Apple has long marketed itself as the privacy company and the firm that controls its own silicon. So a quiet confirmation at WWDC carried real weight: some Apple Intelligence features now run in the cloud on Nvidia GPUs — the first time Apple has openly admitted leaning on its rival’s chips to power its AI.
What Apple confirmed
Speaking at Apple Park, software chief Craig Federighi detailed how Apple built its revamped Siri and the Apple Intelligence layer. The key disclosure: heavier AI features run in the cloud on Nvidia GPUs, as part of Apple’s Private Cloud Compute infrastructure. For a company that prizes on-device processing and vertical integration, outsourcing the heavy compute to Nvidia is a notable concession to the realities of frontier AI.
The privacy engineering
The twist is how Apple squares this with its privacy brand. Nvidia improvements — specifically ‘confidential compute’ — let Apple (and partner Google) build a system that processes data on Nvidia hardware while meeting Apple’s strict privacy standards. Confidential computing keeps data encrypted even while in use, so in principle neither Nvidia nor Apple’s cloud operators can see the user data being processed. It is the technical bridge that lets Apple use someone else’s chips without breaking its promise.
The three-way alliance
The arrangement deepens an unusual triangle. Apple partners with Google (whose Gemini models help power the new Siri) and now relies on Nvidia’s GPUs and confidential-compute features to run it all privately. Three giants — usually rivals — are collaborating because no one of them can deliver cutting-edge, private AI alone. It is a pragmatic admission that the AI era rewards partnerships over pure self-reliance.
Why it matters
The move reveals how even Apple, with its vast resources and chip prowess, cannot fully go it alone at the AI frontier — the compute demands are too great. It also makes confidential computing a strategic battleground: as more sensitive AI workloads move to shared cloud GPUs, the ability to process data without exposing it becomes essential infrastructure, not a nice-to-have.
The risks
Dependence cuts both ways. Relying on Nvidia ties Apple to a supplier whose chips are scarce and expensive, and any flaw in the confidential-compute model would be a privacy crisis for a brand built on trust. Apple is betting the engineering holds — and that users accept ‘cloud, but private’ as good enough.
The bottom line
Apple Intelligence running on Nvidia GPUs, shielded by confidential compute, shows the AI era forcing even the most self-reliant tech giant into partnership. It is a clever reconciliation of privacy and scale — and a sign that the future of private AI may run on someone else’s hardware, encrypted end to end.
Photo: MDGovpics / BY via flickr