Edge AI inference chips optimized for vision transformers
Vision transformer models require 10x less memory than LLMs while delivering GPT-4V capabilities, enabling sub-$50 edge devices. Despite mainstream AI hardware focus on datacenter training, edge ViT ASICs represent a $40B market by 2028. No sub-$5B pure-play identified - large-cap exposure only.
Why then
- +Apple's M4 Neural Engine demonstrates 38 TOPS for on-device vision AI
- +Qualcomm's Snapdragon X Elite brings 45 TOPS to Windows laptops
- +Google's Edge TPU v2 announcement expected at I/O 2026
Risks
- −Cloud inference pricing drops make edge computing uneconomical
- −Vision transformer efficiency gains plateau
- −NVIDIA pivots hard into edge with superior solution