Apple Turns to Google’s Gemini AI in $1 Billion Bet to Rescue Siri

When tech blogger John Gruber asked Siri a year ago, who won the 2004 North Dakota high school basketball championship, he got four different wrong answers in four attempts. Now Apple is paying Google $1 billion annually to ensure that kind of embarrassment becomes a thing of the past.

Years of Internal Dysfunction

Inside Apple, employees have dubbed the company’s artificial intelligence and machine learning group “AIMLess,” according to interviews with three dozen former employees. They called Siri a “hot potato” that gets continually passed between teams without anyone taking responsibility for fixing it. When employees mock their own division’s name, this reveals cultural problems that run deeper than any technical fix can address.

Consider the scale of dysfunction. Removing “hey” from “hey Siri” took Robby Walker’s team over two years to achieve. At WWDC 2024, Apple Intelligence features demonstrated on stage were largely non-functional; only the pulsing, colorful ribbon actually worked on test devices. Meanwhile, Paul Kafasis tested Siri on Super Bowl winners and found it achieved only 34% accuracy.

In reality, these failures reveal a culture that prioritized polish over ambition, small wins over breakthrough capabilities. While competitors raced ahead with large language models, Apple’s Siri team focused on incremental improvements that left the assistant further behind.

The Bake-Off That Led to Google

Faced with mounting evidence of internal limitations, Apple looked externally. An internal “bake-off” evaluation tested AI from OpenAI, Anthropic, and Google before the company made its decision. 

“I think they’re just accepting that they’re not a frontier lab. They’re not going to build to compete with Gemini 3. They need to fix Siri badly. And they’ve accepted that they just aren’t going to get there on their own quickly enough,” thinks Paul Roetzer, CEO of the Marketing AI Institute

Roetzer’s assessment reflects a broader industry reality: building frontier AI models requires specialized expertise and compute investments that even Apple’s resources cannot quickly replicate. This indicates that raw capital alone cannot buy AI leadership.

Anthropic’s Claude reportedly came out ahead in technical evaluations but demanded pricing above Google’s offer. Google, already paying Apple approximately $20 billion annually for default search placement, offered better financial terms. Put simply, Google’s existing relationship gave it an advantage that pure technical superiority could not overcome.

The $1 Billion Partnership

Under the arrangement, Apple will reportedly pay approximately $1 billion per year for access to a custom version of Google’s Gemini model with 1.2 trillion parameters. For context, Apple’s current Apple Intelligence uses just 150 billion parameters, making it eight times smaller than what Google provides. This demonstrates how far Apple’s in-house AI efforts have fallen behind. The deal strengthens Google’s position as a primary AI provider, proving its models are powerful enough for its chief rival to license.

Consider the financial irony: Apple paying Google $1 billion while receiving $20 billion from Google for search defaults. Billions flow in both directions between these rivals, signaling a fundamental shift in how the technology industry operates. In effect, the AI era is forcing even fierce competitors into strategic partnerships where isolation is no longer viable.

However, partnering with Google raised an obvious question: What about privacy? Apple has engineered an architecture that threads this needle.

Gemini will run on Apple’s Private Cloud Compute servers, with Google having no access to user data. According to Apple Security Research documentation:

“Private Cloud Compute extends the industry-leading security and privacy of Apple devices into the cloud, making sure that personal user data sent to PCC isn’t accessible to anyone other than the user, not even to Apple.”

By keeping Gemini invisible to users, with no public branding indicating the partnership, this suggests Apple views the Google connection as a necessary compromise rather than a selling point. Users will interact with Siri without knowing Gemini powers the backend.

What Users Will Get

For users, the payoff comes in March 2026 with iOS 26.4. Users will be able to ask Siri about content displayed on screen through on-screen awareness. Additionally, App Intents will enable cross-application actions, and personal context will let Siri draw on messages, email, and browsing history to provide relevant responses.

Full functionality requires iPhone 15 Pro or iPhone 16 models. For customers with older devices, this means they will not receive these benefits. iOS 26.3 is currently in beta testing ahead of the major release.

Strategic Implications

Leadership changes reflect the strategic shift. John Giannandrea, Apple’s AI chief who reportedly believed chatbots “didn’t add much value for users” after ChatGPT’s 2022 debut, retired in December 2025. Mike Rockwell, who previously launched Vision Pro, now leads Siri development.

With $130 billion in cash reserves, Apple could have invested more aggressively in frontier AI development, as noted by the ongoing antitrust scrutiny of its Google relationship. Instead, the company chose a pragmatic path: buy time with Google’s technology while building its own capabilities. In practice, Apple’s board appears to have recognized that the current AI market is too competitive to enter late without external assistance.

The Race Continues

Yet Apple isn’t surrendering its AI ambitions. Engineers are reportedly developing a 1 trillion parameter model targeted for late 2026. The partnership with Google may be a bridge, not a destination.

Siri’s overhaul launches with iOS 26.4 in March 2026. Whether that gives Apple enough runway to catch up remains the central question. Gruber’s four wrong Siri answers demonstrate how far Apple has to go. For now, the company that built its reputation on doing everything in-house has acknowledged a reality its frustrated users discovered long ago: Siri needs help that only a rival can provide.