AI is gradually moving from the cloud to edge devices and the physical world, noting that its long-term potential may be underestimated, but development will be incremental rather than instantaneous, said Cristiano Amon, the President and CEO of Qualcomm, during a presentation at the 2026 World Economic Forum in Davos on Wednesday. “AI agents are becoming more specialized and able to understand what people say, write, or see,” Amon said. “This is moving AI beyond the data center into devices people carry, wear, and use in daily life.” Beyond smartphones and PCs, new form factors such as smart glasses are beginning to scale. On personal AI devices, Amon said smart glasses shipments have already surpassed 10 million units. He predicted that as various types of personal AI devices proliferate, the total market could grow significantly in the coming years, with a potential acceleration between 2026 and 2027. In enterprise and industrial settings, AI on edge devices is also evolving at multiple levels. Amon explained that AI needs to move to the edge when instantaneous response is required. Applications like payments, recognition, or real-time translation lose effectiveness if they rely solely on the cloud. Similarly, when users prefer data and context to remain local, computing must occur on-device. As a result, some capabilities previously handled in the cloud are now migrating to the edge. From a broader perspective, Amon said historical trends in computing have shown that software ultimately finds the most efficient computing path. Once new capabilities scale on devices, corresponding applications naturally follow. He predicted that future AI will not be a choice between cloud or edge, but a hybrid model, with fast responses on-device and more complex reasoning in coordination with the cloud—a shift that could begin becoming visible as early as 2026. Amon also drew parallels between robots and the automotive industry, reflecting on Qualcomm’s entry into automotive computing. High-power servers are impractical in cars, requiring more efficient computation. He said the same logic applies to robotics: improving battery life from two hours to six or eight hours or reducing costs from $20,000 to $5,000 requires integrating cameras, sensors, and connectivity efficiently. He emphasized that robotics represents a clear, commercially viable AI domain, where training is task-specific and the problem boundaries are well-defined. Regarding data centers, Amon said forecasts for AI infrastructure often do not align with energy consumption estimates, pushing the industry to explore new solutions and further evolve data center architectures. Addressing concerns about an “AI bubble,” Amon likened the current state of AI to the internet in the early 2000s, noting that while growth may exceed initial expectations, development occurs incrementally rather than all at once. He stressed that AI’s expansion will take time and its pace remains uncertain.