📰 AI News: Apple Will Pay Google $1 Billion Annually to Fix Siri—Because Its Own AI Isn't Ready

Nov 7 • AI News

Apple is finalizing a deal to pay Google roughly $1 billion per year for a custom 1.2 trillion-parameter Gemini model to power the long-promised overhaul of Siri, according to Bloomberg. The agreement—codenamed Project Glenwood, represents Apple's admission that its in-house AI models aren't competitive enough to deliver the Siri upgrade promised at WWDC 2024.

The revamped Siri is expected to launch in spring 2026 with iOS 26.4, but Apple plans this as a temporary solution while developing its own 1 trillion-parameter model.

The announcement:

On November 5, 2025, Bloomberg reported that Apple is finalizing an agreement to license Google's 1.2 trillion-parameter Gemini AI model for approximately $1 billion annually to power Siri's upcoming overhaul. The custom Gemini model will handle Siri's summarizer and planner functions—the components that synthesize information and execute complex multi-step tasks—while some features will continue using Apple's in-house models.

The partnership, overseen by Apple executive Mike Rockwell under Project Glenwood, follows an extensive evaluation period where Apple also tested models from OpenAI and Anthropic before selecting Google based primarily on cost considerations. The upgraded Siri is targeted for spring 2026 release alongside iOS 26.4.

What's happening:

Google's 1.2 trillion-parameter Gemini model dwarfs Apple's current AI capabilities, which use a 150 billion-parameter cloud-based model and a 3 billion-parameter on-device model. The massive parameter difference—8x larger than Apple's cloud model—represents a fundamental gap in AI capability that Apple cannot close quickly. For context, parameters measure how an AI model understands and responds to queries, with more parameters generally indicating greater capability.

The Gemini model will run on Apple's Private Cloud Compute infrastructure, meaning user data will not be shared with Google despite Google providing the underlying AI technology. This architecture allows Apple to maintain its privacy-first positioning while leveraging Google's superior AI capabilities. Apple has emphasized that its Private Cloud Compute servers process AI workloads without exposing user data to third parties.

Under the arrangement, Gemini will specifically handle Siri's summarizer and planner functions—the most computationally demanding components that require sophisticated reasoning and multi-step task execution. Apple's own models will continue handling other Siri features, creating a hybrid architecture.

This division suggests Apple wants to limit dependency on Google while acknowledging it needs external help for the most advanced capabilities.

The deal follows earlier discussions in 2024 where Google explored providing general knowledge responses to Siri, though those talks didn't result in product integration. This time, the agreement is broader and tied directly to Siri's core logic rather than peripheral features. The partnership builds on Apple's existing relationship with Google, which already pays Apple approximately $20 billion annually to remain the default search engine on Apple devices.

Apple is simultaneously developing its own 1 trillion-parameter cloud-based model that could be ready for consumer applications as early as late 2026. The company views the Google partnership as temporary and intends to transition to an in-house solution once its models reach competitive capability levels. However, Apple has been bleeding AI talent, including the recent departure of its models team head, raising questions about its ability to catch up.

The revamped Siri represents Apple's response to criticism that its voice assistant has lagged significantly behind competitors like Google Assistant and Alexa in handling complex queries, understanding context, and leveraging generative AI. The current Siri frequently resorts to "I found this on the web" responses when unable to handle requests directly. The Gemini-powered version aims to enable natural conversations, multi-step task execution, and contextually aware responses that competitors have offered for years.

Why this matters:

🎯 Apple Just Admitted It Lost the AI Race – Paying a competitor $1 billion annually for core technology is Apple admitting its decade-long Siri development couldn't produce a competitive AI assistant. This isn't collaboration, it's capitulation.

💡 The "Build Everything In-House" Strategy Is Dead – Apple's legendary vertical integration philosophy just broke on AI's complexity. When even Apple with $160+ billion cash can't build competitive models fast enough, the AI moat belongs to specialized AI companies, not integrated tech giants.

⚡ Google Wins Either Way – Google gets $1 billion annually plus Gemini running on a billion+ Apple devices. If Apple never catches up with its own models, Google locks in a permanent revenue stream. If Apple does catch up, Google still got years of payments while training on insights from Apple's use cases.

🏢 Privacy Theater Gets Exposed – Apple running Google's AI on Apple servers lets them claim privacy protection, but the fundamental AI intelligence comes from Google. That's marketing gymnastics, not technical privacy protection.

What this means for businesses:

🚀 AI Partnerships Trump Ownership – If Apple concludes licensing beats building, enterprises should reconsider whether developing proprietary AI models makes sense versus partnering with specialized providers.

💼 Temporary Solutions Become Permanent – Apple intends this as stopgap until its own models are ready, but "temporary" tech partnerships often become permanent when catching up proves harder than anticipated. Plan AI strategies assuming vendor dependencies persist longer than expected.

📊 Parameter Count Became the New Megahertz – The industry now measures AI capability by parameter count the way it once measured processors by clock speed. Google's 1.2 trillion parameters versus Apple's 150 billion makes the performance gap obvious and quantifiable.

⚖️ Privacy Compliance Gets Complicated – Running third-party AI on your infrastructure doesn't eliminate regulatory concerns about where intelligence originates. Enterprises using similar hybrid approaches should scrutinize compliance implications carefully.

The bottom line:

This deal demolishes three Apple narratives simultaneously: that Apple builds superior technology in-house, that Siri is competitive with other AI assistants, and that Apple's AI strategy is on track. Paying $1 billion annually for a competitor's AI model is an expensive admission of failure.

The spring 2026 timeline matters. Apple promised the Siri overhaul at WWDC 2024—that's nearly two years from announcement to delivery, and only by licensing Google's technology. Apple's inability to ship competitive AI on its own timeline, despite vast resources, shows how different AI development is from Apple's traditional hardware-software integration strengths.

The "we're building our own 1 trillion-parameter model" messaging is face-saving, but the challenge isn't just parameter count—it's whether Apple can attract and retain the AI talent needed when competitors offer equity in companies whose entire valuation depends on AI leadership. Apple's recent talent exodus suggests the company struggles competing for AI researchers who want to work at actual AI companies.

Google secured the deal based on price after Apple evaluated OpenAI and Anthropic, per Bloomberg. That means Apple chose the cheapest option, not necessarily the best technology. If cost was the deciding factor, it suggests Apple expects to pay for this service long enough that price differences matter—undermining claims this is temporary.

The privacy angle deserves skepticism. Yes, Gemini runs on Apple's servers. But the model's intelligence, training, and architecture come entirely from Google. Apple essentially built an expensive wrapper around Google's AI to maintain privacy theater while admitting it can't match Google's capabilities.

Your take: When Apple—famous for "it just works"—needs Google's AI to make Siri work, does that prove AI is impossibly hard, or that Apple just isn't good at it? 🤔

21 comments