Amazon is developing new AZ3 AI chips to enable on-device processing for Alexa devices and future mobile gadgets [1, 2].
This shift toward local silicon reduces reliance on cloud servers, which can minimize latency and improve user privacy. By processing data on the device, Amazon aims to make voice assistants more responsive and capable of functioning without a constant high-speed internet connection for basic tasks.
Panos Panay said to CNBC that "the goal is to bring more intelligence closer to the user" [3]. According to Panay, the AZ3 chip specifically handles sensor fusion and wake words locally, which reduces the latency for the Alexa+ experience [1].
These hardware updates are expected to reach millions of Echo homes by 2026 [1]. The company is also expanding its hardware ecosystem beyond the home. Amazon previously planned a privacy-focused AI bracelet for release in 2025 [4].
The development of the AZ3 chip represents a broader strategy to integrate custom silicon across Amazon's product line. While some complex tasks will still require cloud computing, the local chip handles the immediate triggers that allow a device to feel instantaneous. This approach allows the company to optimize power consumption, and increase the speed of AI interactions.
An unnamed Amazon spokesperson said the company has been working on this technology for a long time [5]. The transition to custom AI silicon puts Amazon in direct competition with other tech giants that design their own chips to power machine learning models on smartphones and wearables.
“"The goal is to bring more intelligence closer to the user"”
Amazon's move toward custom silicon for AI signals a transition from a cloud-first to an edge-computing strategy. By reducing the round-trip time between a user's voice command and the cloud's response, Amazon can improve the reliability of the smart home ecosystem and potentially lower the operational costs associated with massive server farms.
