On January 12, Apple and Google announced a multi-year partnership that will put Google’s Gemini AI at the heart of the next version of Siri. Apple is reportedly paying around $1 billion per year for the arrangement, which is expected to roll out with iOS 26.4 this spring.
Apple CEO Tim Cook told investors that Apple Intelligence will “continue to run on the device and run in Private Cloud Compute and maintain our industry-leading privacy standards.”
That sounded reassuring. Then Google started talking.
The CEOs Can’t Get Their Story Straight
During Alphabet’s Q4 2025 earnings call on February 4, CEO Sundar Pichai described Google as Apple’s “preferred cloud provider” for the partnership. Chief Business Officer Philipp Schindler repeated the same phrase in his own remarks.
This is a problem. Cook said Siri runs on-device and in Apple’s Private Cloud Compute. Pichai said Google is the preferred cloud provider. Both statements can’t be entirely true.
When analysts asked Alphabet executives to clarify the discrepancy, they declined. Cook, for his part, was asked about partnership specifics and responded: “We’re not releasing the details of that.”
The most generous interpretation is that both CEOs are describing different layers of a tiered system. Simple queries stay on your iPhone. More complex requests go to Apple’s Private Cloud Compute. The hardest questions — the ones that require Gemini’s full power — route through Google’s servers.
But neither company will confirm that architecture. And the gap between “your data stays on your device” and “Google is our preferred cloud provider” is exactly where privacy questions live.
What We Know About the Technical Setup
Apple has shared limited technical details. The new Siri architecture has three components: a query planner, a knowledge search system, and a summarizer. Google’s Gemini models power the planner and summarizer. Apple says a “privacy buffer layer” sits between user queries and Google’s infrastructure.
The numbers Apple is citing are significant. The new Siri reportedly handles 1.2 trillion parameters, compared to 150 billion in the current version. Complex instruction success rates jump from 58% to 92%. Response times drop below 0.5 seconds.
Those improvements don’t come from on-device processing. A model that size runs on server farms, and someone has to operate them. The question is whether those server farms belong to Apple or Google — and what data flows between them.
Apple says it anonymizes requests before they reach external infrastructure. Google has confirmed it won’t receive identifiable user data through the partnership. But anonymization isn’t binary. Research has repeatedly shown that “anonymized” data can be re-identified when combined with other data sources — and Google has more data sources than almost anyone.
The $20 Billion Elephant
This isn’t Apple and Google’s first financial entanglement. Google pays Apple roughly $20 billion per year to remain the default search engine in Safari, a deal that came under federal antitrust scrutiny and is currently being appealed by the DOJ.
That deal has always sat uncomfortably next to Apple’s privacy marketing. The company that runs “What happens on your iPhone stays on your iPhone” ads is collecting $20 billion a year from the company whose business model is advertising based on user data. Privacy advocates have called it Apple’s biggest hypocrisy for years.
The Gemini deal adds another layer. Now Apple isn’t just routing your searches through Google’s index. It’s building its flagship AI assistant on Google’s models, processed on infrastructure that Google at minimum refers to as its own. The $1 billion Apple is paying Google is a fraction of what Google pays Apple for search — but the data implications could be larger.
When you search Safari, you can switch to DuckDuckGo. When Siri’s core intelligence runs on Gemini, there’s no toggle for that.
What Google Wants From This
Google isn’t doing this for $1 billion. That’s pocket change for a company earning over $300 billion in annual revenue. What Google gets is distribution.
Gemini is locked in a competitive fight with OpenAI, Anthropic, and a growing field of AI competitors. Getting embedded as the foundation of Siri puts Gemini on more than a billion active Apple devices. Even if Google never sees a single user’s personal data, having its AI models shape how a billion people interact with their phones is an extraordinary competitive advantage.
There’s also the training signal question. Apple says Google won’t get user data. But when Gemini processes billions of queries through Apple’s pipeline, Google learns which types of queries work well and which don’t — aggregate behavioral patterns that improve future model training without technically involving “personal data.”
Neither company has addressed whether aggregate, non-identifiable usage data flows back to Google.
Meanwhile, Google Is Asking Android Users for Everything
The timing of the partnership announcement highlighted a striking contrast. In January, Google launched a “Personal Intelligence” beta asking Android users in the US for access to their Gmail, Google Photos, and YouTube data to train Gemini.
The sales pitch included examples like Gemini referencing “family road trips to Oklahoma found in Google Photos” to suggest tire options. That’s not a helpful assistant — that’s a data-harvesting operation with a friendly interface.
Apple users will interact with the same underlying Gemini models, but presumably without the personal data access. The question is whether that separation holds over time. Partnerships evolve. Features expand. Terms of service update. The history of tech privacy is the history of boundaries that seemed firm at launch and dissolved quietly in subsequent updates.
What This Means
Apple chose Gemini after testing technology from OpenAI and Anthropic. It picked the company with the largest advertising business, the most extensive data collection infrastructure, and a model that’s actively asking users on other platforms for access to their personal emails and photos.
Maybe the privacy buffer layer works perfectly. Maybe anonymization is robust enough that Google gains nothing beyond the $1 billion fee. Maybe the “preferred cloud provider” language was just Pichai talking up the deal to investors.
But Apple built its brand on privacy certainty. “It just works” applied to data protection too. Now the answer to “where does my Siri data go?” is “we’re not releasing the details of that.”
For a company that charges premium prices partly on the promise that your data stays yours, that’s not good enough.
What You Can Do
- Don’t use Siri for sensitive queries until the architecture is clarified. If neither CEO can explain where your data goes, assume the worst.
- Watch for the iOS 26.4 privacy labels. When the update ships this spring, Apple should detail which Siri features use on-device processing, Private Cloud Compute, and Google’s infrastructure. If those disclosures are vague, that tells you something.
- Disable features you don’t need. Apple has historically allowed users to opt out of AI features. Check your settings when the update lands.
- Pay attention to the terms of service. Both Apple and Google will update their privacy policies around the launch. Read the changes. The important shifts are usually in the clauses nobody reads.
- Consider the pattern. The $20 billion search deal. The Gemini partnership. If privacy is genuinely your priority, evaluate whether Apple’s actions match its advertising.