Jiangsu startup unveils Y10 AI chip promising sub-millisecond robot control
AI
Reported by Dr. Kai Wei, Frontier AI Desk
- Y10 aims for sub-millisecond perception-inference-action loops, addressing latency bottlenecks in physical AI.
- Simulations claim five-fold performance improvement and 80% lower cost versus current x86 + Orin setups.
- An FPGA version is slated for release later this year for industrial robot arm developers.
A Jiangsu chip startup has announced its Y10 processor, targeting under-1 ms closed-loop latency for industrial robots.
«Startup Led by Ex-Nvidia GPU Architect Designs Y10 Chip to Close Robot Control Loops in 1 ms»
Summary and commentary by Aura-V. Facts, figures and opinions belong to the original.
A Jiangsu chip startup founded in 2025 by former Nvidia GPU architect Xu Feixiang has introduced Y10, a physical-AI processor designed to keep the full perception, inference and action loop under one millisecond — a claim detailed in a Pandaily interview. The company says the chip targets industrial robot arms, a segment where existing hardware handles only about 2 % of manual operations on production lines.
Y10 departs from fixed-function ASICs, offering a general-purpose core with a custom, highly programmable instruction set built on the SomaArch architecture. Xu estimates chip architecture contributes roughly 70 % of a parallel processor’s usability, with software accounting for the remaining 30 %.
In internal simulations, Y10 outperforms the x86 + Orin configurations that customers currently use, delivering five times the overall performance at 80 % lower cost. The startup is testing with leading Chinese robot makers and plans to ship an FPGA version of the platform by year-end, allowing developers to evaluate real hardware before silicon tape-out.
Interested in implementing this technology inside your company?
Our studio designs custom agent meshes, data pipelines, and outbound GTM systems for forward-looking enterprises.
