Apple software chief Craig Federighi, right, has expressed interest in using outside models.
(Bloomberg) — Apple Inc. is in early discussions about using Google Gemini to power a revamped version of the Siri voice assistant, marking a key potential step toward outsourcing more of its artificial intelligence technology.
The iPhone maker recently approached Alphabet Inc.’s Google to explore building a custom AI model that would serve as the foundation of the new Siri next year, according to people familiar with the matter. Google has started training a model that could run on Apple’s servers, said the people, who asked not to be identified because the discussions are private.
Most Read from Bloomberg
WATCH: Apple is in early discussions about using Google Gemini to power a revamped version of its Siri voice assistant. Source: Bloomberg
The work is part of an effort to catch up in generative AI, a field where the company arrived late and then struggled to gain traction. Earlier this year, Apple also explored partnerships with Anthropic PBC and OpenAI, weighing whether Claude or ChatGPT could serve as Siri’s new brain.
Apple is still several weeks away from making a decision on whether to continue using internal models for Siri or move to a partner. And it hasn’t yet determined who that partner may be. Spokespeople for Apple and Google declined to comment.
Shares of Google and Apple both climbed to session highs on Friday after Bloomberg News reported on the discussions. Google was up 2.9% to $205.46 as of 1:15 p.m. in New York, while Apple rose 1.4% to $227.95.
The possible pivot follows delays to a long-touted upgrade of Siri, which would fulfill commands by tapping into personal data and let users navigate devices entirely via voice. The Siri update was scheduled for this past spring but was postponed by a year due to engineering setbacks.
The failure led Apple to sideline AI chief John Giannandrea from Siri development. The project is now overseen by software boss Craig Federighi and Vision Pro headset creator Mike Rockwell, who are weighing the use of outside help as a possible path forward.
Apple software chief Craig Federighi, right, has expressed interest in using outside models.Photographer: David Paul Morris/Bloomberg
The Siri update was originally designed around technology developed by the Apple Foundation Models team. That group also created the on-device large language models that power Apple Intelligence features like summarizing text and creating custom emoji.
Aiming to fix Siri’s flaws and bring the delayed features to market, Federighi, Rockwell and Apple’s corporate development team run by Adrian Perica started meeting with Anthropic and OpenAI about a potential deal.
Internally, Apple is holding a bake-off to see which approach will work best. The company is simultaneously developing two versions of the new Siri: one dubbed Linwood that is powered by its models and another code-named Glenwood that runs on outside technology.
Executives had long viewed Anthropic as the leading candidate for a partnership, but the financial terms demanded by that company led Apple to broaden the search and bring others into the mix. Apple certainly hasn’t ruled out sticking to its own models either.
These talks are separate from other deals to integrate chatbots into Apple Intelligence. Last year, Apple added ChatGPT as a Siri fallback for general knowledge queries, an area where the assistant has long underperformed. And both Apple and Google have already publicly signaled plans for a similar Gemini integration.
Though Apple and Google are rivals in smartphones, operating systems and services, they already have a search partnership. Google pays billions of dollars a year to make its search engine the default option on Apple products. But that deal is under antitrust scrutiny by the US Justice Department and could ultimately be undone.
The talks about using Google Gemini models to power Siri remain exploratory, with no formal commercial negotiations currently underway. Google has made similar deals before and powers much of the AI functionality on phones sold by Samsung Electronics Co.
Apple’s AI models group, meanwhile, continues to suffer upheaval. In July, Ruoming Pang — the team’s chief architect — departed for Meta Platforms Inc. He was lured by a $200 million package and a senior role in the new Superintelligence Labs unit.
Several colleagues soon followed him, and many of those who remain at Apple are interviewing for jobs elsewhere — either because of the potential shift to third-party technology or to cash in on multimillion-dollar job offers.
A more powerful Siri was originally meant to be a key part of Apple Intelligence last year.Photographer: Jason Alden/Bloomberg
Some software engineering leaders have floated a related idea of also replacing the models used for non-Siri AI features on Apple devices.
That would be a break from Apple’s preferred approach. The company has generally tried to maintain ownership of the AI features that run on consumers’ machines, aiming to ensure security and privacy. That includes Apple Intelligence capabilities like writing tools.
Such a plan isn’t something the company is actively working on, though. In the case of a Siri partnership, third-party models would run on Apple’s Private Cloud Compute servers, which use Mac chips for remote AI processing. That means the external Siri models wouldn’t run on devices themselves.
Apple remains years behind its rivals in AI, and that’s spurred management to consider a range of options. In a recent all-hands meeting, Chief Executive Officer Tim Cook told employees that Apple must win in AI and is stepping up its investments. He took an optimistic tone, noting that Apple is rarely first to new markets but still eventually provides a superior product.
During a quarterly earnings call that same week, Cook declined to comment on the use of third-party models, saying any discussion would reveal the company’s plans. But that in itself was a signal that Apple was considering an external approach.
Already, Apple has started enlisting partners for some consumer features and its internal operations. In the iOS 26 operating system, the company is offering a ChatGPT option for generating images, and it killed a publicly announced project to develop its own generative AI-based coding system. The company is using ChatGPT and Claude instead.
At the same time, work within Apple’s Foundation Models team continues. The company recently began testing its first trillion-parameter model, a major leap from the 150 billion systems currently running in Apple’s AI data centers.
Parameters are a measure of complexity and learning capacity for artificial intelligence models. On this basis, Apple is still well behind the AI leaders. OpenAI uses a several-trillion-parameter approach.
In any case, Apple has no plans to offer this more powerful system to customers yet. For now at least, it will just be used for research.
(Updates with shares in fifth paragraph.)
Most Read from Bloomberg Businessweek
©2025 Bloomberg L.P.