Apple’s Siri Overhaul: The Billion-Dollar Gemini Bet That’s Already Late

A Billion Dollars and a Missed Deadline

On January 12, 2026, Apple and Google announced a multi-year partnership to rebuild Siri around Google’s Gemini AI technology. Under the deal, Apple will pay Google approximately $1 billion annually to license a custom 1.2 trillion parameter Gemini model — an eightfold increase over the roughly 150 billion parameter system currently powering Apple Intelligence.

The agreement makes it one of the largest AI licensing deals in history. The upgraded Siri was originally expected to debut with iOS 26.4 in spring 2026. But when Apple released iOS 26.4 on March 24, the new Siri was nowhere to be found. Internal testing had revealed critical quality problems — Siri cutting users off mid-sentence, struggling with complex multi-step requests, and exhibiting slow response times — forcing Apple to push the features to iOS 26.5 (expected May 2026) and potentially iOS 27 (September 2026).

The delay underscores a difficult truth: even with a trillion-parameter model and a billion-dollar budget, shipping a reliable AI assistant is harder than building the underlying technology.

Why Apple Went Outside

Siri’s core problem has always been architectural. Since its 2011 launch, Siri has operated as an intent-classification system: user requests are matched against a predefined set of commands. “Set a timer for 10 minutes” works reliably. “Find the email from my dentist last week and add the appointment to my calendar” does not, because it requires contextual reasoning, cross-app search, and multi-step execution that intent classifiers cannot handle.

Apple’s internal attempts to modernize Siri — including a team of several hundred engineers assembled around 2022 and the Apple Intelligence initiative announced at WWDC 2024 — produced incremental improvements (Writing Tools, email summaries, notification prioritization) but never delivered the fundamental architectural leap needed to compete with GPT-4-class and later models.

The build-versus-buy calculation ultimately favored buying. Training a frontier model from scratch would require tens of thousands of high-end GPUs running for months. Google has already made that investment for Gemini. Apple’s licensing fee buys access to a model Google spent several times more developing, while letting Apple focus its AI resources on on-device inference, privacy engineering, and hardware-software integration — areas where it has genuine competitive advantage.

Apple evaluated multiple vendors before selecting Google. Notably, Anthropic’s Claude was reportedly in contention to power the new Siri, but Anthropic’s pricing demands — reportedly several billion dollars annually, doubling each year — made the deal untenable. Apple also maintains its existing ChatGPT integration in Siri (introduced in iOS 18.2), which continues alongside the Gemini partnership.

What the New Siri Promises

On-Screen Awareness

The most transformative planned capability is on-screen context awareness — Siri understanding what is currently displayed on the user’s screen and incorporating that context into responses and actions. A user looking at a restaurant webpage could say “make a reservation here for Saturday at 7” and Siri would identify the restaurant, locate its booking system, and initiate the reservation.

The implementation relies on Gemini’s multimodal capabilities, processing a structured semantic description of visible UI elements rather than raw screenshots. On-screen awareness is designed to work across all native Apple apps and third-party apps using standard iOS UI frameworks.

Multi-Step Action Chains

Current Siri can execute single actions or, in limited cases, two linked actions. The Gemini-powered Siri is designed to support chains of up to 10 sequential actions, each informed by the results of previous steps. A request like “Check my calendar for next week, find a free afternoon, draft a meeting email to my team, and attach the project brief” would involve at least five distinct actions executed in sequence.

Each step includes an implicit checkpoint — if any action produces an unexpected result, Siri pauses and asks the user how to proceed rather than continuing blindly. However, it is precisely this multi-step reliability that has proven problematic in internal testing, contributing to the iOS 26.4 delay.

Privacy Architecture

The 1.2 trillion parameter model is too large to run on-device. Apple’s solution is Apple Private Cloud Compute (PCC) — secure cloud infrastructure using Apple Silicon servers where complex queries are processed in encrypted, isolated environments. Google never sees raw user queries or personal data; the Gemini model runs inside Apple’s infrastructure.

Simple requests — timers, phone calls, basic questions — still process entirely on-device using Apple’s smaller proprietary models. The Gemini-powered PCC path activates only for complex tasks.

That said, reports have emerged that Apple is also exploring running some Siri workloads on Google’s own servers, suggesting the PCC infrastructure alone may not be sufficient to handle demand at scale.

What Google Gets

The financial terms — roughly $1 billion per year — are significant but secondary to distribution. Apple’s 2.5 billion active devices represent the largest single distribution channel any AI model has ever achieved. For Google, which has struggled to convert Gemini’s technical capabilities into consumer adoption beyond Search, Apple’s ecosystem solves the go-to-market problem that engineering alone cannot.

The deal also validates Gemini’s technical quality. Apple choosing Google’s model over building its own — or licensing from OpenAI or Anthropic — signals that Gemini is competitive for this specific use case. That Apple, a company that publicly committed to on-device AI, concluded Google’s model was worth a billion dollars annually sends a market signal no marketing campaign could replicate.

The Strategic Risks

Dependency and Delay

Apple is now dependent on a direct competitor for its flagship AI feature’s core intelligence. The multi-year contract provides some protection, but the strategic vulnerability is real. Apple is reportedly continuing to invest in its own foundation model research, viewing the Gemini partnership as a bridge while it develops an in-house alternative.

The iOS 26.4 delay adds another dimension to this risk. Each month without the upgraded Siri erodes the deal’s value proposition and gives competitors like OpenAI (through its own consumer products) and Google Assistant (with native Gemini integration) more time to strengthen their positions.

The Expectation Gap

When the new Siri eventually ships, it will be dramatically more capable than the current version — and dramatically less capable than many users expect. The gap between “can chain 10 actions and understand your screen” and “can do everything a human assistant can” is enormous. Managing expectations may prove Apple’s hardest challenge: a Siri that is ten times better but still fails at tasks users believe it should handle risks feeling frustrating rather than impressive.

The Industry Signal

The Apple-Google deal represents a maturation of the AI industry’s business model. A small number of companies — Google, OpenAI, Anthropic, and Chinese labs like DeepSeek and Baidu — build the foundation models, while a larger ecosystem integrates those models into products, devices, and services.

Apple, the world’s most valuable company, choosing to license rather than build sends a clear signal that this structure is not just viable but optimal. The cost of building a frontier model is so high, and the required expertise so specialized, that even trillion-dollar companies rationally choose to buy rather than build.

For the billions of users waiting on the new Siri, the provenance of the underlying model is irrelevant. What matters is whether the upgrade eventually delivers — and right now, “eventually” is doing a lot of work in that sentence.

Sources & Further Reading

• Apple picks Google's Gemini to run AI-powered Siri coming this year — CNBC
• Google's Gemini to power Apple's AI features like Siri — TechCrunch
• Apple's iOS 26.4 Siri Update Runs Into Snags in Internal Testing — Bloomberg
• No new Siri, but iOS 26.4 is here with a ton of must-have new features — Macworld
• Apple confirms Gemini-powered Siri will use Private Cloud Compute — 9to5Mac
• Apple could have used Claude to power a future Siri, but Anthropic got greedy — AppleInsider
• Apple Explains How Gemini-Powered Siri Will Work — MacRumors