Analysis
A pair of glasses has very little room for a battery, nowhere for heat to go, and a wearer who will take them off if they get uncomfortable. That is the problem Arm describes in its 30 October overview of AI wearables, which names battery life, heat, weight and size as the constraints.
Arm's answer is to split the work. According to the company, tasks should run on the wearable, a connected phone or the cloud, depending on what each one needs. Its example is Everysight glasses using an Alif E4 platform with Arm's Ethos-U85 processor for local eye-tracking inference. Arm presents this as a way to cut transfers of raw sensor data while staying inside tight power limits.
Worth keeping in proportion: that is an architectural choice, described by the vendor. It doesn't establish all-day battery life, broad application performance or a privacy guarantee for every connected feature. We haven't seen independent battery, thermal or privacy testing.
I'd judge these devices on a normal day. Someone walks from a quiet room onto a noisy street, loses the phone connection, and asks for help while the frames are already warm. Each step changes what feels dependable. A slow answer, a battery that drains early or a recording indicator nobody can read can outweigh a good result on one isolated task.
So the test is a sequence, not a single run. Measure response time and energy use across realistic stretches of a day. Write down which functions need the phone or a remote service, and what still works when the connection drops.
If you are weighing a pair, run that sequence yourself, on the actual device, changing one condition at a time.
Source: Why wearables are the next frontier of AI-first personal computing · 2026-10-30
Why it matters
For wearables, efficiency decides how long help stays available and whether wearing it stays comfortable.
