Apple and Google Just Teamed Up on AI — And It's Not What Anyone Expected

For years, the tech narrative has been Apple vs. Google — two giants at opposite ends of the privacy vs. openness spectrum, competing for your attention, your data, and your loyalty. So when Apple announced this week that its third-generation Apple Foundation Models — called AFM 3 — were built in collaboration with Google, I genuinely did a double take. These are not two companies that collaborate. And yet here we are.

What Is Apple Foundation Models 3?

AFM 3 is a family of five AI models that power Apple Intelligence — Apple's on-device and server-based AI system. The lineup spans the full range of deployment scenarios: some models run entirely on your iPhone or Mac with no data ever leaving the device, while others run on Apple's Private Cloud Compute servers for more demanding tasks.

What's new with the third generation is the breadth and the collaboration. Apple partnered with Google specifically to build these models, which is remarkable given how fiercely Apple typically guards its AI development internally. The result is a family of models designed to be both highly capable and deeply privacy-preserving — a combination that Apple has always claimed as its competitive advantage over AI competitors.

Why Google? Why Now?

The Apple-Google collaboration in AI makes more sense when you consider Google's existing relationship with Apple. Google has paid Apple tens of billions of dollars annually to be the default search engine on iOS and Safari. More recently, Apple integrated Google Gemini as one of the AI options within Apple Intelligence. So there's already a business relationship — what's new is it apparently extending into model co-development.

The "why now" part is even more interesting. Apple has been playing catch-up in the generative AI race. Its original Apple Intelligence features, while genuinely privacy-forward, were criticized for being less capable than competing offerings from Google and OpenAI. By partnering with Google on model development, Apple gets access to Google's deep expertise in large language model training while Google gets continued preferred placement in Apple's ecosystem. Either way, both companies benefit from being seen as collaborating on privacy-first AI.

5 Models, One Philosophy

The five-model family is designed around a core principle: the right model for the right task. Smaller, faster models run locally on device for everyday tasks like writing suggestions, photo descriptions, and Siri queries. Larger, more capable models run on Apple's Private Cloud Compute for complex tasks that require more processing power — but with Apple's strong privacy guarantees about data not being used for training or retained after the session.

This tiered approach is smart. It means Apple doesn't have to choose between capability and privacy — it can offer both, depending on what the task requires. And crucially, Apple can credibly say that your most sensitive data never leaves your device, because the most personal interactions are handled by the local models.

The Smart Glasses Delay — A Related Data Point

In the same week Apple announced AFM 3, the company also quietly pushed back its first smart glasses product. The announcement was originally expected ahead of WWDC in 2027 — now Apple says it won't reveal the product until WWDC in June 2027. That's not a huge delay, but it does suggest Apple is being very deliberate about this product category.

Reading between the lines: Apple wants to make sure the AI that powers its smart glasses is genuinely good before it puts them on your face. Given how the Meta Ray-Bans have been criticized for mediocre AI responses, Apple's caution is understandable. Getting AFM 3 right first, then building hardware around it, is probably the right sequencing.

What This Means for Users

If you're an iPhone or Mac user, AFM 3 should mean noticeably better Apple Intelligence performance when it ships in the next major iOS and macOS updates. Smarter Siri, better writing tools, more capable photo and document analysis — all while (Apple promises) maintaining the privacy model that distinguishes it from Google Assistant or ChatGPT.

For the broader AI industry, the Apple-Google collaboration signals something important: even fierce competitors are recognizing that developing capable AI models requires scale and expertise that no single company can easily go alone on. If Apple — the most famously insular major tech company — is co-developing foundation models with Google, the era of fully proprietary AI stacks built in isolation may be ending.

My Take

I'm genuinely impressed by Apple's approach here, even if the Google collaboration raises some eyebrows. Apple has a credibility problem on AI capability — people love the privacy angle but wish the AI was sharper. If AFM 3 closes that gap, Apple Intelligence could become the de facto standard for people who want AI assistance without the surveillance business model. That's a compelling value proposition for a huge portion of the market. Let's see if the models live up to the billing when they ship.

What's your experience? Drop a comment below! 👇

Does it change how you feel about Apple Intelligence knowing that Google helped build the underlying models — or does Apple's privacy architecture make the collaboration irrelevant to you?