According to AppleInsider, the restructuring of Apple’s AI teams throughout 2025 and the retirement of AI chief John Giannandrea are part of a deliberate internal strategy shift, not a reaction to individual failure. The reorganization, which moved robotics to hardware and Siri to the Vision Pro team, leaves a core group focused on developing the Apple Foundation Model. This new structure, with former Google engineer Amar Subramanya now reporting to Craig Federighi as VP of AI, is set for the planned relaunch of Apple Intelligence in early 2026 with iOS 26.4. The central feature will be a new LLM-backed Siri that can leverage on-device models, supplemented by third-party models like a custom Gemini in Private Cloud Compute. Despite about a dozen reported departures of AI-adjacent staff in the past year, the team is revealed to be massive, with nearly 200 contributors identified from public research papers alone.
The restructure was probably overdue
Look, when a high-profile exec like Giannandrea retires and a bunch of team leads bail, the easy story is “Apple’s AI project is in disarray.” But here’s the thing: the timeline suggests this was a cleanup operation, not a panic move. Giannandrea’s responsibilities had ballooned from Siri and ML to include the Apple Car and then robotics. That’s a wild scope for one person. By the time 2025 rolled around and Apple Intelligence was delayed, it makes perfect sense to streamline. Spin off the hardware-adjacent stuff, put Siri with the spatial computing team, and let the core AI researchers just focus on the foundation model. It seems less like a punishment and more like finally giving the key project the dedicated focus it needed. The fact that the new structure stayed in place after Giannandrea left tells you this was the plan all along.
The “Google brain drain” narrative is overblown
So a bunch of people are leaving, many from Google originally. Big deal. The report points out that senior folks staying 6+ years is normal churn at that level. And in the frenzied AI talent market, lower-level engineers hopping every 9 months? That’s practically the industry standard now. The more interesting tidbit is that Apple’s AI team is huge. Trying to catalog them from their public research papers quickly turned up nearly 200 names. That’s not a skeleton crew. They’re publishing constantly. The real issue might be that their research isn’t making waves because, true to Apple form, it’s not in a consumer product yet. But the engine is clearly running, and it’s massive.
The real strategy: on-device, plus a cloud backdoor
Apple’s stance hasn’t wavered: the core intelligence happens on your device. That’s Giannandrea’s legacy, and Subramanya apparently believes in it too. But they’re not dogmatic purists. The planned integration of third-party models like Gemini into Private Cloud Compute is the smart hedge. Need a super-powered web search or a super-specific task? Your on-device model can hand it off securely to a cloud model. It’s the best of both worlds: privacy and efficiency for most things, access to brute-force cloud models when you need it. This is Apple playing the long, integrated game, not racing to ship the chattiest chatbot.
What to actually expect in 2026
Forget the personnel drama. The roadmap is what matters. Early 2026 is the real target. iOS 26.4 is supposed to bring the “serious revamp” with a Siri that can actually use app intents via the on-device foundation model. That’s the holy grail – moving beyond a Q&A bot to an actual system-level assistant. If they pull that off, the past year’s restructuring will look like genius prep work. If they stumble again, then the narrative will shift. But betting against Apple when it finally focuses its resources on a core software problem? That’s usually a bad bet. They’re aiming for a thoughtful, private ecosystem in a world getting sick of AI slop. The play is classic Apple: be late, be different, be better. We’ll see if it works.
