Apple โคAppoints New AI โlead as โStrategy Shifts
CUPERTINO,โ CA – November 6, 2025 – Apple has named Subramanya Sharmaโค as its new head of Artificial Intelligence, replacing John Giannandrea, who is stepping down from the role. The move comes amid increasing pressure on Apple too accelerate itsโค AI progress and follows reportsโ of internal challenges and a strategic pivot towards utilizing external AI models.
Giannandrea,who joinedโ Apple from Google inโข 2018 where he led Machine Intelligence and Search,previously oversaw Apple’s AI โstrategy,machine โคlearning infrastructure,and Siri development. During his tenure, Apple focused on โprocessing AI tasks โdirectly on users’ devices usingโค its custom Apple Siliconโฃ chips, prioritizing user privacy by minimizing data collection. More complex requests were routed through Private Cloud Compute, servers designedโ for temporary dataโ processing and immediate deletion.
However, recent investigations by Bloomberg revealed organizationalโค dysfunction within โฃApple’sโ AI teams, including communication โbreakdowns between AI and marketing, budget misalignments, and an exodus of โAI researchersโ to competitorsโฃ like OpenAI, google, and Meta.The report detailed a climate of internal criticism, wihtโค some employees reportedly referring toโข Giannandrea’s group as โ”AI/MLess.”
In March 2025, โคapple reassigned Siri oversight from Giannandrea to Mikeโค Rockwell, creator of the Vision Pro. The company also removed its secretive roboticsโ division โfrom giannandrea’sโ control.
Now, Apple โขis reportedly planning to integrate Google’sโ Gemini model to power the โnext iterationโ of Siri, a notable development given the longstanding rivalry between the two tech giants.
Sharma will report to Craig Federighi and is tasked with helping Apple catch up in the rapidly evolving โAI landscape. Apple’s approachโ of on-device AI processing presents trade-offs, โคresulting in smaller, less โคcapable models compared โto those running in competitors’โ data centers. apple’sโฃ privacyโข focus has also limited its access to the vast real-world data sets used to train AI โฃsystems at companies like Google and microsoft.