On Monday, Apple and Google announced a multi-year deal for Apple to use Google’s Gemini models and cloud for Apple Intelligence and Siri. Details were not revealed but prior reports suggest that Apple could be paying Google about $1B annually, structured as a cloud-computing agreement. While the deal has been brewing for months, the announcement puts a capstone on a key decision that’s likely to shape how consumers will use AI in their everyday lives. It means that Siri may no longer be “stuck in the mud,” with a more personalized version potentially coming out later this year. Given this new direction, iPhone users might finally allow themselves to wonder about the possibilities: “What if Siri was actually really good?”

As a reminder of how we got here, Apple first revealed its vision for Apple Intelligence back in Jun 2024 – a year and a half ago. What it called Apple Intelligence was actually supposed to be a smaller on-device model and a larger model available through Apple’s “Private Compute Cloud” (PCC). On-device processing would be used to the extent possible for blink-of-an-eye latency, powered by Apple’s powerful chips. In instances when a request exceeded the system’s on-device capabilities, Apple would send the request for handling to PCC, which could tap larger models – either its in-house model or from partners like OpenAI for more complex queries (with user consent). (Google, Perplexity, Anthropic, xAI, and DeepSeek have reportedly been in Apple’s partner consideration set for this.)

The idea was that Apple Intelligence would lean on users’ “personal context” from all of Apple’s knowledge about a user – including their relationships, routines, communications, and content (e.g. contacts, calendars, messages, emails, photos, apps, files). With this extraordinarily valuable (and sensitive) information, Apple could theoretically provide a more capable and relevant AI.

This vision was oriented around a personalized Siri, which would have on-screen awareness and contextual memory, was capable of taking action across multiple apps, and could communicate more naturally. Apple has 2.2B active device users, which means Siri as its native assistant has an incredible distribution advantage, seeing 1.5B voice requests a day.

Unfortunately, despite being the first to launch a mainstream AI assistant in 2011, Apple’s team failed to keep pace with the industry’s state-of-the-art and struggled to achieve the vision for Siri. In part, this was due to departures from Apple’s team, given the AI industry’s revolving doors and outsized pay packages relative to Apple’s typical comp (not to mention hotly anticipated industry IPOs).

Apple had intended to have its next-gen Siri use Apple’s own on-device and cloud-based models (except for complex queries). As development fell behind and delays looked to stretch into 2026, Apple began exploring partnerships to accelerate – reportedly opening up discussions with Google, OpenAI, and Anthropic. Apple considered using partners for the 3 different core elements of the new Siri – a planner to take the input and plan how to respond, search systems for the web and devices, and a summarizer to synthesize the output.

Anthropic was at one point the leading candidate whose models were reportedly viewed as slightly better than Gemini, but it sought $1.5B+ per year, with sharp ramp-ups. OpenAI, in turn, said it made the “conscious decision to not become the custom model provider for Apple,” in the fall of 2025. (Its current partnership with Apple for more complex queries, which requires user consent to pass along, reportedly doesn’t generate much traffic for OpenAI.)

Around Aug 2025, Google began training a custom 1.2T-parameter model that would run on Apple PCC servers. (This is far larger than the 150B-parameter model that Apple uses today for its cloud-based model.) It was essentially a white-label custom model licensed by Apple from Google. It does not carry Google/Gemini branding, and Apple will be able to finetune the model on its own. This “behind the scenes” approach is in stark contrast to Google’s preexisting $20B+ deal with Apple to be the default search provider for Safari. With Gemini 3 and Gemini emerging as the leading model family, Apple decided to turn to Google for the planner and summarizer, and part of web search. (Apple would handle the device/user-data search for privacy reasons.)

The deal between Apple and Google is nominally nonexclusive – not surprising given Google’s antitrust challenges related to exclusive distribution agreements. Apple still hopes to eventually move away from Gemini and develop its own 1T+ parameter model, perhaps as soon as this year. Industry analysts are estimating the deal in total could still be worth $5B for Google. For Google, $1B annually isn’t that significant, especially given Google has been paying Apple $20B+ annually to be the default search provider for Safari. It is, however, a mark of confidence from Apple, which called Gemini the “most capable foundation” for its models in its announcement.

While Apple was likely reluctant to default to Google given their rivalry in mobile operating systems, the two are natural partners with a significant longstanding relationship in search. Both are consumer tech firms who understand the mobile operating system, who are stewards of massive stores of sensitive user data, and who are thoughtful about considerations like data privacy and user safety. Furthermore, Gemini has lately been outpacing its rivals in key categories like Text, Vision, and Search. If the alternatives are Anthropic, OpenAI, Perplexity, xAI, and DeepSeek, Google has the edge.

Not surprisingly, xAI founder Elon Musk has been intensely critical of Apple’s deals with OpenAI and now Google. xAI sued Apple/OpenAI last year, and this week has Musk saying, “This seems like an unreasonable concentration of power for Google, given that [they] also have Android and Chrome.” For OpenAI, the disappointment of its Apple relationship has led it to double down on developing its own AI hardware.

Nov 21 2025 (3 Shifts): Gemini 3 and Google’s progress

Jun 14 2024 (3 Shifts): Apple Intelligence and Apple’s ambitions