Google is accelerating its push toward on-device artificial intelligence (AI) for Pixel smartphones, introducing new tools that allow developers to build AI-powered applications capable of running entirely on users’ devices without requiring an internet connection.
The move marks Google’s latest effort to reduce reliance on cloud-based AI processing by enabling faster performance, enhanced privacy and offline functionality across its Pixel ecosystem.
Google introduces Gemma 4 E2B for Pixel devices
At the centre of the announcement is Gemma 4 E2B for TPU, a lightweight version of Google’s open-source Gemma AI model family designed specifically for the Tensor Processing Unit (TPU) inside Pixel smartphones.
The model allows AI tasks to be processed directly on the device instead of remote servers, enabling applications to continue working even when users are offline.
By keeping AI processing on the handset, Google said sensitive user data can remain stored locally, offering improved privacy alongside lower response times.
Broader Gemma AI expansion
The launch forms part of Google’s broader expansion of the Gemma AI family.
Earlier this year, Google introduced the Gemma 4 12B model, bringing multimodal AI capabilities and native audio input to laptops without requiring dedicated AI hardware.
The company also released quantization-aware training (QAT) versions of Gemma 4, reducing memory requirements while maintaining model quality, making the technology more accessible across a wider range of consumer devices.
Pixel 10 showcases offline AI features
Google also demonstrated the technology running on the upcoming Pixel 10 during Google I/O India, highlighting several AI features that function without an internet connection.
The demonstrations included offline AI assistants capable of planning trips, suggesting recipes and controlling smart home devices.
Another feature, called Mobile Actions, enables users to perform tasks such as turning Wi-Fi on or off or launching Google Maps using voice or text commands processed entirely on the device.
Multimodal AI works without internet
Google said Pixel 10 users will also be able to hold AI-powered conversations offline, identify landmarks, plants and objects from photos, and transcribe lectures or voice recordings directly on the device.
These capabilities are expected to benefit users travelling in areas with limited connectivity or working in environments where internet access is unavailable.
Enterprise applications
Beyond consumer features, Google highlighted several enterprise use cases for on-device AI.
The company said retailers could use Pixel devices to generate offline shopping maps, while automotive technicians could identify faulty vehicle parts using photographs processed locally by AI.
Google’s latest announcement reflects the growing industry trend toward running AI models directly on smartphones, reducing dependence on cloud infrastructure while improving speed, security and reliability.