Apple rebuilds Siri with Gemini and Nvidia help
AI
Cloud
Gemini
Apple
Siri
Google
TLDR: SAN FRANCISCO, Calif.āApple says its new Siri is co built with Google Gemini foundation models and runs cloud workloads on Nvidia confidential compute within Private Compute Cloud.
Key Takeaways:
- Apple now runs Siri as a multi model system that splits work between an on device AFM Core Advanced model and larger cloud models.
- A System Orchestrator routes each request and uses a scarce model method to lock parameters for the full request on device.
- Private Cloud Compute expands to Google and Nvidia while Apple keeps a hard rule that devices only connect to software signed by Apple.
Apple wants Siri to feel instantly personal, but the heavy lift now happens in a choreographed partnership. The result is less about using Google and more about making its brains disappear behind Apple shaped privacy.
Apple wants Siri to feel instantly personal, but the heavy lift now happens in a choreographed partnership. The result is less about using Google and more about making its brains disappear behind Apple shaped privacy.
Q&A
What happens when Apple turns Siri to a different model choice mid request?
Apple says the on device scarce model method locks parameters for the full request. That design implies fewer mid request reroutes, reducing inconsistency when Siri shifts between local understanding and cloud reasoning.
Why would Apple insist Siri never āfeelsā like Gemini to users?
Appleās framing suggests the user experience layer stays under Apple control, including speech, context, and app handoffs. That helps Apple preserve brand style and privacy expectations even when models come from partners.
How does Appleās āsoftware signed by Appleā rule change the risk profile in third party cloud compute?
It creates a gate that blocks unsigned or unverified software from connecting. In practice, this limits what can run in the PPC pathway and helps Apple keep enforcement tighter than a purely open cloud deployment.
What is the real purpose of the System Orchestrator beyond sending prompts?
It decides where each query runs and builds the prompt accordingly. That orchestration also determines latency, cost, and whether Siri uses on device models or routes to AFM Cloud Pro style workloads.
Could confidential compute become a competitive expectation for premium assistants?
Appleās approach shows how privacy positioned as a product feature can force infrastructure choices. If other assistants chase similar guarantees, confidential compute style designs may move from technical detail to baseline requirement.
No comments yet. Be the first to share your thoughts!