Concept render of Google's Astra edge AI chip on a circuit board with a smartwatch and smart glasses in the background

Google’s Astra Edge AI Chip Cuts Power Use 40%: What It Means for Wearables

ByChris Wade

Key takeaway

Google Research previewed the Astra line of low-power edge AI chips, co-developed with Synaptics, at its Research@ event in Mountain View on October 14, 2026. Built on the Coral NPU M2, the chip uses 40% less energy than existing edge chips, a step toward always-on AI on wearables. It is not shipping yet: Synaptics will reveal detailed specs and a live demo on October 20, and no Google device or release date has been confirmed.

What Google announced

At its annual Research@ showcase on October 14, Google Research previewed the Astra line of low-power AI chips designed for edge computing. The chips are being co-developed with Synaptics, the chipmaker Google has partnered with on its Coral neural processing platform.

The headline spec: the new chip uses the M2, a successor to the Coral Neural Processing Unit (NPU), and consumes 40% less energy than existing edge chips, according to Avinatan Hassidim, vice president at Google Research.

The design targets voice, video, motion, and context detection processed in real time on wearable devices — the kind of workloads that would make an AI assistant on your wrist or glasses feel instant instead of cloud-delayed.

Why edge AI needs a chip like this

Edge AI means running AI computation directly on the device instead of sending data to cloud servers. That cuts response times — no round trip to a data center — and reduces the risk of sensitive data leaking in transit. The catch is physical: wearables and smart glasses are small, battery-powered, and sit on your body, so the chip must be highly power-efficient and generate very little heat.

“Conventional edge chips are slow and use too much energy, making an always-on (ambient) AI experience impossible,” Hassidim said at the event. The M2 is Google’s answer to that bottleneck: always-on ambient sensing that doesn’t drain the battery by noon.

Why it matters

Google is pushing on both ends of AI hardware at once. In data centers it scales its own Tensor Processing Units; on the device side, the Astra line is its play for the “ambient AI” market — glasses, watches, hearables, and cameras that react to what you see and hear without a cloud connection. Competitors like Qualcomm and Apple already ship on-device neural engines; Google’s 40% power advantage, if it holds up in shipping silicon, would give Android and Pixel hardware a tangible edge in battery life for AI features.

The move also fits the wider shift of inference away from data centers toward the edge — a trend playing out across wearables, industrial vision, and smart cameras this year.

What we still don’t know (analysis)

The following is analysis, not verified fact: the preview was deliberately light on specifics. Google and Synaptics have not published the chip’s process node, TOPS rating, memory configuration, or power envelope. Synaptics CEO Rahul Patel confirmed the chip “won’t stay in the lab” and will be built into Google devices, but declined to say which devices or when. More concrete answers arrive October 20, when Synaptics holds its own event with detailed specs and a live demonstration.

Worth watching: whether the 40% figure refers to per-inference energy or full-system draw, and whether shipping products hit the promised always-on battery budgets.

Background: the Google–Synaptics Coral story

This is not a first-time collaboration. Synaptics unveiled the Astra SL2600 series in October 2025 — the first chips to commercialize Google’s Coral NPU platform. In March 2026, the companies launched a next-generation Coral Dev Board (Astra SL2610) aimed at developers, with an open-source MLIR/IREE software stack and support for Gemma, Google’s open model family for the edge. The M2 chip extends that same line, which means developers already experimenting with Coral tooling may get an early path to prototype on it.

FAQ

What is Google’s Astra edge AI chip?

The Astra chip is a low-power AI processor previewed by Google Research on October 14, 2026, co-developed with Synaptics. It uses the Coral NPU M2 compute engine and is designed to run AI tasks like voice, video, motion, and context detection directly on wearable devices instead of in the cloud.

How much power does the Astra chip save?

According to Google Research vice president Avinatan Hassidim, the new chip uses 40% less energy than existing edge AI chips. The goal is to make always-on, ambient AI practical on battery-powered wearables without draining them or generating too much heat.

When will the Astra chip be available, and in which devices?

No release date or specific devices have been announced. Synaptics CEO Rahul Patel said the chip will be built into Google devices but did not say which ones or when. Synaptics plans to reveal detailed specifications and a live demo at its event on October 20, 2026.

What is edge AI, and why does it matter for wearables?

Edge AI runs AI computation on the device itself rather than sending data to cloud servers. This makes responses faster and keeps sensitive data on the device, lowering leak risk. For wearables like smart glasses and watches, it enables instant voice and context features — but only if the chip is power-efficient and cool-running enough to fit the form factor.

Sources


Similar Posts