"Wrapping" Google's Gemini, but Apple hasn't given up on developing its own model yet

By Xiao Jinya, Alphabet AI

Apple has "bowed down."

On January 12 local time, Apple and Google issued a joint statement:

"After careful evaluation, Apple believes Google's artificial intelligence technology provides the strongest foundation for Apple’s base models, and we are excited about the innovative experiences it will unlock for Apple users."

The two companies said these models will help support future Apple Intelligence features, including a more personalized Siri coming later this year.

As soon as the news broke, Elon Musk was the first to express dissatisfaction, criticizing on X that such "concentration of power" is unreasonable—especially since Google owns both Android and the Chrome browser.

This implies that Google now controls even Apple's AI. But Apple would certainly never agree with that.

For Apple, this is a temporary compromise—not surrender or giving up.

Bringing in Google's Gemini is more like hiring a helper during a transition period. It won't fully replace Apple’s self-developed model. And Apple is still developing its own trillion-parameter model.

Gemini Is Not Replacing Apple's Self-Developed Model

Public details from Apple and Google are limited and deliberately vague, but one thing is clear: Gemini will be involved in building Apple’s next-generation base model.

Back in November last year, Bloomberg’s Mark Gurman reported this already, citing insiders saying Apple would pay Google around $1 billion annually.

However, this isn’t a simple “one-for-one” swap where Gemini directly replaces Apple’s base model. Instead, it’s a system architecture allowing Apple to continue using its own models while sharing no user data with Google.

As initially reported, the Gemini model won’t be directly embedded into Apple’s operating system.

Rather, everything will still appear externally as powered by Apple’s base model, but at the underlying level, Gemini will serve as its “foundation.” In other words, Apple’s use of Gemini sits somewhere between “repackaging” and “knowledge distillation”—and of course, legal under mutual agreement.

This point is crucial: interpreting this news as “Apple’s AI will be driven by Gemini” is inaccurate.

When you buy an iPhone, it contains chips made by TSMC, displays from Samsung, and various components sourced globally. Yet no one claims Samsung “drives” the iPhone.

On-device Apple Intelligence will still be supported by Apple’s own base model.

In short, the Gemini model will act as a tool to help train Apple’s base model, enabling it to better perform tasks defined by Apple.

Gurman previously reported that under the Apple-Google agreement, Google’s Gemini model will handle summarization and planning functions for Siri—components that help the voice assistant synthesize information and decide how to execute complex tasks. Some Siri features will continue using Apple’s proprietary models.

The report also noted the model will run on Apple’s private cloud servers, ensuring complete isolation of user data from Google’s infrastructure. Apple has already allocated dedicated AI server hardware to support this.

Given the lack of public rebuttals from either side, these reports remain unchallenged.

This deal is fundamentally different from collaborations like the “Google Search” partnership because it’s entirely invisible to end users. During interactions with Siri, there will be no indication of Gemini—and there shouldn’t be.

Apple Hasn’t Given Up Yet

Further collaboration with Google is a transitional move for Apple.

Since 2024, Apple users have been eagerly awaiting the new Siri, which was previewed at WWDC with promises of gradual rollout. But repeated delays turned this into a major embarrassment for Apple.

Currently, multiple media outlets report the new Siri will launch alongside iOS 18.4 in March.

Beyond that, Apple Intelligence itself hasn’t progressed smoothly. Early features like “news summarization” faced criticism due to errors.

Worse yet, when Meta suddenly launched an “AI talent war” last year, Apple suffered heavy losses. Most notably, Ruoming Pang, who led Apple’s base model team, was poached—reportedly offered a staggering $200 million compensation package by Meta.

Under these circumstances, further postponement wouldn’t just mean falling behind in the AI race—it would severely damage market confidence.

At this moment, turning to third parties to push forward critical initiatives (like the new Siri) on schedule and deliver tangible results to users has become Apple’s top priority.

Choosing Google isn’t surprising.

Apple and Google have a long-standing relationship—for years, Google has paid Apple over $20 billion annually to be the default search engine in Safari.

Their cooperation has drawn antitrust lawsuits, but recent rulings only adjusted specific practices—the $20 billion deal remains intact.

Moreover, Gemini is currently among the strongest models available. According to prior reports, Apple tested several leading models including Gemini, GPT, and Claude, ultimately selecting Gemini.

As a side note, Apple’s emphasis on “user privacy” in this Gemini collaboration—highlighting operations within Apple’s private cloud without sharing any data with Google—is informed by past lessons. In 2012, a “Safari tracking cookie scandal” erupted, with Google accused of exploiting Safari’s privacy settings loophole to place tracking cookies, bypassing policies blocking third-party cookies. Google was eventually fined $22.5 million by the FTC and required to reform. Additionally, it reached a $17 million settlement with 37 U.S. states and Washington D.C. to resolve class-action litigation.

In summary, Apple’s deepened collaboration with Google can be seen as a temporary “bow,” but absolutely not a surrender or abandonment.

Apple continues developing its own trillion-parameter model, expected to debut around 2027.

Even after that model launches, Apple may still retain Gemini for enhanced training, benchmarking, and optimization. Using Gemini as a “tool” to strengthen itself—that’s Apple’s true expectation.

The Biggest Winner Is Google

Although we’ve spent considerable space clarifying what this deeper collaboration between Apple and Google is—and isn’t—the perception that “Apple’s AI relies on Gemini” will likely persist.

One can easily imagine: What if, by March or April, Siri suddenly becomes incredibly capable…?

It’s not hard to envision. Apple now faces an awkward reality: If bringing in Gemini delivers immediate, noticeable improvements to Apple Intelligence and Siri, then the narrative that “Apple simply couldn’t do it alone” becomes cemented.

Even if Apple successfully develops its trillion-parameter model later and transitions to full independence, Gemini may still be credited as the key contributor—not Apple itself.

As for Google, the market has already responded positively.

After the joint announcement, Alphabet’s stock rose as much as 1.7%, closing the day with its market value surpassing $4 trillion for the first time—a company record—while Apple’s shares gained less than 1%.

This partnership also puts pressure on others in the industry.

Elon Musk, who voiced objections, criticized Google—controller of both Android and Chrome—for now penetrating Apple—as evidence of excessive “power concentration,” hinting at monopolistic concerns.

Another uncomfortable party is Sam Altman of OpenAI. Previously, OpenAI reached a cooperation agreement with Apple: ChatGPT would serve as a “supplementary option” when Siri answers user queries. That is, just as Siri sometimes lists web search results today, it could provide ChatGPT responses when needed in the future.

That collaboration once brought significant prestige to OpenAI, but now pales in comparison to Google’s new role with Apple.

Clearly, being a visible “supplementary option” on the front stage holds far less weight than serving as the invisible “foundational helper” behind the scenes. It’s like Apple hosting a show: OpenAI performs on stage, while Google operates the lighting backstage. Altman has not commented yet.

In China, whether Apple is also seeking a “transitional helper” before completing its trillion-parameter self-developed model remains to be seen.