Source: PYMTS – https://www.pymnts.com/artificial-intelligence-2/2026/google-pushes-ai-onto-devices/
Artificial intelligence development has focused on cloud-based systems for years. Large models run in centralized data centers. These systems support chatbots, enterprise software, and consumer tools. Relying on the cloud increases latency, infrastructure costs, and data transfers across networks.
Google is now taking a different approach. The company is expanding edge AI alongside cloud-based Gemini models. Google Edge tools and a compact model called FunctionGemma are central to this effort. This strategy makes local execution a key part of AI infrastructure.
FunctionGemma runs directly on mobile devices. The model translates natural language commands into executable actions. Processing occurs without cloud inference. Devices respond immediately to user input. Functionality remains available even with limited or no connectivity. VentureBeat reports that the model focuses on device control rather than conversation.
FunctionGemma builds on the Gemma 3 270M model. Its training differs from typical language models. The system specializes in function calling. Outputs follow structured formats designed for software execution. MarkTechPost reports that targeted fine-tuning improved accuracy for mobile action tasks.
Local execution eliminates network round trips. User data stays on the device. This approach supports privacy-focused design and real-time responsiveness. The model’s footprint fits constrained hardware environments. Google sees FunctionGemma as an embedded part of applications, not just a standalone assistant.
FunctionGemma supports a hybrid AI architecture. Lightweight edge models handle frequent operational tasks. Cloud models take on complex reasoning and generation. This setup reduces demand for cloud inference and stabilizes performance costs. Local data processing also lowers the risk of exposure to centralized data as regulatory scrutiny increases.
