BofA Securities: Apple WWDC Marks a Major Positive Turning Point for AI Strategy
Miles Bennett
BofA analyst Wamsi Mohan calls Apple's WWDC 2026 a major positive inflection in its AI strategy — Siri is repositioned as a system-level AI agent, and Apple's privileged access to device context and private user data is a moat generic AI assistants cannot replicate.
What actually changed with the new Siri?
Siri moves from voice assistant to a context-aware, multimodal, cross-app system-level agent.
Mohan breaks the core capability into four dimensions: personal context, on-screen awareness, broad world knowledge, and cross-app operations.
This means → Siri is no longer just "set a timer for me" — it reads what's on your screen and orchestrates different apps to get things done.
Apple's strategic intent: make AI a native layer spanning iPhone, iPad, Mac, Apple Watch, Vision Pro, and the services ecosystem.
What is the real breakthrough in the AFM 3 architecture?
Apple's third-generation foundation model, AFM 3, runs a hybrid inference stack: routine tasks stay on-device; heavy workloads go to Private Cloud Compute (PCC) and Google Cloud with Nvidia GPUs.
The key on-device disclosure is AFM 3 Core Advanced — a sparse architecture with 20 billion parameters that activates only 1–4 billion per request.
In plain terms = the model is large, but each task "wakes up" only a fraction of the parameters — preserving capability while keeping power draw in check.
This reflects Apple's core trade-off: run strong AI on the phone without requiring a constant internet connection.
Why is "token economics" underappreciated by the market?
Mohan's 2030 scenario model assumes 50% of requests handled on-device, 50% routed to Private Cloud, with 5% going to AFM 3 Cloud Pro.
After weighting for image computation, inference load, and model-cost intensity, Cloud Pro carries 33% of workload units but accounts for 67% of weighted cloud costs.
This means → Apple's AI cost structure hinges not on "how many people use it" but on how often users trigger image generation and heavy-duty agentic reasoning.
In plain terms = the slice of requests that looks small on a pie chart is the slice that burns the most cash — margin sensitivity is extreme around usage patterns.
What defends Apple's moat?
Mohan maintains a Buy rating on Apple with a $380 price target.
The core logic: Apple holds privileged access to device context, app capabilities, screen state, and private user data — things generic AI assistants simply cannot reach.
This reflects a deeper judgment: in the AI-assistant race, whoever sits closest to user data is the hardest to displace.
The verification checkpoint is equally clear — whether Apple can convert this architecture into sustainable services revenue and deliver margin returns on its AI investment.
Content is for reference only, not financial advice.