TheAI Academy
中文

Apple Unveils New Siri Powered by Google Gemini at WWDC, Introduces Co-Developed "Apple Foundation Models" for iOS 27

By TheAI Academy · Jun 11, 2026

The WWDC 2026 conference kicked off on June 8 at Apple Park, where Apple introduced a completely rewritten Siri and the next generation of Apple Intelligence. In a rare acknowledgment, the company revealed that this system is built on "Apple Foundation Models co-developed with Google, utilizing Gemini technology," which is embedded in the Dynamic Island and features a standalone Siri App. According to multiple media reports, Apple pays Google approximately $1 billion annually for a customized Gemini model with around 1.2 billion parameters. What does this mean for users and developers in Taiwan?

Last Monday night, I was glued to my computer, watching the WWDC live stream with moderate expectations - after all, Apple had been all talk and no action on AI for the past two years. But as soon as Craig Federighi started speaking, I put down my coffee: he almost seemed to be saying "we know you're going to ask" as he directly brought up Google's name on the big screen at Apple Park.

For a company that's used to touting "our own technology," Apple's acknowledgment that the new Siri is backed by Google is a significant shift. The WWDC 2026, which kicked off on June 8th, Taiwan time, is Tim Cook's last keynote as CEO (Apple has already announced the upcoming handover), and the fact that he chose to unveil a Siri co-developed with arch-rival Google at this juncture is quite a spectacle.

Event Background

In the current wave of generative AI, Apple is widely recognized as a laggard. As ChatGPT, Gemini, and Claude have successively advanced conversational AI to the point where it can write code, look up data, and plan trips, Siri has remained stuck in the "can't understand human language" phase, earning the ire of users for years. Last year's promised "smarter Siri" was repeatedly delayed, and there were rumors of internal team restructuring.

So, the core of this WWDC is essentially Apple's decision to stop trying to go it alone and instead leverage external forces. According to the official statement, the new Apple Intelligence and Siri are built on "Apple Foundation Models co-developed with Google," utilizing Gemini technology and running on both device-side (on-device) and Apple's private cloud computing (Private Cloud Compute). The former prioritizes real-time performance and privacy, while the latter handles complex tasks that require larger models - this "device plus private cloud" hybrid architecture is consistent with Apple's longstanding narrative on privacy, with the only difference being the use of Google's technology as the engine.

Key Highlights

  • Siri completely rewritten, engine replaced with Gemini: The official statement clearly indicates that the new Siri and Apple Intelligence run on "Apple Foundation Models co-developed with Google, utilizing Gemini technology," which balances device-side and private cloud computing.
  • Siri integrated into Dynamic Island: The new Siri is directly embedded in the iPhone's Dynamic Island, enabling more instantaneous interactions; a standalone Siri App is also launched, with conversation records synced across devices via iCloud.
  • Operating systems upgraded to version 27: iOS 27, iPadOS 27, watchOS 27, tvOS 27, visionOS 27, and macOS 27 (codenamed "Golden Gate") are all unveiled.
  • Tim Cook's last WWDC keynote: Apple has already announced the upcoming CEO handover, making this keynote a significant passing of the torch.
  • Annual bill of $10 billion (according to foreign media reports): According to multiple foreign media reports, Apple pays Google around $10 billion annually for a customized Gemini model with approximately 1.2 trillion parameters. This figure is not mentioned in Apple's official statement, so it should be treated as market information.

Market Impact Analysis

Taiwanese users: The most direct benefit is that Siri finally has a chance to become a usable assistant. In the past, when I asked Siri slightly more complex questions, I would often get "I found these on the web" followed by a list of links, which was slower than just using Perplexity myself. With the upgrade to a Gemini-level engine, Siri should theoretically be able to understand more colloquial and jumping Chinese commands, and multi-turn conversations won't be cut off after just a few sentences. However, I must pour some cold water: support for Traditional Chinese and local Taiwanese contexts (addresses, stores, phonetic input) has always been the last to arrive and most frequently compromised in Apple's AI features. It's recommended that users wait and see the actual online performance before getting too excited about the keynote's English demo. To accumulate useful prompts, it's still best to practice with prompt words in your daily life.

Enterprise applications: For Taiwanese enterprises, the signal sent by this event is more important than the functionality itself. The fact that a company like Apple chooses to "not insist on self-research and instead directly purchases the strongest external model" sets an example for many companies that are still hesitating about whether to develop their own models - in most cases, spending money on integrating mature models and focusing on in-house data and processes is more cost-effective than training from scratch. For companies that intend to introduce conversational customer service or internal assistants, this also serves as a reminder: the underlying model will change and upgrade, and what's truly important is to bind your data and workflow, not a specific model. To explore how to implement this, you can refer to our compiled AI implementation tasks and enterprise AI chatbot guide.

Developers: For iOS developers, the key point is whether Apple will open up the capabilities of this basic model to third-party apps through APIs. If App Intents and the Foundation Models framework allow developers to directly call device-side models, it's equivalent to giving every app a free AI engine, which is a significant boon for independent developers and small teams in Taiwan. However, it's also necessary to pay attention to the costs and quotas associated with running on private cloud computing, which Apple has not yet clarified. Developers who are used to writing programs with Cursor and habitually stringing together various model APIs can put this architecture on their watchlist and re-evaluate it when the developer documentation is released.

Future Development Trends

I think the most memorable aspect of this WWDC is not any specific feature, but rather the shift in Apple's mindset. A company that's most concerned with "controlling the entire stack" is willing to publicly acknowledge that its core AI relies on a rival's support, which indicates that even Apple recognizes that "it can't catch up, so it's better to buy the best." It can be expected that in the future, the division of labor between major manufacturers will become increasingly clear, with "you do hardware and channels, and I do models" becoming a more common arrangement, and pure self-research no longer being the only politically correct approach.

The second trend is that AI assistants are transitioning from "a feature within an app" to "a fundamental layer of the operating system." Apple's efforts to integrate Siri into Dynamic Island, launch a standalone Siri App, and enable cross-device synchronization demonstrate a clear intention to make AI the default entry point for using an iPhone, rather than just a tool that's occasionally summoned. This is the same logic behind Google's decision to embed AI in its search bar - whoever controls the "first place you ask" controls the traffic.

Thirdly, this $10 billion annual cooperation also carries variables. Apple and Google are already embroiled in antitrust lawsuits over search, and now they're deeply intertwined in AI; it's worth tracking how regulatory agencies will view this. From a technical standpoint, whether Apple can successfully and stably implement Chinese language support and local experiences is the true point of concern for Taiwanese users.

TheAI Academy Summary and Commentary

Honestly, my attitude towards this new Siri is "cautiously optimistic." I'm optimistic because Apple has finally put aside its pride and taken the most practical approach to addressing its biggest weakness; I'm cautious because Apple's historical record on Chinese language support and localization is not exactly confidence-inspiring. No matter how impressive the English demo is, it's essential to wait until Taiwanese users can get satisfactory answers in Traditional Chinese before getting too excited.

Commentary: Apple's biggest breakthrough this time is not technological, but rather an acknowledgment - when even the most proud company chooses to buy the strongest model, the question of "whether to self-research or purchase" has essentially been answered for most people.

Our specific advice for Taiwanese readers: don't rush to upgrade your device just for the new Siri. After iOS 27 and the Chinese language experience are officially released, spend a few minutes testing a few scenarios you commonly ask Siri about (setting alarms, checking traffic, writing messages, multi-turn questioning), and then decide where it fits in your workflow. The true amplifier of AI benefits is not which phone you use, but rather how well you've mastered the tools and prompts at your disposal.

Sources

This article is compiled based on publicly available information and should be considered accurate as of the official announcement. The figures of approximately $10 billion annually and 1.2 trillion parameters are reported by multiple foreign media outlets and are not officially disclosed by Apple, so they should be treated as market information.

Frequently Asked Questions

Is the new Siri really using Google's technology?

Yes, and it's officially acknowledged by Apple. According to the company, the next-generation Apple Intelligence and Siri are built on "Apple Foundation Models co-developed with Google, utilizing Gemini technology," which runs on both devices and Apple's private cloud computing (Private Cloud Compute). In other words, Apple controls the interface and privacy architecture, while the model's core is powered by Google's Gemini technology.

Is it true that Apple pays Google $1 billion annually?

This needs to be taken with caution. While multiple media outlets report that Apple pays Google around $1 billion annually for a customized Gemini model with approximately 1.2 billion parameters, this figure is not mentioned in Apple's official press releases. Therefore, it should be considered market information for reference, rather than officially confirmed data.

Will the new Siri in iOS 27 provide a good experience for Traditional Chinese users?

It's still uncertain. Apple's AI features have historically been slower to support Traditional Chinese and local Taiwanese contexts (such as addresses, stores, and phonetic notation). It's recommended to wait for the official release of iOS 27 and test the Chinese experience in various scenarios before making a judgment, rather than just relying on the English demo shown in the keynote.

What are the other key takeaways from WWDC this year?

The operating systems have been updated to version 27, including iOS 27, iPadOS 27, watchOS 27, tvOS 27, visionOS 27, and macOS 27, codenamed Golden Gate. Siri is now integrated into the Dynamic Island and has a standalone App, with conversations that can be synced across devices via iCloud. This is also the last WWDC keynote speech by Tim Cook, as Apple has announced that a CEO transition is imminent.

繁體中文版 →