Apple Just Rebuilt Siri on Google Gemini — And Samsung Is All In Too

Two announcements this week just changed how 2 billion people will interact with AI — and most people are sleeping on it.

Apple officially confirmed that the next version of Siri has been completely rebuilt from the ground up on Google’s Gemini model, running in part through Apple’s Private Cloud Compute infrastructure. Meanwhile, Samsung announced a goal to deploy Gemini across 800 million devices by the end of 2026 — doubling its current footprint.

For context: that’s essentially both of the world’s largest mobile platforms converging on a single AI model. Not ChatGPT. Not Claude. Gemini.

The implications are massive — for Apple users, for Google’s business, for OpenAI’s ambitions, and for anyone building AI-powered apps. Let’s break it all down.

What Apple Actually Announced

Apple’s rebuilt Siri isn’t just an upgrade. It’s a complete architectural replacement. The old Siri ran on Apple’s own on-device models with cloud fallback — models that, frankly, embarrassed the company every time someone compared them to ChatGPT or Gemini.

The new Siri runs on a hybrid architecture:

• On-device inference using Apple Silicon (A18 Pro and M-series chips) for low-latency, privacy-sensitive requests
• Private Cloud Compute for more complex queries — Apple’s encrypted cloud layer that processes data without storing it
• Google Gemini (specifically the 1.2 trillion-parameter multimodal version) for the hardest tasks requiring deep reasoning, image understanding, or cross-app workflows

The key new capability Apple is leading with is on-screen awareness. The new Siri can see what’s on your screen — an email you’re reading, a photo you’re viewing, a document you’re editing — and act on it intelligently. Ask it to “summarize this contract” or “find flights to the city in this photo” and it doesn’t just respond with information. It does the task.

“We’ve rebuilt Siri from the foundation. The version you’ve known is gone. What’s coming understands context the way a person does — across your apps, your content, and your life.” — Apple marketing materials, as reported

This is a major philosophical shift for Apple. For years, the company resisted using external AI models, preferring to build everything in-house for privacy reasons. The Gemini partnership signals that Apple has decided it can’t match Google’s training infrastructure, and that a hybrid architecture with privacy guarantees is acceptable.

Why Google? Why Not OpenAI or Anthropic?

Apple reportedly evaluated partnerships with OpenAI, Anthropic, and Google before choosing Gemini. The decision reportedly came down to three factors.

1. On-Device Capability

Google has spent years optimizing Gemini for edge deployment. Gemini Nano — the smallest tier — runs natively on Pixel phones and is already deployed on Samsung Galaxy devices. Apple needed a partner who could work with Apple Silicon’s Neural Engine, not just sell API access. Google had the on-device story. OpenAI doesn’t.

2. Multimodal Depth

The new Siri’s most impressive demos involve understanding images, screens, and documents simultaneously. Gemini’s multimodal architecture was purpose-built for this. GPT-5.4 is also multimodal, but Gemini’s native vision capabilities are tighter for real-time on-device inference scenarios.

3. Privacy Architecture Compatibility

Anthropic and OpenAI’s models primarily run via cloud API with standard data retention policies that conflict with Apple’s privacy commitments. Google agreed to a custom data handling arrangement that routes Apple traffic through Private Cloud Compute with zero-retention guarantees — a deal structure that required Google’s cooperation at the infrastructure level.