AI computing architectures are shifting from centralized cloud processing toward the edge. Consequently, hardware requirements for edge AI chips are rising steadily, demanding a careful balance between computing performance and power efficiency, according to DIGITIMES.
Historically, most edge AI chips were manufactured on mature process nodes below 22nm to control costs. Recently, however, startups such as DeepX have begun adopting advanced process technologies below 5nm, seeking to meet flagship product requirements through increased transistor density.
As a result, a clear bifurcation is emerging in foundry strategies for edge AI processors: more conservative products that prioritize cost control and supply stability continue to rely on TSMC's mature nodes, while more aggressive, flagship-oriented on-premises datacenter products are willing to assume higher risk by adopting Samsung Electronics' advanced processes.
