Summary: Apple and Google confirmed a multiyear deal on January 12, 2026 to build the next generation of Apple Foundation Models on Gemini models and cloud technology, with a more personalized Siri promised later that year. Apple did not confirm terms, but Bloomberg reported an approximately $1 billion annual license for a bespoke 1.2-trillion-parameter Gemini model.
A joint statement landed on January 12, 2026: Apple and Google confirmed a partnership running across multiple years, under which Apple Foundation Models, in their next generation, are to be built atop Google’s Gemini models and cloud technology to drive the Apple Intelligence functions still to come, and the two companies promise a “more personalized Siri” before the year is out. Reuters framed the arrangement as a significant victory for Alphabet. According to CNBC, Apple would not confirm any terms; Bloomberg, however, had reported earlier that Apple intends to pay on the order of $1 billion a year for a license to a bespoke Gemini model carrying 1.2 trillion parameters.
What makes this deal remarkable is not what it costs but where it points. A company whose identity rests on vertical integration — its chips designed in-house, its models developed in-house — has decided to have a competitor it has spent years fighting in court over search build the base for its AI assistant. From Google’s side, Gemini has, for the first time, become the Apple ecosystem’s foundation-model layer — and that same day, Alphabet’s worth on the market briefly pushed past $4 trillion.
The Terms on the Table: A Model License Bundled with Cloud Technology
The statement itself is brief, yet three phrases merit unpacking. One: “multi-year collaboration” — not bought once, but a long-term tie-up. Two: “Apple Foundation Models built on Gemini” — Google provides the base models together with the cloud technology; Apple, for its part, keeps everything above them, the layers where the product and the experience live. Three: “a more personalized Siri this year” — CNBC notes the reworked assistant should land later on in 2026, which would close the book on a delay Apple once framed as “it’s going to take us longer than we thought to deliver on these features.”
Apple’s official posture on price and scale is silence. Every figure on the record traces back to Bloomberg’s reporting: roughly $1 billion a year, for a custom model of 1.2 trillion parameters. What Apple says publicly stays at the qualitative level: “After careful evaluation, we determined that Google’s technology provides the most capable foundation” for Apple Foundation Models.
Why Licensing Won Out Over Building at Apple
Siri’s record of missed deadlines tells the story. The overhaul of Apple’s assistant has cycled through announcement and postponement more than once, and the company itself has conceded the features are arriving more slowly than promised. The in-house route could not match the pace of competition, and in 2026 the market for assistants offers no grace period, with OpenAI, Google, and Anthropic each moving agentic assistants into mainstream interfaces.
Strip it down and Apple’s move is a purchase of time — paying cash to shrink a gap in capability that in-house work had failed to close on Apple’s own timetable. The division of technical labor also lines up neatly: capability at the base comes from Gemini; the privacy architecture — Private Cloud Compute plus on-device — remains Apple’s; and so does control of the product.
Google’s Side of the Ledger
CNBC’s reporting sorts the significance into three layers. The first is endorsement of Google’s AI turnaround — Alphabet has just closed its finest year since 2009; its market cap, for the first time since 2019, now sits above Apple’s. The second is distribution: Gemini now occupies the spot every vendor of models dreams of, the Apple ecosystem’s foundation-model layer, with the features Apple Intelligence adds down the line to be layered on top of it. The third is revenue: should Bloomberg’s figures prove accurate, roughly $1 billion a year flows in as recurring revenue from licensing that never touches consumer subscriptions — licensing in its purest B2B form.
Privacy: The Architecture, Plus the Partnerships That Stayed
One point gets special emphasis in the statement itself: Apple Intelligence is to keep running on Private Cloud Compute and on Apple’s devices; its privacy standards stay as they are — still leading the industry. Put differently, the models and the cloud technology may come from Google, yet the path that inference follows stays inside Apple’s own architecture — a precondition that matters greatly to corporate purchasers and privacy-minded users.
Two arrangements carry on untouched. Apple is still working with OpenAI on ChatGPT, and Google keeps paying to hold the default search spot on iPhones — the antitrust ruling let that arrangement continue. Apple thus runs a multi-track strategy for sourcing AI: Google, OpenAI, and Apple’s own models each keep a place at the table.
Product Teams and Developers: The Takeaways
Three things stand out. To start, the platform layer of assistants is converging: once Gemini sits under Siri, teams working on AI products inside Apple’s ecosystem will, in practice, plan features and write prompts against Gemini-compatible capabilities. Next, the fight over where foundation models get distributed has entered fresh territory — licensing models has turned into a platform-level deal, kin to the old fights over default search placements, and those placements historically drag antitrust tails behind them. Finally, our opening outlook flagged the industry-wide speedup in model cadence this year — Apple’s deal shows the same trend from another side, now that even hardware giants no longer gamble their release plans on in-house schedules.