🤝 The deal: $1 billion a year for the AI Apple couldn't build

The company that built the most successful closed ecosystem on the planet just admitted it can't do it alone. Apple will pay Google $1 billion a year for a customized version of Gemini with 1.2 trillion parameters — eight times larger than its current model. It's not a rumor: it's a signed deal. And it's not a defeat — it's the smartest move Apple could have made.

🤝 The deal: $1 billion a year for the AI Apple couldn't build

On January 12, 2026, Apple and Google announced a multi-year alliance. It's not a rumor or a leak: it's a signed agreement, confirmed by both companies in a joint statement.

The numbers are brutal. Apple will pay approximately $1 billion a year to Google (up to $5 billion total) for a customized version of Gemini with 1.2 trillion parameters — eight times larger than the cloud model Apple has today. The model uses Mixture-of-Experts architecture, where only a fraction of parameters is activated per query, enabling massive capacity without multiplying inference cost.

But the detail that changes everything isn't how much Apple pays — it's how they deploy it.

🔐 What nobody expected: Apple privacy, not Google's

The key detail of the agreement is that Google's model runs on Apple's servers, not on Google Cloud. Apple licenses the trained model, but inference — every time Siri processes a request — happens on Private Cloud Compute, Apple's own cloud infrastructure with its privacy guarantees.

This means Google doesn't receive your data when you use Siri. Surprising, but strategically logical: Apple wasn't going to sacrifice its biggest competitive advantage (privacy) for a technical alliance.

From data control to the master play: Apple isn't just renting — it's learning.

🧠 More than a rental: Apple will use Gemini as a "teacher"

According to The Information's March 2026 report, the agreement goes far beyond simple licensing. Apple has full access to the Gemini model and will use it for distillation — the process where a large model "teaches" a smaller one, transferring knowledge without the small one needing to match its size.

In plain English: Apple will use Gemini to train its own local models (the ones that run Siri directly on your iPhone or Mac). So Siri can respond fast and offline, leveraging Gemini's intelligence without depending on it for every query.

And this brings us to the uncomfortable question: why couldn't Apple, with a $3 trillion market cap, do it alone?

📉 Why Apple gave up (temporarily)

Apple is two years behind in AI. Apple Intelligence launched with fewer features than promised, the Siri overhaul was repeatedly delayed in 2025, and the ChatGPT integration (announced as "for complex queries") was never more than a patch.

The reality is uncomfortable for Cupertino: neither the $3 trillion market cap, nor 2 billion active devices, nor its team of thousands of engineers have been able to build a competitive frontier model. While Google, OpenAI, and Anthropic launch models every week, Apple still doesn't have a decent conversational assistant.

This deal is a bridge. Apple is still developing its own 1-trillion-parameter cloud model and its "Baltra" chips for AI servers, with mass production targeted for the second half of 2026. But they needed something that worked now, and Gemini was the only option at scale.

— Max