Apple Integrates Google Gemini, Uses Nvidia Chips

Per reporting by The Information, Apple will run some user queries to a new Siri in Google Cloud on a licensed version of Google's Gemini model, and Apple has approved the use of Nvidia confidential compute in that setting, Aaron Tilley reports. The Information quotes sources saying Apple is "using a version of Google's large Gemini model to train a smaller version of the model that can run locally on Apple devices, a process known as distillation." The same reporting says the full Gemini model contains trillions of parameters and requires more compute than Apple's internal Private Cloud Compute has been able to handle.

Per The Information, Nvidia's confidential compute is a security feature inside Nvidia GPUs that encrypts data and AI models while they are processed, and enabling it can slow inference slightly while providing stronger protections for data in use. The Information reports that Apple's decision to allow the technology in Google Cloud is recent, occurring "in recent weeks," according to its sources.

Companies that split workloads between on-device distilled models and larger cloud-hosted foundation models are using a hybrid pattern to balance latency, capability, and privacy constraints. Distillation into smaller local models reduces bandwidth and latency for many queries, while licensing a full Gemini instance in the cloud handles more complex requests that exceed on-device capacity. Confidential compute is emerging as a practical control point for vendors that need to combine third-party cloud compute with privacy commitments.