Apple weighs outsourcing Siri to Anthropic or OpenAI in major strategic reversal

Summary

Apple is considering outsourcing Siri to either Anthropic or OpenAI rather than relying on its own AI models, according to Bloomberg's Mark Gurman. The move would acknowledge that Apple's in-house AI efforts have fallen short in the generative AI race.

The evaluation

Mike Rockwell, who took over Siri engineering in March after John Giannandrea was sidelined, ordered a comprehensive assessment of whether Siri should run on Apple's foundation models or third-party LLMs. After testing Claude, ChatGPT, and Google's Gemini, Rockwell and other executives concluded that Anthropic's Claude was most promising for Siri's needs. Apple asked both Anthropic and OpenAI to retrain versions of their models to run on Apple's cloud infrastructure for testing, a significant technical requirement that signals serious intent.

The internal project, dubbed "LLM Siri," was originally scheduled to launch in 2026 using Apple's own models. That timeline now appears in doubt.

The pricing impasse

Anthropc is seeking a multi-billion-dollar annual fee that increases sharply each year, a structure Apple finds difficult to accept. The opex model creates a cost center that scales with usage, which conflicts with Apple's goal of showing margin expansion in its services business. OpenAI appears more flexible on deal structure and could theoretically offer free or discounted access early on while monetizing through ads inserted into Siri responses, mirroring Google's search model.

Internal talent crisis

Apple's foundation models team of roughly 100 people, led by Reuming Pong (who joined from Google in 2021), faces serious attrition. Senior researcher Tom Gunter left after eight years, and this month the team behind MLX, Apple's open-source system for machine learning on Apple chips, threatened to leave entirely before Apple made counter-offers to retain them. Meta and OpenAI are offering AI researchers packages worth $10 million to $40 million annually, multiples of what Apple pays. The prospect of being relegated to integration work rather than frontier model research is pushing talent out the door.

Some team members have signaled unhappiness that Apple is considering third-party technology, fearing they will be blamed for the company's AI shortcomings.

Competitive position

Siri has fallen behind competitors since its 2011 launch. Apple already allows ChatGPT to handle web-based search queries in Siri and to generate images in iOS 18. The shift to outsourced LLMs would bring Siri closer to parity with Google's Gemini on Android, though adoption and user perception of Gemini's UX remain weak.

Apple briefly considered acquiring Thinking Machines Lab (founded by former OpenAI CTO Mira Murati) and Perplexity, but nothing materialized. The company has already approved a multi-billion-dollar budget for 2026 to run its own models via cloud, but plans beyond that remain unclear.

Brand and privacy

If Siri becomes visibly powered by OpenAI or Anthropic, Apple risks eroding the "aura of vertical magic" around the service. Precedent suggests consumers may not care. Siri itself was acquired, not built in-house, and users never objected. Apple could maintain privacy commitments by having OpenAI and Anthropic run custom versions on Apple's own cloud infrastructure, preserving the on-device or private-cloud narrative without requiring homegrown models.

Wall Street reacted positively, with Apple shares up over 2% after Bloomberg's reporting, suggesting investors favor pragmatism over pretense.

Apple is spending billions on AI infrastructure and talent but lacks the velocity and researcher density to compete in a field that ships multiple model updates per month. Licensing third-party technology is faster and cheaper than building from scratch, but it also signals that Apple's AI strategy has stalled.