Apple's Siri Reboot: A $1 Billion Bet on Google's AI to Fix Its Most Embarrassing Product

For nearly two years, Apple has been promising a smarter Siri. On March 24, 2026, Bloomberg's Mark Gurman reported what that actually looks like: a standalone app, a chatbot-style interface, and a systemwide "Ask Siri" button—all built on top of Google's Gemini AI models [1]. The overhaul, codenamed "Campos" internally, is expected to debut at Apple's Worldwide Developers Conference on June 8 as part of iOS 27 and macOS 27 [2].

The announcement caps a string of delays stretching back to WWDC 2024, when Apple first promised an AI-powered Siri that could understand context, act across apps, and use personal data to answer queries. That version was initially slated for spring 2025, then pushed to early 2026, and delayed again [3]. The new Siri is now promised before the end of 2026 [1].

What the New Siri Actually Does

The redesigned Siri abandons the glowing-edge animation introduced in iOS 18. In testing, a new interface places Siri within the Dynamic Island at the top of the screen, prompting users to "Search or Ask." While processing, a pill-shaped "Searching" indicator appears alongside a glowing Siri icon. Results expand into a translucent panel using Apple's Liquid Glass design language [2].

The most significant functional change is the shift to a conversational, chatbot-style experience. Like ChatGPT or Claude, the new Siri will support both text and voice input, with conversations displayed in iMessage-style chat bubbles [4]. Apple is also testing a systemwide "Ask Siri" toggle that appears in menus across built-in apps—if you're reading something in Safari or Mail, you can send that content directly into a Siri conversation without leaving the app [5].

Capabilities reportedly under development include web search, image generation, article summarization, and file analysis—features that put Siri in direct competition with standalone AI assistants from OpenAI, Google, and Anthropic [6]. Siri will also gain the long-delayed ability to use personal data and on-screen context to answer queries, and to perform actions across multiple apps [1].

The Google Gemini Deal

Perhaps the most consequential detail is what powers the new Siri under the hood. In January 2026, Apple and Google announced a multi-year partnership under which Apple's next generation of foundation models will be based on Google's Gemini technology and cloud infrastructure [7]. Apple is paying approximately $1 billion per year for access [8].

The deal came after OpenAI reportedly declined a similar arrangement to power Siri [9]. The partnership represents a striking departure for Apple, which has historically developed core technologies in-house. Google's Gemini will power not only Siri but a range of future Apple Intelligence features [10].

This raises a fundamental question about differentiation. If the intelligence behind Apple's AI assistant comes from the same company that builds Android's competing assistant, what makes Siri distinctive? Apple's answer, predictably, centers on integration and privacy—Gemini provides the underlying model capability, but Apple controls the user experience, on-device processing, and data handling [7].

The Spending Gap

Apple's approach to AI investment diverges sharply from its peers. The company spent approximately $12.7 billion on total capital expenditure in fiscal 2025—roughly 3% of revenue, a ratio it has maintained for five years [11]. Compare that to the competition:

Source: CNBC / Company earnings reports

Data as of Mar 25, 2026CSV

In 2025, Amazon, Microsoft, Meta, and Alphabet collectively spent approximately $400 billion on AI infrastructure [12]. For 2026, those four companies plan to spend roughly $650 billion—a 60% year-over-year increase [13]. Alphabet alone has guided between $91 billion and $93 billion in capex for 2025 [12], and up to $185 billion for 2026 [13]. Apple's entire capital expenditure is a rounding error by comparison.

Apple CEO Tim Cook and CFO Kevan Parekh have said they expect capex to grow "substantially," driven by investments in Private Cloud Compute and first-party data centers [11]. But "substantially" from a $12.7 billion base still leaves Apple spending an order of magnitude less than its rivals.

The strategic logic, as articulated by analysts at Fortune, is that Apple is betting AI models will become interchangeable commodities—and that the durable competitive advantage lies in hardware integration and user experience rather than model development [14]. The Gemini deal supports this thesis: rather than spending tens of billions to build a competitive large language model from scratch, Apple is licensing one.

Who Gets Left Out

Apple Intelligence currently requires an A17 Pro chip or newer on iPhones (iPhone 15 Pro and later), an M1 chip or newer on iPads and Macs, and a minimum of 8 GB of RAM [15]. This excludes the base iPhone 15, all iPhone 14 models and earlier, and every Intel-based Mac.

Apple reported 2.5 billion active devices worldwide as of January 2026, up from 2.35 billion a year earlier [16]. No public breakdown exists of how many of those devices meet the Apple Intelligence hardware threshold. However, given that the A17 Pro debuted only in the iPhone 15 Pro lineup in September 2023, and older iPhones constitute a large share of the active installed base—particularly in developing markets—it is likely that the majority of Apple's 2.5 billion devices cannot run Apple Intelligence features.

Apple Intelligence also remains unavailable on devices purchased in mainland China or using an Apple Account set to mainland China [15], a significant exclusion given China's importance to Apple's revenue.

Whether iOS 27 will impose additional hardware requirements beyond the current Apple Intelligence baseline has not been confirmed. If the new Siri app requires more processing power for on-device inference, the exclusion list could grow.

Privacy: The Tightrope

Apple's privacy positioning faces new tension with an AI assistant that relies on Google's cloud infrastructure. Apple's stated architecture processes requests on-device whenever possible—searching through Messages, Notes, or providing widget suggestions happens locally [17]. When cloud processing is required, Apple uses Private Cloud Compute, which the company says does not store user data and only processes it to fulfill the specific request [18].

Under the Gemini partnership, Apple says it will maintain its privacy standards, with Apple Intelligence continuing to run on Apple devices and Private Cloud Compute [7]. But the details of what data flows to Google's infrastructure, and under what conditions, have not been fully disclosed.

Apple's privacy record is not unblemished. Research presented at Black Hat USA 2025 found that Siri sent the content of dictated messages and commands—including WhatsApp communications—to Apple servers even when transmission wasn't necessary to complete user requests [19]. Apple retains request data for up to two years to improve Siri and related services, using random identifiers rather than Apple Account information [20].

The introduction of an "Apple Intelligence Pro" subscription tier at $15 per month, which unlocks advanced agentic capabilities such as multi-step cross-app actions [21], adds another dimension. Premium AI features that require more intensive cloud processing may involve different data handling than basic on-device functions. Apple has not clarified whether Pro-tier queries will have different privacy characteristics.

The Developer Problem

Apple's relationship with third-party developers on Siri integration has been rocky. SiriKit, introduced to let apps work with Siri, initially expanded to support messaging, payments, and other domains. But starting with iOS 15 in 2021, Apple pulled back, deprecating several third-party Siri APIs and reducing the types of voice commands available to external apps [22]. The company pushed developers toward the Shortcuts framework instead.

The standalone Siri app introduces fresh uncertainty. Apple has not detailed how existing SiriKit integrations and Shortcuts will work with the new architecture. If "Campos" represents a ground-up rebuild—as the codename and standalone app approach suggest—there's a real possibility that current integrations will need to be rewritten.

More broadly, a capable Siri chatbot that can summarize articles, generate images, search the web, and analyze files competes directly with the AI apps that currently generate substantial App Store revenue for Apple. OpenAI's ChatGPT alone accounts for approximately 75% of generative AI subscription revenue on the App Store, and Apple is projected to collect over $1 billion from rival AI apps in 2026 [23]. A fully capable built-in Siri could cannibalize that revenue stream—though Apple may calculate that keeping users within its ecosystem is worth more than App Store commissions.

How Siri Stacks Up

Independent benchmarks paint a mixed picture of voice assistant accuracy. In controlled conditions, Google Assistant leads at roughly 95.8% accuracy, with Siri at 95.2% and Alexa at 94.7% [24]. But these figures measure voice recognition, not the kind of complex reasoning and task completion that defines modern AI assistants.

On those more demanding benchmarks—multi-step reasoning, contextual follow-ups, creative generation—Siri has lagged significantly behind ChatGPT and Gemini. The Gemini partnership is explicitly intended to close this gap. Whether it succeeds will depend on how effectively Apple integrates Gemini's capabilities while maintaining its privacy commitments and user experience standards.

Amazon has taken a parallel path with Alexa Plus, an AI-powered upgrade that makes Alexa more conversational and capable of handling complex multi-step requests [24]. The voice assistant market is converging on similar capabilities, which reinforces Apple's commodity thesis—but also raises the stakes for execution.

The Monetization Question

Apple's AI monetization strategy operates on multiple levels. At the base tier, Apple Intelligence features ship free with compatible devices, functioning as a hardware sales driver. The Apple Intelligence Pro subscription at $15 per month [21] targets users willing to pay for advanced capabilities—priced slightly below ChatGPT Plus and Microsoft Copilot Pro at $20 per month.

Analysts at Wedbush have suggested that AI monetization could add $75 to $100 per share to Apple's valuation over the coming years [25]. The bull case rests on Apple's unique position: 2.5 billion active devices, deep hardware-software integration, and the ability to offer an AI assistant that works seamlessly across iPhone, iPad, Mac, Apple Watch, and Vision Pro.

The bear case is that Apple is late, dependent on a competitor's technology, and asking users to pay for capabilities that ChatGPT and Gemini already offer as standalone products. The $1 billion annual Gemini licensing cost also represents a recurring expense that grows with usage, unlike a one-time infrastructure investment.

Source: GDELT Project

Data as of Mar 25, 2026CSV

What This Means

Apple's Siri overhaul is the company's highest-stakes product bet since the Apple Watch. The two-year delay from initial announcement to expected delivery has eroded credibility. The reliance on Google's Gemini models—while pragmatic—undercuts Apple's narrative of technological self-sufficiency. And the hardware restrictions mean the new Siri will be unavailable to a large share of Apple's installed base.

But Apple retains significant advantages. No other company can ship an AI assistant pre-installed on over a billion phones, integrated with the operating system, the app ecosystem, and the full suite of first-party apps. If the new Siri works well—genuinely understanding context, executing multi-step tasks, and respecting privacy—the distribution advantage alone could make it the most widely used AI assistant in the world.

The June 8 WWDC keynote will determine whether "Campos" represents a genuine leap forward or another round of promises. After two years of delays, Apple has run out of room for the latter.