Siri Now Runs on Google’s Nvidia Hardware, Not Apple’s
What You Need to Know
- Apple will run new Siri on Google’s infrastructure using Nvidia Blackwell B200 hardware instead of its own servers.
- Apple spent two years building Private Cloud Compute to avoid relying on external infrastructure for cloud AI processing.
- Apple’s silicon couldn’t run Gemini at acceptable speed, forcing reliance on Nvidia hardware Apple cannot build quickly enough.
- Apple Intelligence faced critical reception at WWDC 2024, with improved Siri repeatedly delayed past announced release dates.
Apple’s plan to rebuild Siri on Google’s infrastructure runs deeper than a software licensing deal. The cloud processing will run on Google’s Nvidia Blackwell B200 data center hardware, with user data protected through Nvidia’s hardware-based confidential compute feature rather than Apple’s own server stack.
That detail matters because Apple spent two years building Private Cloud Compute specifically to avoid this kind of arrangement. The system runs on Apple silicon and was positioned as proof that cloud AI didn’t have to mean handing your data to someone else’s infrastructure. Apple apparently tried to run a modified version of Gemini on that system and found it too slow to ship.
What Apple Is Giving Up
Private Cloud Compute may keep its branding even as the underlying hardware shifts to Google and Nvidia. That would be a tidy piece of marketing: the name survives, the architecture doesn’t. How the two systems coexist, or whether Private Cloud Compute handles any part of the new Siri at launch, remains unclear.
The Blackwell B200 is Nvidia’s current flagship data center chip, designed for large language model inference at scale. Apple’s own chips excel at on-device tasks, but running a frontier-scale model like Gemini at acceptable latency apparently required hardware Apple doesn’t own and can’t build fast enough to deploy.
The timing sharpens the stakes for WWDC on June 8. Apple Intelligence launched at WWDC 2024 to a reception that ranged from underwhelmed to openly critical, and the more capable, personalized Siri kept slipping past its announced windows. This year’s conference is effectively a relaunch, and the product Apple will demo is running on a competitor’s chips inside a competitor’s data centers.
For a company that treats vertical integration as both a business strategy and a brand identity, shipping Siri on Google’s Nvidia cluster is a more candid admission of where Apple’s AI infrastructure actually stands than any slide deck will say out loud.
0 Comments