In-House AI Stalls: Apple's New Siri Leans on Google's Gemini
Claire Weston
Apple plans to roll out a fully redesigned Siri in September, with its core model built on Google Gemini and heavy compute tasks running on Google Cloud servers equipped with Nvidia B200 chips — the first time Apple has relied on outside cloud infrastructure at scale for a flagship product, marking a clear pivot in its AI strategy.
How will the new Siri actually run?
The new Siri will handle as much as possible on-device — on iPhones and other Apple hardware — but tasks requiring heavier compute will be routed to Google Cloud servers.
Those servers run Nvidia's Blackwell B200 data-center chips — Nvidia's latest-generation AI chips, purpose-built for large-scale inference.
This means → Siri's "brain" no longer lives solely on Apple's own servers. Google Cloud is now a critical piece of Siri's compute stack.
How is user privacy protected?
Apple will activate Nvidia's Confidential Compute feature — a chip-level technology that encrypts data while it is being processed — so that even on Google's servers, Google cannot read user data.
In plain terms = the data is cooked in someone else's kitchen, but the recipe and ingredients stay locked the entire time — the kitchen owner sees nothing.
This is Apple's compromise between tapping external compute and honoring its privacy commitments.
Why did the in-house route fail?
Apple tried deploying a modified version of Google's Gemini model on its own Private Cloud Compute servers — built on Mac-series chips — but the system ran too slowly, and the effort was abandoned.
Private Cloud Compute launched two years ago to deliver more private, secure cloud processing. Its role in the new Siri is now unclear.
This means → Apple's custom silicon dominates on mobile, but for the data-center-grade compute that large-model inference demands, it currently cannot match Nvidia.
Why is this a strategic pivot for Apple?
The new Siri's core model relies on Google Gemini; its compute relies on Google Cloud + Nvidia chips. This is the first time Apple has brought in outside partners at this scale for a flagship AI product.
That breaks sharply with Apple's long-standing approach of controlling every core component — from chips to operating systems — through vertical integration.
This reflects how high the compute barrier in today's large-model race has become: even Apple cannot be fully self-sufficient in the short term.
What to watch next?
Apple will officially unveil the new Siri at next week's WWDC, showcasing what it calls "a more personalized, more sophisticated conversational experience."
The key question ahead: whether Apple's in-house compute can gradually replace its reliance on Google Cloud — and whether its AI strategy can return to a self-sufficient path.
In plain terms = right now Apple is borrowing someone else's ship to go to sea. Whether it ultimately wants to — and can — build its own is the real long-term story.
Content is for reference only, not financial advice.