A billion dollars a year to Google for AI model access. Twenty billion a year from Google for search defaults. Two financial pipelines running in opposite directions between the same two companies, and somehow the conversation is about whether Siri can process queries fast enough.
This is not a tech story. It is a power story.
The Dependency Nobody Wants to Name
You already know the shared facts: the Gemini deal, the Siri delays, the leadership shuffle, the class action settlement. What deserves more scrutiny is the structural picture these events compose when you step back far enough.
Apple now relies on Google for two of its most strategically important functions. Google pays Apple roughly $20 billion annually for Safari search placement, money that makes up an estimated 20% of Apple's entire Services revenue. Simultaneously, Apple is paying Google approximately $1 billion per year so that the next generation of Apple Foundation Models can be built on Gemini. These are not unrelated transactions. They are the architecture of mutual dependency between the two companies that already control how most humans access the internet.
Regulators noticed the first relationship. A federal judge ruled Google had illegally maintained a search monopoly, though the remedies phase preserved Google's ability to keep paying Apple for default placement. The DOJ is now appealing. And while that case crawls through the courts, Apple and Google quietly deepened their entanglement in an entirely new domain: the AI layer that will sit between users and every app, every search, every question they ask their phone.
Nobody is asking the obvious question. If the same company that provides your search results also provides the foundation model powering your voice assistant, how many meaningful chokepoints remain between you and the open internet?
The Privacy Claim Deserves More Than a Press Conference
Apple's pitch is that Gemini will be customized, containerized within Private Cloud Compute, and that no user data flows to Google. Tim Cook said it directly: "We're not changing our privacy rules." The joint statement was careful and coordinated. But as one security expert noted, "Private Cloud Compute is only as private as the weakest link." And the joint statement, as multiple analysts observed, was notably light on implementation details.
The concern here is not that Apple is lying. It is that the architecture of trust required to make this work has never been tested at this scale, with this level of model complexity, involving a partner whose entire business model is built on harvesting user data for advertising. Google's core business depends on data collection and targeted advertising. Apple runs ads positioning itself as the opposite of that. Partnering on the foundational AI layer powering 1.5 billion devices creates tension that no press release can resolve.
Consider what happens when Siri, powered by Gemini-derived models, gains the "personal context" and "screen awareness" features Apple has promised. That means the model processes what is on your screen, your messages, your calendar, your documents. Apple says this all stays on-device or within its private cloud. But the model's architecture, its biases, its behavioral tendencies, were shaped by Google. As one researcher put it: "You can audit data flows, but you can't audit the black-box reasoning that determines user experience." Control over model behavior is a form of infrastructure power that we do not yet have regulatory frameworks to address.
Follow the Incentives
The pattern here is older than AI. Apple has always been a distribution company as much as a technology company. Its genius is controlling the last mile: the device in your hand, the operating system, the default settings. Google has always been an infrastructure company disguised as a consumer brand. Its genius is controlling the pipes: search, advertising, cloud, and now foundation models.
What the Gemini partnership does is formalize a division of power that was already emerging. Google builds the intelligence layer. Apple controls the interface. Users get a polished experience. And the two most valuable companies on Earth lock each other into a relationship that no competitor can easily disrupt and no regulator has yet figured out how to govern.
A SellCell survey found 73% of iPhone users who tried Apple Intelligence features felt they added little to no value. A CIRP poll showed only 14% of iPhone 17 buyers cited new features like AI as their primary motivation for upgrading. The supposed "AI supercycle" that was meant to justify all of this? Analysts at CNBC reported it never materialized in 2025. The actual upgrade driver was mundane: pandemic-era phones aging out, with an estimated 315 million users overdue for replacements. Apple's record $143.8 billion quarter had less to do with intelligence and more to do with batteries dying and screens cracking.
So who does Apple Intelligence actually serve right now? Not the users, most of whom barely notice it. Not the people whose data trained the models, since Google's Gemini was built on a web-scale corpus with the usual questions about consent and compensation. It serves Apple's narrative to Wall Street that it has an AI strategy. It serves Google's narrative that Gemini is the industry standard. And it serves both companies' interest in maintaining the duopoly over mobile computing that they have carefully constructed over the past fifteen years.
The features themselves are fine. Live Translation is useful. Visual Intelligence has real utility. The Foundation Models framework gives developers something genuinely valuable: free, on-device inference. These are not nothing. But describing them as "intelligence" obscures the structural choices underneath. Every on-device model that routes a complex query to Private Cloud Compute makes a decision about what stays local and what goes to a server whose ultimate architecture we must take on faith.
The question is not whether this technology works. It is who it works for. Right now, the answer is two companies in Cupertino and Mountain View, who have decided that the future of AI on the most popular consumer device ever built will be a joint venture between the world's largest advertising company and the world's most valuable hardware company. That arrangement may produce great products. It will certainly produce great earnings calls. Whether it produces something that genuinely serves the public interest is a question that deserves better than a carefully worded statement to CNBC.
