Apple is in early discussions with Google about using Gemini AI to power a redesigned Siri voice assistant, marking a potential shift toward outsourcing more artificial intelligence technology as the company struggles to catch up in generative AI development.

The iPhone maker recently approached Alphabet’s Google to explore building a custom AI model that would serve as the foundation for a new Siri launching next year, according to people familiar with the discussions.

Google has begun training a model that could operate on Apple’s servers, though the talks remain exploratory with no formal commercial negotiations underway.

The collaboration discussions follow significant delays to Siri’s long-promised upgrade, which was originally scheduled for spring 2025 but was postponed by a year due to engineering setbacks.

The enhanced Siri would fulfil commands by accessing personal data and allow users to navigate devices entirely through voice commands.

Apple has sidelined AI chief John Giannandrea from Siri development following the delays.

The project is now overseen by software executive Craig Federighi and Vision Pro creator Mike Rockwell, who are evaluating external partnerships as potential solutions.

The company is conducting an internal competition between two Siri versions, “Linwood” powered by Apple’s own models and “Glenwood” using external technology.

Apple has also held discussions with Anthropic and OpenAI about potential partnerships, initially viewing Anthropic as the leading candidate before financial terms led to a broader search.

Both Apple and Google shares rose following news of the discussions, with Google climbing 2.9% to $313 and Apple increasing 1.4% to $348 during Friday trading.

Apple’s AI models team faces significant upheaval, with chief architect Ruoming Pang departing for Meta Platforms in July after receiving a $305 million package and a senior role.

Several colleagues followed, while the remaining team members are interviewing elsewhere due to the potential shift toward third-party technology or lucrative external offers.

CEO Tim Cook has emphasised to employees that Apple must win in AI while stepping up investments.

During a recent earnings call, Cook declined to comment on third-party model usage, stating that discussion would reveal company plans, itself signalling consideration of external approaches.

If implemented, Google’s models would run on Apple’s Private Cloud Compute servers using Mac chips for remote processing, meaning external Siri models wouldn’t operate directly on consumer devices.

This approach aims to maintain Apple’s security and privacy standards while accessing advanced AI capabilities.

Apple currently uses a 150 billion parameter system in its AI data centres and recently began testing its first trillion-parameter model, though this remains well behind industry leaders like OpenAI’s several-trillion-parameter systems.

The discussions represent a significant strategic consideration for Apple, which typically maintains control over AI features running on consumer devices.

The company has already begun partnering for some features, including ChatGPT integration for image generation in iOS 26 and abandoning internal coding system development in favour of existing solutions.