Overview
- Multiple reports on Thursday say Apple will route certain Siri requests that need cloud power to Google Cloud where they will run on Nvidia Blackwell B200 GPUs.
- Apple plans to enable Nvidia’s confidential compute feature so data is encrypted while it is being processed on those chips, a hardware step meant to keep information private in shared cloud servers.
- The change reflects an engineering tradeoff after Apple found a modified Gemini model ran too slowly on its Private Cloud Compute servers, which use Apple silicon.
- Apple has licensed Google’s Gemini models for the cloud fallback and will combine those large models with smaller on‑device models that handle other Siri tasks.
- WWDC 2026 and the iOS 27 rollout are the next milestones where Apple is expected to explain which Siri features use the cloud, how Private Cloud Compute fits in, and what users should expect for privacy and timing.