Apple Rebuilt Siri on Google's Gemini. That's Not a Concession — It's a Strategy.
Agentic AIApple’s revamped Siri doesn’t run on an Apple-built large language model. It runs on Google’s Gemini — processed partly on-device and partly through Apple’s Private Cloud Compute infrastructure. That partnership, more than any specific feature announcement, tells you something significant about how Apple has chosen to compete in AI.
The Architecture Decision
The new Siri retains Apple’s on-device dictation through a model called AFM Core Advanced. But the substantive capabilities — understanding context from your messages and photos, reasoning across apps, executing multi-step tasks — run on Gemini.
This is not a temporary bridge. Building and training a foundation model competitive with Gemini for agentic tasks requires infrastructure investment that would take years to deliver what Apple needed now. The Gemini partnership delivers capability today. Apple’s Private Cloud Compute layer handles data routing and privacy obligations, letting Apple maintain its privacy positioning even while routing queries to Google’s models.
The practical result: a user asking Siri to “find the pasta recipe Mom mentioned last month and add the ingredients to my grocery list” now gets a useful answer. That specific task requires reading Messages, understanding context across a weeks-long conversation thread, extracting a list, and writing to another app. No prior version of Siri could do this end-to-end.
What Apple Gets From the Deal
Capability on the timeline it needed. Apple was publicly embarrassed by the gap between its AI announcements and the delayed delivery of advertised features. The Gemini partnership compresses years of model development into a deployment window.
Privacy positioning maintained. Processing through Private Cloud Compute means Apple can credibly claim that user data isn’t stored or used for Google’s training. Whether that’s fully verifiable is a separate question, but the architecture supports the claim.
The EU and China delay. Apple cites privacy compliance and regulatory concerns as reasons for not launching the Gemini-powered Siri in the EU and China. The EU’s AI Act and DMA create specific obligations around third-party AI model integrations that Apple hasn’t resolved yet. The EU delay reflects real regulatory complexity, not a soft launch strategy.
What This Means for the AI Assistant Landscape
The competitive picture among AI assistants has been Google Assistant vs. Alexa vs. Siri, with each running on its developer’s own model. That’s no longer true. Apple’s most capable assistant now runs on Google’s model, creating a strange alignment between two companies that compete in almost every other dimension.
The more interesting question is what Apple builds next. The AFM Core Advanced on-device model suggests Apple is investing in local processing. If Apple can close the capability gap with a proprietary foundation model over the next hardware cycle, the Gemini layer becomes replaceable — a ramp, not a permanent dependency. If not, the partnership deepens.
The So What
Apple choosing to build Siri’s agentic capabilities on Gemini signals a maturation in how large technology companies think about AI foundation models: they’re increasingly infrastructure, not moat. The companies that will differentiate aren’t those that train the best models — it’s the companies that deploy them most effectively within trusted data surfaces. Apple’s bet is that Private Cloud Compute plus tight device integration is the moat, not the model itself.
Whether that thesis holds over the next three years, against competitors who are both building models and deploying them at scale, is the question worth tracking.
Source: The Decoder
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →