Apple to Route Siri to NVIDIA Blackwell via Google Cloud

Apple will route larger Siri requests to Google Cloud servers running on NVIDIA Blackwell B200 data-center chips beginning in September 2026. The assistant will process most requests on-device and forward only heavier or more complex queries to the cloud.

The cloud-hosted instance will use a licensed large model on Google Cloud that runs on Blackwell hardware. Apple has approved NVIDIA’s confidential computing feature to keep models and user data encrypted while they are processed on third-party servers. Apple determined that its internally operated Mac-chip servers could not deliver the necessary performance for the larger model workloads.

NVIDIA introduced Nemotron 3 Ultra at Computex in Taipei on June 1. The company described Nemotron 3 Ultra as an open-source model with roughly 500 to 550 billion parameters designed for long-running agents and multi-step workflows.

NVIDIA reported that Nemotron 3 Ultra can deliver up to five times faster inference and reduce the cost of complex agentic tasks by as much as 30 percent. The company also noted that the Nemotron 3 family recorded more than 50 million downloads in the year up to April 2026. NVIDIA wrote in a company statement: “Nemotron 3 Ultra is built for that new workload. It’s a frontier smart model that delivers up to 5x faster inference and lowers the cost of complex agentic tasks by up to 30%.”

NVIDIA’s Blackwell B200 chips are built for large-scale model training and inference. The combination of a licensed cloud model on Google Cloud and Blackwell hardware will form the backend for heavier Siri queries when the update goes live in September.

Apple is scheduled to present more of its AI plans at WWDC 2026, which begins June 8. Market responses to the arrangement and NVIDIA’s product announcement were limited: TradingView data showed NVIDIA shares at $216.18, up about 0.7 percent over 24 hours, and Apple shares at $310.04, up about 0.2 percent over the same period.