Apple nears $1B Google deal for custom Gemini model to power Siri (9to5mac.com)

Bloomberg reports Apple is finalizing an agreement to pay roughly $1 billion a year for a custom Google Gemini model — reportedly about 1.2 trillion parameters — to power a major Siri overhaul slated for next spring. The deal (part of internal “Project Glenwood”) would have Gemini handle Siri’s summarizer and planner components — the parts that synthesize information and decide how to execute multi-step tasks — while some capabilities will remain on Apple’s smaller in-house models. Crucially, Google’s model would run on Apple’s servers (inside its Private Cloud Compute), so user data reportedly won’t be shared with Google. Apple’s current cloud model is ~150 billion parameters and its on-device model is ~3 billion, and Apple is still developing its own ~1 trillion-parameter cloud model aimed at consumer use next year. This is significant because it marks a major tech-industry shift: a top consumer device maker outsourcing core assistant intelligence to a rival’s massive model rather than relying solely on in-house R&D. For the AI/ML community, the deal underscores trade-offs between model scale, cost, privacy architecture (on-prem/enterprise-style hosting), and product timelines. It also highlights strategic pressures—Apple’s talent losses and the ballooning cost of state-of-the-art models—while signaling that hybrid architectures (mixing third-party large models with internal models) may become a preferred route for deploying advanced, privacy-conscious AI features.