Apple will reportedly use a custom version of Gemini to power the new Siri (www.engadget.com)

Bloomberg reports Apple will power its revamped Siri (slated for a 2026 launch) with a custom version of Google’s Gemini running on Apple’s Private Cloud Compute. Gemini will handle Siri’s summarizer and planner functions—components that synthesize information and decide how to carry out multi-step or app-driven tasks—while Apple retains some homegrown models for other roles. The deal is reportedly worth about $1 billion per year, will be kept quiet publicly, and follows earlier explorations with Anthropic and OpenAI; Apple also already surfaces ChatGPT for certain requests via Apple Intelligence. For the AI/ML community this underscores a growing trend: major platform owners are mixing licensed, large third‑party models with proprietary systems to accelerate product timelines while maintaining control over data and execution. Technically, the arrangement implies Google delivered a version of Gemini engineered to run securely on Apple’s servers, prioritizing integration, latency and privacy constraints over pure cloud hosting. It also highlights strategic tradeoffs—outsourcing critical capabilities now to meet product goals, then planning to replace them with Apple’s own cloud model (reportedly a one‑trillion‑parameter system targeted for consumer use next year). The move raises questions about model portability, auditability, commercial licensing dynamics, and how top-tier LLMs will be packaged inside consumer ecosystems.