Apple new Siri reportedly runs on Nvidia chips in Google data centres

Apple spent a decade building its own chips specifically so it would not have to depend on anyone else. Reporting now suggests the redesigned Siri will run on Google Gemini models, hosted in Google Cloud, on Nvidia Blackwell B200 hardware.

That is three dependencies on two direct rivals, inside the product Apple has staked the most reputation on fixing.

What Apple is reportedly giving up

Apple has separately used Google silicon to train models for Apple Intelligence, so the pattern is not new, it is newly visible. The company that markets vertical integration harder than anyone has quietly worked out that building this particular piece itself would cost more time than it has.

The privacy framing gets awkward too. Siri is the assistant Apple has spent years positioning as the one that does not ship your life to a competitor. Running it on a rival infrastructure does not automatically break that promise, since private cloud architectures exist precisely to handle this, but it does move the promise from something Apple controls end to end into something Apple has to attest to on someone else hardware.

Why it matters beyond Apple

The usual Apple move is to swallow a large one-off cost rather than accept a permanent dependency. Choosing not to, on its highest-profile product, is a fairly loud admission about how far behind it fell and how expensive catching up would have been.

It also says something about where the compute actually sits. TSMC July sales were up 45% year on year and the company has raised its 2026 revenue growth outlook past 40%, which reads as real orders rather than press releases. The same buildout shows up in big tech 2026 AI spending commitments and in the cost base behind the OpenAI IPO filing. Frontier inference capacity is concentrated in a handful of places, and Apple is not one of them.

For Google this is close to a best case. Having already passed a billion monthly Gemini users, it now gets to serve the assistant on Apple hardware as well, and gets paid for the compute either way.

What to watch for next

Whether Apple confirms any of this publicly, and how carefully it words the privacy explanation when it does. Whether the arrangement is a stopgap while in-house models catch up, or a settled architecture. And whether the new Siri actually lands on the devices people already own, including the HomePod mini, or gets held back for newer hardware.

Product links are Amazon affiliate links. We may earn a commission on qualifying purchases, at no extra cost to you.