Analyst(s): Olivier Blanchard
Publication Date: October 14, 2025
The Amazon AZ3 chips and AZ3 Pro enable on-device AI acceleration, advanced audio and language capabilities, and multimodal sensor fusion to power Alexa+ experiences across the Echo lineup.
What is Covered in this Article:
- Amazon AZ3 and AZ3 Pro chip architecture, including the AI Accelerator and model execution capabilities.
- Advanced conversation detection and >50% improvement in wake-word recognition.
- Integration of state-of-the-art language models and vision transformers with local processing.
- Omnisense sensor fusion platform built on top of the chips.
- Strategic implications for Amazon’s voice-first AI ecosystem and data infrastructure.
The News: Amazon has unveiled two new custom silicon chips – AZ3 and AZ3 Pro – at the center of its Alexa+ hardware lineup. The AZ3 powers the Echo Dot Max, while the AZ3 Pro is built into the Echo Studio, Echo Show 8, and Echo Show 11. Both come with an integrated AI Accelerator designed to handle future edge AI models, improving responsiveness and contextual awareness.
These chips are built specifically for ambient AI tasks, allowing smoother, natural conversations and more advanced language processing. The AZ3 focuses on better wake-word and audio performance, while the AZ3 Pro adds support for language models and vision transformers. The Omnisense platform builds on this technology, enabling private, multimodal interactions that combine input from cameras, microphones, radar, and sensors.
Amazon’s AZ3 Chips Help Advance Voice-First AI Agentic UX
Analyst Take: Designed for Voice-First AI – The AZ3 and AZ3 Pro represent a major shift in Alexa’s compute model – from cloud-based processing to increasingly more on-device intelligence. To that end, Amazon is now building silicon tailored for voice-first, ambient AI instead of using general-purpose chips. This approach centers on faster response times, sensor integration, and local inference to better align with Alexa+’s growing conversational and contextual abilities. Paired with Omnisense, these chips form the backbone of how Alexa+ aims to expand across smart home, health, and commerce ecosystems.
The AZ3 boosts wake-word detection accuracy by over 50% compared to older Echo models, helped by upgraded microphones that block background noise. This means users can speak naturally from anywhere in the room without shouting or facing the device. It is built to detect conversation cues, a key part of ambient AI that depends on ongoing, context-aware interactions. With its built-in AI Accelerator handling real-time inference locally, the chip reduces the need for cloud processing. This makes the AZ3 a central driver of Alexa+’s speed and dependability in everyday environments – a critical component of the types of ubiquitous, remarkable natural language interactions that Amazon needs to embed in its devices in order to drive utility, customer engagement, and increase both the quality and the frequency of user interactions with Alexa devices.
Scaling Language Models on AZ3 Pro
Building on AZ3’s foundation, the AZ3 Pro adds the ability to run advanced language models and vision transformers, supporting both speech and visual understanding. This allows Alexa+ to interpret complex multimodal inputs directly on the device, improving both accuracy and response speed. Its extra computing power lets it handle more advanced models without always relying on the cloud.
Several advantages to that: The first is speed. Any delay in responses during verbal interactions with an assistant or with sensor-activated actions is a friction point and a negative hit to overall user experience (UX). Moving more intelligence and inference to an Amazon device means less lag between prompts and responses (or detection and action), translating into better, more rewarding user experiences. The second is a device’s ability to remain helpful to users even when network bandwidth is limited or inconsistent. The third is security: The more inference workloads remain local on a device, the more likely they are to remain secure for the user. I believe that this aspect of on-device intelligence will play a much bigger role in Amazon’s device silicon story in the future than it seems today, so keep an eye on this topic. Lastly, the design also supports future model upgrades without requiring new hardware, essentially extending the lifecycle of products for users. With these capabilities, the AZ3 Pro sets a new standard for high-performance Alexa+ devices both now and in the coming years.
Omnisense Sensor Fusion as an Execution Layer
Omnisense uses the AZ3 and AZ3 Pro as its processing base to run local sensor fusion involving 13-megapixel cameras, audio input, ultrasound, Wi-Fi radar, accelerometers, and Wi-Fi Channel State Information. Rather than sending raw sensor data to the cloud, these chips process it locally, allowing Alexa to react intelligently to real-world situations. For example, it can recognize when someone enters a room to trigger a reminder or send an alert if a garage door stays open after 10 p.m. This local-first approach helps protect user privacy while scaling ambient computing. By moving critical processing to the device, Omnisense strengthens Amazon’s ability to expand contextual AI across its product lineup.
Strategic Positioning Through Custom Silicon
Amazon’s move to design its own AZ3 and AZ3 Pro chips marks an essential step toward the type of vertical integration Amazon needs to build into its device and services ecosystem: It lets Amazon fine-tune performance for its specific AI needs instead of relying on third-party chips. This will allow for increasingly precise optimization of audio, language, and sensor workloads while balancing cost and efficiency. Owning its own silicon roadmap also gives Amazon more control over how to bring Alexa+ features to both its own devices and partner products.
Lastly, Amazon’s development of its own silicon creates a critical foundation for not only the unique features and performance specs Amazon needs to lead in the AI-enabled space, but also the kind of clarity it needs to easily differentiate itself from AI competitors such as Gemini, OpenAI, Copilot, Siri, and others. Note that UX differentiation is already shaping up to be a critical competitive vector for AI assistant and agentic vendors. Given Amazon’s unique market positioning, I believe this will be a core value proposition for its combined ecosystem of devices, services, and Alexa+ IP.
What to Watch:
- Performance improvements in conversational responsiveness driven by >50% better wake-word detection.
- How AZ3 Pro supports larger and more complex language and vision model deployments.
- Scalability of Omnisense sensor fusion across different Echo form factors.
- Potential use of the AZ3 architecture in partner devices beyond Echo.
- Integration pace of third-party services and applications on the Alexa+ platform.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Other insights from Futurum:
Amazon Q2 FY 2025 Revenue Tops View; Profit Outlook Disappoints Street
Application Security, Identity, & More at AWS re:Inforce 2025
Author Information
Olivier Blanchard is Research Director, Intelligent Devices. He covers edge semiconductors and intelligent AI-capable devices for Futurum. In addition to having co-authored several books about digital transformation and AI with Futurum Group CEO Daniel Newman, Blanchard brings considerable experience demystifying new and emerging technologies, advising clients on how best to future-proof their organizations, and helping maximize the positive impacts of technology disruption while mitigating their potentially negative effects. Follow his extended analysis on X and LinkedIn.
