Apple's AI overhaul and Siri reboot under Tim Cook

Apple has long sold itself as the vertically integrated, privacy-first alternative to Google's data-hungry model. The Siri reboot inverts that posture: the rebuilt assistant runs on a custom Google Gemini model, and its heaviest reasoning routes out to Google Cloud on Nvidia Blackwell B200 GPUs ^[1]. The arrangement is a multi-year deal reported at roughly $1 billion per year ^[7], with Apple architecting around the obvious objection: a three-tier routing system keeps simple requests on-device, sends moderate ones to Apple's own Private Cloud Compute, and only escalates the hardest queries to Google, which is contractually barred from using Siri queries to train future Gemini models ^[1]. The strategic tension is real regardless of the safeguards. Apple's flagship consumer AI now depends on the company it competes with most directly, and that dependency was struck precisely because Apple could not ship a competitive model on its own timeline ^[2].

The rebuilt Siri is built on a 1.2-trillion-parameter mixture-of-experts model ^[1]. A mixture-of-experts design does not fire all 1.2 trillion parameters on every request; it activates only the relevant subset per query, which keeps a very large model fast and cheaper to run than a dense model of the same size. Apple pairs that with a three-tier routing scheme that decides where each request is answered: trivial tasks resolve on the device itself, mid-weight tasks go to Apple's Private Cloud Compute, and only the heaviest reasoning leaves Apple's perimeter for Google Cloud's Blackwell B200 GPUs ^[1]. In practice that means a quick timer or unit conversion never leaves your iPhone, while a multi-step reasoning request may be handled by Google's hardware. It is a rebuild, not a patch: technical observers in developer communities have noted the same point, that this is a ground-up architecture rather than incremental tuning of the old Siri.

The reboot is also a reorganization. A 2025 crisis meeting convened near Apple's software engineering department to confront the Apple Intelligence failure and the slipping Siri update; Federighi led most of the discussion, and Vision Pro creator Mike Rockwell volunteered to fix Siri ^[2]. Rockwell reportedly drove the Google deal alongside Federighi and Eddy Cue, but he reports to Federighi rather than to Cook, will not take the WWDC stage, and has received little public credit ^[2][3]. Reporting indicates he did not get the senior VP promotion he sought and has considered stepping back ^[8]. Former AI chief John Giannandrea, in whom Cook reportedly had little confidence, was effectively displaced ^[2]. Bloomberg's Mark Gurman frames the launch around exactly this cast: Cook's final act, Federighi as the new AI leader, and Rockwell as the man fixing Apple's biggest shortcoming.

Investors hoping the new Siri triggers an upgrade wave will likely be disappointed. Apple Intelligence already failed to move the needle on sluggish iPhone sales, per analyst Ming-Chi Kuo ^[5], and Gene Munster's read is that Apple has not blown anyone away but the upgraded Siri could help it catch up to rivals ^[6]. The framing across coverage is that this is Cook's long-term bet to leave Apple credible in AI before handing the company to John Ternus on September 1 ^[1], rather than a near-term hardware catalyst. The hardware story is muted further by iOS 27 dropping the iPhone 11 family, which narrows the install base that can access the new Siri at launch ^[1]. The smart-home angle, including a previewed homeOS and a HomePad with a 7-inch display, points to where Apple wants a capable Siri to matter next ^[1].

Apple's hardest problem may be credibility rather than capability. Analyst Avi Greengart's critique is pointed: the issue was never the delay but that Apple promised a smarter Siri as a reason to buy devices that could not yet deliver it ^[4]. That promise traces back to WWDC 2024, where Apple announced three Siri improvements that then repeatedly slipped ^[7]. The result is a wary audience. Across Reddit communities the dominant sentiment is skepticism and delay-fatigue, with buyers who upgraded for Siri features that never shipped voicing anger and many adopting an I-will-believe-it-when-I-see-it stance, alongside references to the false-advertising concerns; a vocal minority defends Apple's restrained on-device approach and argues a ground-up Gemini rebuild is the right call. Apple's own executives have leaned into a public post-mortem, defending a take-its-time philosophy in candid interviews about what went wrong, signaling the company knows it has to earn back trust before users will lean on Siri again.