Google just landed an AI model inside the iPhone, and Apple apparently said sure, go ahead. The Google AI Edge Gallery app, now available on iOS, lets users download and run Gemma 4 models entirely on-device. No cloud. No internet. No Google servers. Just a frontier-class AI running locally on Apple hardware.

This should embarrass Apple. While Siri still struggles to set a timer without misunderstanding you, Google is shipping models that run multi-step agentic tasks on the same phone. The Gemma 4 family includes models from 2 billion to 31 billion parameters, with the smaller ones designed specifically for mobile and edge deployment. Google announced the Edge Gallery app with Agent Skills support on April 2, and it is already trending on Hacker News.

The strategic implications are significant. Google is using open source to do what its search engine used to do: become the default layer that sits between users and their devices. By shipping Apache 2.0 models that run anywhere, including on competitors' hardware, Google is making Gemma the default AI stack for developers who do not want to be locked into OpenAI's API pricing or Anthropic's rate limits.

For Apple, this is a calculated concession. The company has been quietly letting third-party AI models onto iOS because its own Apple Intelligence has failed to keep pace. Reports suggest iOS 27 will let ChatGPT, Gemini, and Claude compete as Siri backends. Apple is becoming the platform, not the AI. That might be the smartest surrender in tech.

Meta should also be nervous. Google's Gemma 4 under Apache 2.0 is a direct shot at Llama's dominance in the open-source AI space. Gemma 4 models are already showing competitive benchmark performance, and Google's developer ecosystem is deeper than Meta's. The open-source AI crown is officially up for grabs.

The company that once got fined for pre-installing apps on Android is now getting invited onto iPhones. The AI race changes everything.