From cloud to factory floor: KETI outlines four directions shaping next-gen physical AI chips

KETI’s roadmap signals an irreversible shift of AI compute from data centers to industrial edge environments. Technically, this accelerates adoption of in-memory computing, neuromorphic architectures, and low-precision on-device training—forcing EDA, advanced packaging, and RISC-V ecosystems to evolve in lockstep. Regulatory pressures in the U.S. and EU around energy efficiency and data localization will raise compliance costs, disproportionately hurting smaller players reliant on generic GPU stacks. Strategically, NVIDIA may leverage Grace-Hopper for end-to-edge-cloud integration, while MediaTek (Taiwan, China) and Cambricon (Mainland China) could double down on custom NPUs to bypass CUDA lock-in. Within 18 months, fragmented standards will emerge, but vendors offering vertically integrated sensor-chip-control software stacks will dominate the long-tail industrial market and establish new moats.