Key Points
- Apple will run Private Cloud Compute on Google Cloud using Nvidia Blackwell B200 GPUs.
- Private Cloud Compute was built in 2024 to run only on Apple silicon.
- Nvidia confidential computing encrypts data in use to preserve privacy.
- Full capacity is targeted by the end of summer 2026.
Apple built Private Cloud Compute to run on Apple chips, in Apple-controlled facilities, and nowhere else. That promise just got a carve-out. Apple's privacy-critical cloud is expanding onto Google Cloud infrastructure powered by Nvidia's Blackwell B200 data-center GPUs — the first time Apple's confidential AI workloads will run on hardware Apple neither makes nor owns.
What's actually changing
The move powers the new, Gemini-based Siri Apple unveiled at WWDC. Simple requests still run on-device; the heavier, compute-intensive queries get routed to Google's servers. Private Cloud Compute, which Apple introduced at WWDC 2024 as a system that would run exclusively on Apple silicon, is the piece being extended off Apple's own racks. According to The Information, Apple's in-house PCC ran too slowly in testing with the new Siri model — its servers, in plain terms, weren't enough for advanced AI at scale.
The privacy bet
Privacy is the whole reason PCC exists, so the company that brands itself on it can't just ship customer data to a rival's data center. The hedge is Nvidia's confidential computing, which encrypts data while it is being processed, not only at rest or in transit. Apple is betting that encrypted-in-use processing on Google's Blackwell GPUs preserves the same guarantee it made when PCC ran only on Apple chips. It is a meaningful philosophical concession from the company that built a brand on keeping the silicon, the model, and the data under one roof — and a tacit admission that even Apple's roof isn't big enough for frontier AI. The same privacy instinct shows up elsewhere in the Siri rollout, including a standalone app that will reportedly offer auto-deleting chats.
Timeline
The rollout is expected to reach full operational capacity by the end of summer 2026, ramping gradually until then, with more technical detail promised at the Confidential Computing Summit later in June.
Source: Macworld, Crypto Briefing, CNBC
Comments