing: The network itself acts as a data center, with inference occurring at radio base stations.
3. Sensing: Using RF signals as physical AI input. By triangulating these signals, networks can create "digital twins" of cities, enabling real-time tracking of objects, traffic, and aerial drones.

5. Qualcomm’s Strategic Advantage

Qualcomm positions itself as the leader in the "compute continuum," spanning from sub-2 milliwatt devices (earbuds) to multi-kilowatt data center solutions.

• Mobile Heritage: The company argues that its decades of experience in the mobile industry—where computing power is constrained by battery life—provides a unique advantage in creating power-efficient AI solutions for the data center.
• Dragonfly: The announcement of the "Dragonfly" brand marks Qualcomm’s formal entry into the data center market, aiming to provide high-performance, efficient AI processing at scale.

Synthesis and Conclusion

The transition to Agentic AI is not a future prospect but an ongoing reality. The core takeaway is that AI is the new form of computing. As agents become the primary interface for technology, the hardware ecosystem must evolve to support a distributed model where intelligence is processed locally for privacy and latency, and in the cloud for scale. Qualcomm’s strategy is to provide the underlying hardware and software orchestration to ensure this "compute continuum" functions seamlessly across every device, from the smartphone in a user's pocket to the industrial robots in a factory.