Apple’s Next Siri: Inside the Reported $1 Billion Custom Gemini Deal

A new report suggests Apple is preparing one of its boldest AI pivots yet: licensing a custom-built version of Google’s Gemini to supercharge Siri. According to the reporting, Apple is negotiating a roughly $1 billion-per-year agreement for access to a tailored 1.2 trillion-parameter large language model. If finalized, the move would mark a strikingly pragmatic partnership between two fierce competitors – and a decisive attempt by Apple to close the perceived capability gap between Siri and newer AI assistants.

Here’s what that figure and that model size really mean, why Apple is making this bet now, how it fits with the company’s privacy-first architecture, and what everyday users should expect when the upgraded Siri finally arrives.

What a 1.2 Trillion-Parameter Model Buys Apple

Parameter counts aren’t the only measure of intelligence, but they do correlate with the breadth of patterns a model can learn. Apple’s cloud side of “Apple Intelligence” reportedly tops out near 150 billion parameters today – solid for single-turn tasks and short summaries, but much less capable for extended reasoning, multi-step planning, and complex context juggling. A 1.2T-parameter Gemini variant sits in a different league. In practice, that scale can enable richer memory within a conversation, more reliable tool use, and improved grounding when orchestrating actions across apps.

In concrete terms, the bigger model should be better at turning a messy request into a precise execution plan. Ask future Siri to “rebook my missed flight, find a 24-hour pharmacy near my hotel, and text my team a new arrival time,” and the assistant must parse multiple constraints, reach into calendars and messages (with permission), and maintain a coherent state between steps. That kind of compound task is exactly where today’s mainstream assistants struggle and where very large models begin to shine.

Why Google – and Why Now

Apple reportedly evaluated multiple partners – including OpenAI and Anthropic – before zeroing in on Google’s Gemini family. The calculus looks straightforward: Apple needs a top-tier, general-purpose reasoning engine today while its own long-range model matures. By paying for a tuned Gemini, Apple buys time, capability, and optionality without permanently hitching its identity to an external brain.

Crucially, Apple’s plan is framed as transitional, not permanent. Internally, the company is said to be developing a roughly 1 trillion-parameter, cloud-based model of its own, with ambitions to put it into consumer-facing use as soon as next year. Whether that aggressive timetable holds is an open question – training and serving models at this scale require relentless spending, talent, and infrastructure – but the intent is unmistakable: license what’s best now, build what’s best next.

Privacy Architecture: Gemini’s Brains, Apple’s House

Licensing Google’s model doesn’t mean Apple will run your queries on Google’s machines. On the contrary, the report indicates the custom Gemini will execute on Apple’s Private Cloud Compute (PCC) infrastructure. PCC is Apple’s answer to a hard problem: offer cloud-scale AI while preserving the company’s privacy promises. Requests are processed on Apple-managed servers hardened to minimize data exposure, and the system is engineered so that sensitive content isn’t casually retained or repurposed to train generalized models.

This split architecture – some tasks on device, some in PCC – allows Apple to keep a tight ring around user data while still tapping state-of-the-art AI. Expect Apple to keep certain capabilities (especially those touching local content or device control) on its own in-house models, reserving the heavyweight third-party model for deep reasoning, cross-app planning, and complex synthesis.

Timelines, Codenames, and Who’s in Charge

Inside Apple, the effort to use a third-party model to lift Siri reportedly carries the codename Glenwood, while the long-anticipated “Personal Siri” release has been repeatedly adjusted, now rumored to target an iOS 26.4 window in the spring under the codename Linwood. The project is steered by Mike Rockwell – the executive who led Vision Pro – and Craig Federighi, Apple’s longtime software chief. Their mandate is to turn Siri from a narrowly scoped voice interface into a truly capable, context-aware assistant that understands what you mean, not just what you say.

How a New Siri Could Actually Feel Different

In its best form, the Gemini-powered Siri should deliver three practical upgrades:

• Richer understanding of intent: Better at following multi-step instructions and asking the right clarifying question only when needed.
• Cross-app orchestration: Scheduling, messaging, reminders, file handling, and navigation chained together without the user micromanaging each step.
• Longer, steadier context: Sustained conversations that remember prior details within a session and keep plans aligned as you change constraints.

If you’ve used Gemini on Android – some people even replace Google Assistant with Gemini on devices like the Pixel 6 Pro – you’ve seen glimpses of that deeper reasoning: more complete answers, better synthesis, fewer dead ends. The Apple twist is the company’s tight integration and privacy guardrails, which could make these capabilities feel less like a chatbot bolted on top and more like a native layer across the OS.

The Money and the Precedent

The reported $1 billion annual fee is eye-popping but not unprecedented in the context of Apple–Google deals. The two companies already have a massive search partnership under which Google effectively rents default status inside Safari; in some years that arrangement has delivered tens of billions to Apple. By comparison, the AI licensing fee is smaller but strategically sensitive – Apple gets cutting-edge capability while retaining brand control and privacy posture; Google deepens platform reach for its flagship model without owning the end-user relationship.

Don’t expect Apple to plaster “Powered by Gemini” labels across iPhone marketing. Apple will almost certainly present the experience as “Siri, upgraded,” not “Siri, outsourced.” The delineation matters: the software brains may be sourced, but the trust and presentation must remain distinctly Apple.

What It Means for Developers

If Apple executes, developers should gain a more reliable Siri for intents and actions, with fewer brittle command grammars and better disambiguation. Apple could pair the new assistant with expanded APIs so apps can expose structured capabilities the model can safely invoke. That would turn Siri from a voice veneer into a genuine runtime for “AI actions,” enabling flows like: summarize a PDF in Files, draft the recap in Notes, attach it to Mail, and schedule follow-up in Reminders – without writing a tangle of one-off shortcuts.

Risks and Open Questions

Large models still hallucinate. They can overconfidently invent citations, misinterpret rare edge cases, and struggle with strict correctness in domains like finance or health. Apple’s challenge is to bound that behavior with system prompts, tool constraints, and UI that keeps the user in control. There’s also the question of reliability and cost at scale: serving trillion-parameter-class models is expensive, and Apple must keep latency low enough that Siri feels instantaneous, not academic.

Another variable is talent. The report notes Apple has seen notable AI staff churn in recent years. Shipping a first-party 1T-parameter model next year would be a tour de force; even if it slips, the Gemini bridge gives Apple a viable path to ship meaningful upgrades sooner.

The Bottom Line

Apple appears ready to rent the best AI it can get while building the best AI it can own. A custom 1.2T-parameter Gemini running inside Apple’s Private Cloud Compute could make Siri meaningfully more capable at planning, reasoning, and executing across apps. If Apple follows through with a robust developer surface and clear privacy guarantees, the next Siri won’t just answer questions – it will finish tasks.

And if Apple’s own 1T-class model lands on schedule, the company can swap the engine without changing the car. Until then, a licensed Gemini gives Apple what it needs most: time, capability, and a credible way to make Siri feel modern again.