Apple is putting the future of Siri—and much of its AI strategy—in the hands of a longtime rival.
Apple’s AI Bet on Google
In a multi-year deal announced January 12, Apple said it will rely on Google’s Gemini AI models and cloud infrastructure to power a major Siri overhaul and a broader slate of “Apple Intelligence” features due later this year.
In a joint statement, the companies said they concluded Gemini offered the strongest base for what Apple calls its “Foundation Models” and promised a wave of new experiences built on top of it.
The decision marks a rare break from Apple’s longstanding preference to build core technologies itself—and a quiet acknowledgment that its own AI work wasn’t keeping up with the rapid progress at Google, OpenAI and others.
A Rare Alliance After Years of Missed Deadlines
Under the agreement, Gemini will sit beneath a redesigned Siri expected with iOS 26.4 in March or April, an upgrade Apple first previewed at its 2024 developer conference and then delayed at least twice.
The new Siri is meant to do what users have expected for years: understand what’s on screen, chain together multi-step tasks across apps, and remember personal context such as a parent’s flight details pulled from Mail and Messages.
Inside Apple, the delays had become a source of frustration. The company originally targeted iOS 18 in 2024 for the reboot, but internal tests showed Apple’s own large language models lagged badly behind systems from Google, OpenAI and Anthropic, according to people familiar with the work.
“It’s a very public course correction,” said one former Apple AI engineer who worked on Siri until last year. “We spent years saying on‑device models would be enough. Then ChatGPT happened, Gemini happened, and the gap was just too wide.”
Apple had already dipped a toe into outsourcing by allowing Siri to route some optional queries to OpenAI’s ChatGPT beginning in 2024. Making Gemini the default reasoning engine for the assistant—and for future Apple Intelligence features—goes much further.
How It Will Work: Apple’s Face, Google’s Brain
Apple says its basic architecture will stay the same. Simple tasks—timers, alarms, short text edits, offline commands—will continue to run on-device on Apple’s A‑ and M‑series chips.
More demanding requests, like summarizing long email threads or orchestrating multi‑app workflows, will be sent to Gemini models running in what Apple calls its Private Cloud Compute.
According to Apple, those cloud requests will be encrypted end‑to‑end and processed on servers it owns and operates. Google supplies the models and cloud stack underneath, but is not supposed to see user identities or raw content.
“Apple Intelligence will continue to run on Apple devices and on Apple’s Private Cloud Compute servers, with industry‑leading privacy standards,” the company reiterated Monday.
Privacy advocates are unconvinced so far.
“Any time you embed a rival’s neural network at the heart of your assistant, you owe the public independent verification,” said a security researcher at a Bay Area university, who described current disclosures as “marketing, not a threat model.”
Key details remain missing from public documents, including whether Gemini can log anonymized interaction data, how long intermediate outputs are stored, and what, if any, data can be used to further train Google’s models.
Money, Power and Antitrust
Financial terms were not disclosed, but Apple is expected to pay around $1 billion a year for a custom 1.2‑trillion‑parameter Gemini model, according to earlier reporting.
That figure is small next to the more than $20 billion a year Google reportedly pays Apple to remain the default search engine on iPhones. Still, it deepens a partnership already at the center of high‑profile antitrust cases in the United States and Europe.
In 2024, a federal judge ruled that Google illegally maintained a search monopoly in part through default‑search deals, and later restricted exclusive agreements that run longer than a year. Regulators are now expected to look closely at whether putting Gemini at the heart of Siri effectively recreates that leverage in AI, even if Apple is technically free to plug in other models.
Dan Ives, an analyst at Wedbush Securities, called the pact “a major validation moment for Google” and “a stepping stone for Apple to get its AI strategy on track into 2026 and beyond.”
Investors seemed to agree. Alphabet, Google’s parent company, briefly crossed a $4 trillion market value for the first time after the announcement, a sign of market confidence that Gemini will now sit in front of more than two billion Apple devices worldwide.
What Changes for Users—and What Doesn’t
For iPhone owners, the first difference will be how Siri behaves in everyday conversations.
The assistant is expected to handle more natural, multi‑part questions, interpret mixed inputs such as “this photo plus that email,” and keep context over days instead of treating each query as a one‑off. A user might say, “Can you move Mom’s flight to tomorrow and update lunch?” and expect Siri to parse airline emails, calendar entries and messages with fewer errors.
On Macs, Apple will pitch Siri and Apple Intelligence as a productivity co‑pilot, able to summarize long documents, draft responses and coordinate actions across apps. On Apple Watch and HomePod, the gains may be more subtle, in part because of hardware constraints.
The experience will also depend heavily on connectivity. Because the most advanced features rely on cloud processing, users with spotty or slow internet connections may still see the older, stripped‑down Siri behavior for complex tasks.
And even with Gemini, the assistant won’t be infallible. Large language models still hallucinate, misunderstand instructions and miss nuance at the edges. Apple has signaled that much of 2026 will be spent observing how the system behaves in the real world and adjusting it accordingly.
For privacy‑conscious users, the trade‑off is stark. To get the most out of Siri, they will need to allow more queries to leave the device for cloud processing. Turning off those features will mean a noticeably less capable assistant. As of Monday, Apple had not outlined any new, granular opt‑out controls beyond its existing privacy settings.
Winners, Losers and the Next Fight
In the near term, Google is the clearest beneficiary. After spending much of 2023 and 2024 portrayed as trailing OpenAI in the AI race, it now becomes the default brain behind the most valuable consumer hardware ecosystem on the planet.
OpenAI, by contrast, is pushed to the sidelines on Apple platforms. Its ChatGPT integration, once marketed as a marquee partnership, becomes just one more optional backend rather than the main engine driving Siri.
Inside Apple, the deal buys time. The company is still training its own trillion‑parameter cloud model and has told investors and developers that it ultimately wants to reduce its dependence on Google. For now, Gemini serves as a bridge between Apple’s ambitions and what its own models can reliably do.
The next big question is how open this ecosystem will be. Apple and Google both describe the collaboration as “non‑exclusive,” but neither has committed to letting users choose a different default AI provider for system‑level tasks—say, Anthropic’s Claude or another rival—rather than using them as standalone apps.
For regulators, developers and everyday users, that choice may define the next phase of the AI era: if your phone’s most personal decisions and recommendations run through someone else’s neural network, how much real control do you actually have over which brain is in charge?
