Siri’s Core Requests Will Route Through Google Cloud Starting This Fall

Apple’s revamped Siri will process some requests through Google Cloud rather than Apple’s own servers, using Nvidia Blackwell B200 chips with confidential computing enabled to encrypt data during processing. The timing is tight: WWDC is days away, and a September launch is reportedly on track.

The more telling detail is what Apple tried first. The company explored running a modified version of Gemini on its own Private Cloud Compute infrastructure and hit performance limits. That’s not a minor footnote. It means Apple’s privacy-first architecture, which the company has marketed as a competitive advantage, couldn’t keep up with what the new Siri actually needs to do.

What the Google Cloud arrangement actually involves

Confidential computing is a hardware-level feature that encrypts data while it is being actively processed, not just stored or transmitted. Apple’s use of it on Google’s infrastructure is an attempt to extend its privacy guarantees into someone else’s data center, which is architecturally different from simply promising that data won’t be logged.

Apple has used third-party AI providers before, most publicly with OpenAI for ChatGPT integration announced at last year’s WWDC. Routing Siri’s own core requests through Google Cloud is a different category of dependency, one that touches the assistant’s default behavior rather than an optional escalation path.

The Blackwell B200 is Nvidia’s current flagship data center GPU, built for the kind of large-model inference that on-device chips can’t match at scale. Apple’s A-series and M-series silicon are genuinely capable, but the gap between what fits on a phone and what runs in a warehouse full of B200s is not closing fast enough for Apple’s timeline.

What Apple is quietly admitting here is that building a competitive AI assistant and maintaining full infrastructure control are, for now, in tension. The privacy framing is real, but so is the compromise.