Google's chip team just showed off something that could quietly change how AI runs on the devices you wear. At its annual Research@ event in Mountain View on October 14, Google Research previewed the next generation of its Astra line of edge AI chips — co-developed with chipmaker Synaptics — and the headline number is striking: 40% less power consumption.
The new chip is built around the M2, the successor to the Coral Neural Processing Unit (NPU), the AI compute engine Google introduced last year. Edge AI — running AI models directly on a device instead of sending data to cloud servers — is the whole bet here. Keeping computation on the device means faster responses and less risk of sensitive data leaking out. The catch has always been physics: tiny devices can't tolerate power-hungry, heat-generating chips.
That's exactly the problem Google says it cracked. Avinatan Hassidim, vice president of Google Research, put it bluntly at the event: conventional edge chips are too slow and burn too much energy, making an always-on AI experience impossible. The M2-based Astra chip flips that math, using 40% less energy than existing products.
In practical terms, this chip is designed to process voice, video, motion, and context detection in real time on wearables — think smart glasses that understand what you're looking at, or a smartwatch that handles AI tasks without draining the battery by noon. It runs on smartphones, smart glasses, and smartwatches, where every milliwatt matters.
Synaptics CEO Rahul Patel added the part that matters most for consumers: this chip isn't staying in the lab — it will be built into Google devices. He didn't say which devices or when, but Synaptics will reveal full specifications and a live demonstration at its own event on October 20, so the details are coming soon.
WHY IT MATTERS
The AI race is increasingly moving off the cloud. Between Apple's smart-home push and Google's on-device ambitions, the next big AI battleground isn't a bigger data center — it's the chip inside your glasses. A 40% power cut is exactly the kind of jump that turns “always-on AI” from a demo into something you'd actually use all day.
THE QUICK TAKE
Google's Coral NPU line is growing up fast. Watch October 20 for the full spec reveal — and the first real look at what on-device AI feels like when the power budget stops being the bottleneck.
Comments
Post a Comment