Apple is in early discussions with Alphabet about licensing the Google Gemini large-language model to underpin a redesigned version of Siri, according to people familiar with the talks. Google has begun training a custom variant of Gemini that could run within Apple’s Private Cloud Compute infrastructure, a move that would keep user data on the company’s own servers while adding more conversational capabilities to its voice assistant. The iPhone maker has not yet chosen a partner and is conducting parallel tests of three approaches: an internally-developed model, Gemini, and rival systems from Anthropic and OpenAI. A final decision is expected to take several weeks, and the overhauled Siri is pencilled in for release in 2026. Turning to an external supplier would mark a break from Apple’s longstanding preference to build core technologies in-house, but executives have been seeking ways to accelerate their generative-AI roadmap after engineering setbacks delayed the Siri upgrade by roughly a year. Siri has lagged Alexa and Google Assistant on complex queries and deep app integration, and Apple is under pressure to close that gap as competitors roll out device-level AI features. News of the talks lifted Alphabet shares 3.7% and Apple 1.6% in New York trading on 22 August, reflecting investor expectations that a deal could expand Google’s AI footprint while helping Apple bring a more capable voice assistant to market.