The quiet rise of on-device AI — and why your next phone may not need the cloud
Smaller, faster models are moving onto handsets, with big implications for privacy, battery life and data costs.
For most of the past few years, using an AI assistant meant sending your words to a data centre. That is starting to change.
Chipmakers are now shipping processors with dedicated neural units capable of running compact language models entirely on the phone.
The benefits are obvious: lower latency, no data charges and, crucially, less personal information leaving the device.
The trade-off is capability. On-device models are smaller and can be less accurate on complex tasks, which is why most products now use a hybrid approach.