苹果终于承认，Siri老了

Original Author: Sleepy

In the early morning of June 9, 2026, Beijing time, Apple's WWDC 2026 arrived as scheduled.

At the press conference, it renamed Siri to Siri AI, announced a deep cooperation with Google, using Gemini's model capabilities to train its own new generation of foundational models, and extended Private Cloud Compute to Google Cloud and Nvidia GPUs for the first time.

It released five Apple Foundation Models, with the smallest edge-side model having 3 billion parameters and the largest cloud model specifically optimized for Nvidia GPUs. Almost every daily-use App has been rewritten. Siri also gained its own standalone application, capable of saving conversations, syncing across devices, and having memory.

This is the most information-packed press conference Apple has held in years.

Taming the Future

Apple's AI story can be traced back to the autumn of 2011, when Siri first took the stage at the iPhone 4S launch event.

At that time, Steve Jobs was gravely ill, and Apple stood at the crossroads of an era. Siri was like a little thing that ran out of a sci-fi movie. You'd ask about the weather, ask for restaurants, tell it to set an alarm, and it would answer in a slightly mechanical tone. For the first time, you felt that a phone wasn't just a piece of cold glass.

Siri originated from SRI International's CALO project, a military-grade AI assistant funded by DARPA. Apple acquired it in 2010, with TechCrunch reporting the deal might have exceeded $200 million. A year later, Siri debuted with the iPhone 4S. Apple claimed it could understand natural language and act like a personal assistant.

At that moment, Apple had secured the world's best personal smart interface. Then it wasted over a decade.

Looking back today, Siri's earliest achievement was changing the way people interact with machines. In 2011, the iPhone was transforming phones from communication tools into personal computing devices. The App Store redefined software distribution, moving the mobile internet from the PC desktop into the palm of your hand. Siri emerged at the crest of a rising wave. But once inside Apple, it quickly went from an ambitious personal assistant to an obedient voice remote control.

At its core, Apple believes in closed systems and control. But a true personal assistant must integrate more services, understand more context, and tolerate more uncertainty. Uncertainty means errors, means privacy risks, means the disorder Apple is least adept at handling.

Thus, Siri was only allowed to perform deterministic tasks, like a tamed future. It had a name, a voice, a personality package, but lacked the initiative and memory essential for a true personality. Initially, users were amazed by it, then they joked about it, and later, they simply stopped using it much.

Apple was the first to put a "personal assistant" into a phone, and also the first to lock it away.

Looking at the industry's current focus on Agents, the Siri of 2011 was almost its prototype. One could say Apple was the first company to create the prototype of an Agent, yet ended up being the last to finish it.

AI That Doesn't Look Like AI

During all these years of Siri not growing up, did Apple's AI stagnate?

The answer is quite the opposite. Apple did a lot of AI, but it did it in a way that didn't look like AI at all.

Judging by the fanfare at press conferences, it seems like Apple only started seriously talking about AI in 2024. But if you trace back along the technology path, Apple has been taking action for a decade.

In 2015, it acquired two companies in a row: one to supplement natural language dialogue, another to explore running deep learning directly on phones. That same year, WWDC introduced Proactive Assistant, trying to make the system offer suggestions before the user even spoke. The idea was ahead of its time, but it seemed more like a slogan under the technological conditions of the time.

The following year, it launched SiriKit, cautiously opening a crack for developers to access Siri. It also publicly discussed Differential Privacy, signaling its intention to learn from large-scale data while protecting individual privacy. In 2017, the iPhone X brought the Neural Engine, with Face ID and cameras relying more on on-device machine learning. Apple also launched Core ML, allowing developers to run models on Apple devices, and acquired Workflow, later known as Shortcuts.

This was a very Apple-like set of answers. It wanted AI but didn't want to bet everything on the cloud and massive amounts of personal data like Google. It wanted developers but didn't want Siri to become a mishmash. So Apple chose the hardest and slowest path: focusing on edge computing, privacy, and system integration.

Around 2020, Apple acquired several more companies working on low-power edge AI and voice understanding. That same year, the M1 chip was released, bringing a 16-core Neural Engine to the Mac, pushing edge AI compute power from phones in pockets all the way to computers. The next year, Live Text and Visual Look Up were implemented, allowing text in photos to be copied directly and the camera to recognize plants and animals. More voice requests could be processed locally on the device.

Over this decade, Apple didn't launch a separate AI App, but it did make its phones smarter.

There was a reason for choosing this path. AI on a phone isn't just a question-answering machine. It needs to look at photos, listen to voice, understand contacts, invoke Apps, and perceive battery status, location, and time. It's better if it can do some things without an internet connection, and ideally, it shouldn't need to package and upload every aspect of the user's life to the cloud for every request. Apple's hardware control allows it to take this path.

But there's a deep chasm between localized intelligence and overall intelligence. Apple excels at breaking technology down into reliable components, but generative AI requires it to piece those components back together into a whole.

These components lay quietly within the system, waiting for an opportunity.

But the opportunity didn't come first. ChatGPT came first.

When ChatGPT emerged in late 2022, Apple wasn't completely unprepared. Tim Cook repeatedly emphasized on various occasions that AI and machine learning were core technologies for Apple products for years. Bloomberg also reported in 2023 that Apple had an internal Ajax large model framework and an internal Chatbot project.

However, the problem wasn't whether Apple had cards to play. The problem was that the rules of the game had changed.

ChatGPT shifted user attention from "features" to "capabilities." Users now assume phones must have AI by default, and then compare which one is better. While ChatGPT could already organize a messy set of thoughts into an email, Siri was still saying "I found these on the web."

At WWDC 2024, Apple put Apple Intelligence on the table. Writing tools, notification summaries, photo search, personalized Siri understanding, ChatGPT integration. Apple finally admitted that relying solely on its own models, at least in 2024, couldn't meet user expectations. But the promises it made ultimately failed to materialize according to the announced timeline.

Hiring Google as a Tutor

Behind the delay of Apple Intelligence wasn't just technology falling short; it was the entire Siri team's structure failing to keep up with this wave of AI.

Multiple media outlets confirmed that Apple's former AI head, John Giannandrea, stepped down, with Craig Federighi taking over the AI direction. Vision Pro head Mike Rockwell was brought in to lead the Siri team, and a large number of Siri engineers were sent to learn AI programming tools. This wasn't a dignified rotation. Internally, Apple had realized that the old people and the old pace couldn't keep up.

In January 2026, Apple and Google issued a joint statement stating that Apple would leverage Gemini technology to customize Apple Intelligence features for iPhones and other products. It was reported that Apple plans to pay Google approximately $1 billion annually to use a customized 1.2 trillion parameter Gemini model to power the Siri overhaul. Apple had also tested models from OpenAI and Anthropic but ultimately chose Google.

This was completely different from the ChatGPT integration in 2024. That time, ChatGPT seemed more like a rescue summoned by the user when Siri couldn't answer; the branding was OpenAI's, and the interface was pop-up based. This time, Gemini goes directly to the bottom layer, becoming a part of Apple's new generation of foundational models.

The key action was distillation. Google gave Apple full access to Gemini. Apple used the large model within Google's data centers to generate high-quality answers and reasoning processes, then used these results to train smaller, cheaper models capable of running on iPhones.

The technical article Apple published the day before WWDC packaged this cooperation as the third-generation Apple Foundation Models, developed in cooperation with Google across five models. On the edge, there's the 3 billion parameter AFM 3 Core, and the 20 billion parameter sparse model AFM 3 Core Advanced, which activates only parts of its parameters per request. On the cloud, there are AFM 3 Cloud, the image model ADM 3 Cloud, and the most powerful, AFM 3 Cloud Pro.

A more realistic change lies in computing power. Even the smartest edge model cannot handle all tasks, and Apple's Private Cloud Compute infrastructure alone struggles to bear the full load of Gemini-class reasoning. Some requests will run on Nvidia GPUs in Google Cloud. Apple subsequently confirmed that PCC was extended beyond Apple's own data centers for the first time, with the tech stack covering Nvidia Confidential Computing, Intel TDX, and Google Titan chips. Apple emphasized it still controls the PCC software, devices only trust programs encrypted and approved by Apple, and related binary files will be open for inspection by security researchers.

Apple hasn't truly given up control, but it has given up the pretense of complete self-reliance.

Bones Are Borrowed

To understand Apple's position in the AI era, one must first recognize its most core asset.

It's not the chip, not the model, it's the device. The device holds albums, emails, calendars, maps, and payments, carrying the fragments of many ordinary people's lives. Whichever AI can leverage these fragments isn't just a chatbot; it becomes a true personal intelligence hub.

Apple began paving the way for this hub early on. The Workflow acquired in 2017 later became Shortcuts, deeply integrated with Siri and system automation. The App Intents launched in 2022 allowed third-party apps to expose their capabilities to the system entry point. By the Apple Intelligence era, these interfaces have become the hands and feet for the AI to access real-world actions.

With these interfaces, OpenAI could come in, Gemini has come in, and the Chinese market could find local partners in the future. But they won't take over the iPhone directly. Instead, they will be contained within Apple's permission framework and privacy rules.

What Apple fears most isn't someone having a better model. It fears users starting to bypass the system and entrust their lives directly to another entry point. If, one day, users open an AI assistant that can manage everything for them instead of an App, Apple would be reduced to a nicely crafted shell.

So, from now on, the "Apple" in Apple Intelligence represents product control, but no longer complete technological sovereignty. The skin is its own, the clothes are tailored by itself, but the bones are borrowed. Google provides the skeleton, Nvidia provides the joints, and Apple's job is to dress this body in its own clothes and walk it out the door.

What Google gets from this deal is massive validation – even Apple admits Gemini's underlying capabilities are more reliable. What Nvidia gets is another proof point – even with the strongest consumer-grade chips and ambitions for self-built servers, when it comes to frontier inference and complex agent tasks, Nvidia's GPU cloud is unavoidable.

But the more bones you borrow, the less the body is truly your own. Behind every borrowed bone lies the supplier's commercial calculations, regulatory requirements, and technological cadence. If someday someone wants to pull those bones back, can Apple stand on its own? It doesn't need to answer this question just yet, but it will eventually have to.

A New Tenant Living in the System

Ordinary people don't care about model parameters. They care if their phones can bother them less.

At WWDC26, Apple said on stage: "There are times when you expect more from Siri."

For Apple, this was almost an apology.

Then it tried to show you a different morning.

You wake up, and the screen is piled with twenty notifications. In the past, you had to clear them one by one. Now, the system has already prioritized them for you. Your boss's messages are at the top, while ads and promotions are bundled into a single line of gray text. You open your email, and a long work email has already been summarized into three sentences. You decide to reply, and Siri drafts a response based on your typical tone with that person. You remember you need to call a merchant in the afternoon about a return, and before you even dial, the system has already retrieved the order number from an email you received two days ago and pasted it onto the call interface.

This is the story Apple wants to tell: a layer of intelligence underlying the system, handling the cognitive overhead you repeat every day. Read less fluff, spend less time finding files, get interrupted less by notifications.

To tell this story, Apple almost completely rebuilt Siri's entry points. On the iPhone, it's placed in the Dynamic Island, accessible by swiping down. On iPad and Mac, it's merged with Spotlight. It has its own independent App, capable of saving and continuing past conversations, syncing across devices via iCloud. Apple wants Siri to become an AI assistant living inside the system, with memory and context, while trying not to make it look like ChatGPT.

Vision is also an important direction. The camera has a new Siri mode: point it at food to get nutritional information, point it at something you don't understand to identify and search. The system-level dictation not only converts speech to text but also automatically adds punctuation and adjusts formatting, turning spoken language into text ready to be sent.

The developer side is also being paved. Apple opened up the Core AI framework, allowing third parties to load their models on the device. With the upgrade of App Intents, Siri can more easily understand third-party apps. The Foundation Models Framework no longer only calls Apple's own edge models but also supports connecting to external providers like Claude and Gemini. Apple is paving a path for the entire ecosystem. In the future, if Siri wants to do things across multiple apps, developers must expose their content and actions for the system to understand.

If these plans materialize, Apple AI will no longer be just a "chatty Siri."

However, this time Apple is much more cautious than before. Siri AI will only open to users in beta form later this year, starting with English. And the same Apple Intelligence, when it arrives in China, might not be the same product at all.

For Chinese users, watching Apple AI is mostly just for entertainment. The conference is exciting, the features look nice, but in China, it's "not supported yet."

The Chinese market has a full set of rules regarding generative AI: filing, content security, and data localization. Apple needs to find local model partners and pass regulatory approvals. Apple Intelligence in China isn't just about being a few months late; its underlying architecture might be a completely different set.

What US users see is a combination of self-developed models and Gemini. What Chinese users might see is a version blended from Apple's system permissions, local cloud services, local models, and regulatory requirements. They are both called Apple Intelligence, but their actual capabilities and reachable boundaries could be entirely different.

iCloud's services in mainland China are operated by Guizhou Cloud. The cloud disk stores files; AI needs to understand files. The cloud disk stores photos; AI needs to interpret photos. The cloud disk syncs notes; AI needs to extract your plans, habits, and interpersonal relationships from those notes. This data has entirely new uses in the AI era, naturally facing different levels of regulation.

A more realistic threat comes from competition. Domestic phone manufacturers are acting very quickly on edge-side large models, Chinese assistants, and imaging AI. For Chinese users, spending one or two thousand dollars on a new iPhone, only to find its core AI features unusable, might make them switch brands.

The daily scenarios of the Chinese market are particularly tricky for Apple. WeChat, Alipay, Meituan, Douyin (TikTok), ride-hailing apps, government services, hospital registrations – these are the real things most people use their phones for daily. An AI assistant that can't enter these scenarios, can't understand group chats, invoices, verification codes, and various expressions that only locals instantly get, will hardly be called "intelligent."

Understanding a Person

Apple Intelligence also has another problem: it doesn't cover all iPhones.

iOS 27 can cover the iPhone 11 and the second-generation iPhone SE, but Apple Intelligence requires at least an iPhone 15 Pro or newer, M-series iPads and Macs. The strongest edge models have even higher requirements: iPhone 17 Pro, iPhone Air,