Apple Intelligence launched less than two years ago built around Apple's own on-device neural networks, with ChatGPT as the add-on for requests that needed more processing power. That architecture is now being replaced from the ground up. Apple announced on June 8 that the new Apple Intelligence runs on foundation models developed with Google, using Gemini technologies at the core.
From Homegrown to Google-Powered
The redesigned system introduces a "system orchestrator" layer that routes tasks across Apple Intelligence features. The underlying foundation models are no longer Apple-built. They're Gemini-based and run both on-device and through Apple's Private Cloud Compute infrastructure. Apple calls the collaboration with Google "deep" and describes the result as a "huge upgrade," though the company hasn't released benchmarks showing exactly how much performance improved.
New capabilities include realistic image creation, advanced photo editing, visual question answering, better dictation accuracy, and speech generation on higher-end devices. Several of these existed in some form before. The real test is whether they work well enough that daily users actually notice a difference.
What This Costs Apple Strategically
Apple has spent years building its own AI research teams and promoting the idea that tight hardware-software integration (its custom silicon paired with its own models) produces better results than anything else. That message is now difficult to deliver when the foundation models come from a competitor.
The position is also unusual for Apple: the company now relies on two external AI providers (OpenAI for ChatGPT requests and Google for the core intelligence layer) to power its most-promoted software feature. Apple typically controls every layer of what it ships.
On privacy, Apple says "user data is only used to execute the immediate request and is not accessible to Apple or third parties." That's the same language used for Private Cloud Compute at launch. What isn't yet clear is how the Gemini integration affects that in practice. Apple hasn't published technical documentation of the collaboration to verify the data handling independently.
For people using Apple devices daily, this probably means Apple Intelligence gets noticeably better. For anyone watching where the big tech AI strategies are heading, the real signal is that even a company with Apple's resources and hardware advantage concluded it couldn't build competitive frontier AI models entirely in-house.