Apple and Google announce Gemini-based collaboration for Apple's foundation models

Background

Apple builds its own foundation models, the large language models that power Apple Intelligence. Bloomberg's Mark Gurman reported on 5 November 2025 that Apple would pay about $1 billion a year to use a 1.2-trillion-parameter Gemini model to power Siri ^[2]. That was a press report. The formal statement came on 12 January 2026.

A parameter is one of the numbers inside a neural network that are set during training. The parameter count is a rough way to describe a model's size, and it is often quoted in reports about new models.

What happened

On 12 January 2026 Google and Apple issued a joint statement announcing a multi-year collaboration ^[1]. Under it, Google's Gemini models and cloud technology would underpin Apple's next generation of foundation models. These models would support future Apple Intelligence features, including a more personalised Siri ^[1].

Apple said it had concluded that Google's AI technology offered the strongest foundation for its models ^[1]. The conclusion was Apple's own, and it was stated in the joint announcement.

Key details

The statement set out two commitments about how the features would run. Apple Intelligence would continue to run on Apple devices and on Private Cloud Compute, Apple's system of servers for requests too heavy for a device. Apple's privacy standards would stay unchanged ^[1]. These two points tell readers where the processing happens and which privacy rules apply, even though the models themselves came from a partner.

The wording describes Google's contribution as the underpinning of Apple's models. That places the partnership beneath the features rather than making it a separate product that users would see directly ^[1].

Later developments

When Apple launched AFM 3 in June 2026, Craig Federighi said the new Siri did not use Gemini's models or infrastructure ^[3]. Wikipedia, citing MacRumors, reports that the models were refined using outputs from Gemini frontier models ^[3][2]. Frontier models are the most capable models a lab offers at a given time. Federighi's remark concerns the Siri product, while the refinement report concerns how the models were developed. Readers should weigh the two statements separately.

Impact

The announcement shows how large AI models are now built. A company does not always create every part of its model stack itself. Apple chose to combine its own products and privacy design with a partner's models and cloud technology, and it set out the basic terms publicly in a joint statement. For users, the practical question was whether privacy standards would change, and the statement said they would not.

Later statements from Apple and reports in the press describe the relationship in different terms. For that reason, readers who want to know what was agreed should go back to the joint statement from Google and Apple as the primary document, and treat the later reports as commentary on it.