ProductJun 10, 2026, 01:49 AM· United States
On-device AI agents hit a hard memory limit. Apple's new architecture routes around it.
Summary
Apple announced third-generation foundation models AFM 3 at WWDC26, breaking on-device AI memory constraints. The 20-billion-parameter AFM 3 Core Advanced stores weights in NAND flash instead of DRAM, using per-prompt routing decisions to avoid token-by-token weight swapping. The family includes two on-device and three server-based models, with server models running on NVIDIA GPUs in Google Cloud and on-device architecture being Apple's own.
Why it matters
This architecture solves the core bottleneck of on-device AI models being limited by DRAM capacity, potentially reshaping edge AI deployment.
Source links
⚠ Content is from official & reputable-media public sources, AI-assisted and auto-published, for information only — not investment advice.
Market reaction
The following is market reference information for related companies, and does not constitute investment advice.
NVIDIA
NASDAQ · NVDA
Anthropic
Private / not listed