Apple Rebuilds Siri with Google Gemini, Cuts Legacy Hardware Support
Summary
Key Takeaways
Apple rebuilds Siri using Google Gemini technology, but emphasizes it's not a rebrand: it uses none of Google's customer-facing models or search infrastructure, routing all queries through its Private Cloud Compute system with no retention of user data or chat logs.
Five new AFM 3 foundation models form the technical backbone: AFM 3 Core, AFM 3 Core Advanced, AFM 3 Cloud, AFM 3 Cloud (Image), and AFM 3 Cloud Pro, with the 20B-parameter multimodal model as the most capable on-device option. These power the revamped Siri across iPhone, Mac, Watch, and AirPods.
The most immediate hardware impact is watchOS 27: it drops Series 6/7/8, first-gen Ultra, and second-gen SE, only supporting S9 or S10 chips—the largest single-year compatibility cut for Apple Watch. Similar trends appear in macOS Golden Gate (ending Intel Mac support) and iPadOS 27 (requiring A14 or M1+).
Apple also introduced a minimalist Sports app and updated App Store guidelines to crack down on low-effort submissions.
Why It Matters
Apple's move is a defensive play against OpenAI and Samsung: leveraging Google Gemini to close Siri's context gap without building its own frontier models. However, it hides three lock-in mechanisms:
- Forced hardware refresh: watchOS 27's compatibility cut is not purely technical; Apple uses AI model requirements (larger neural engine, memory bandwidth) to push users to new devices, downplaying that the 20B-parameter model still heavily relies on Private Cloud Compute for complex tasks, exposing users to tail latency.
- Model dependency risk: Outsourcing core AI to Google Gemini creates a strategic vulnerability. If Google changes API terms or model weights, Apple loses end-to-end control over Siri's intelligence, contradicting its 'full stack' narrative.
- Ecosystem lock-in: Developers building on Core AI frameworks are now tied to Apple-Google joint AI stack, reducing flexibility to use other LLMs like Llama or Claude. The App Store guideline tightening further raises barriers, forcing developers to invest in this new AI infrastructure.
PRO Decision
【Vendors (Competitors)】
- Samsung should accelerate Galaxy AI's on-device self-developed models, highlighting Apple's dependency on Google for data sovereignty risks, and leverage One UI's open ecosystem to attract developers.
- OpenAI should integrate directly with Apple devices via Siri Shortcuts, offer transparent privacy controls, and emphasize ChatGPT's independent reasoning superiority over Apple's hybrid architecture.
- Qualcomm must boost Snapdragon AI Engine on-device inference performance, demonstrating ability to run open-source models like Llama 3 at similar power, undermining Apple's hardware lock-in with the 20B-parameter model.
【Enterprises (CIOs/Architects)】
- Assess long-term dependency risk on Apple devices for AI workloads: test Private Cloud Compute latency and availability for internal automation, and maintain cross-platform backup plans (e.g., Android + OpenAI).
- Perform TCO audit on watchOS 27's hardware cut: forced replacement every 3 years instead of 4-5 increases capital expenditure.
- Monitor App Store guideline updates for internal apps: simple utility apps must be regularly updated to avoid removal.
【Investors】
- See through Apple's PR: the Google partnership sacrifices AI independence, narrowing long-term moat. Consider shorting Apple and going long on Google (model licensing revenue) and Samsung (potential defectors).
- Watch independent benchmarks of the 20B-parameter model: if on-device inference shows high latency or low accuracy, Apple's AI strategy weakness will be exposed.
Get 3-5 key AI infrastructure signals weekly →
💬 Comments (0)