On 21 April 2026, RoboSense held its Tech Day in Shenzhen and for the first time laid out a systematic chip roadmap. Two flagship SPAD-SoCs on its new “Genesis” digital architecture debuted: the Phoenix line for monolithic native ultra-high line count, and the Peacock line for all-solid-state high-resolution arrays. Both reach mass production in 2026. … Read more
RISC-V chip-design firm SunZhan and founder Zeng Yi argue AI has shifted architecture power from chip makers to AI firms, and are building tooling to auto-generate custom processors as the semiconduct
Moore Threads, the Chinese GPU designer that listed on Shanghai’s STAR Market in December 2025, plans a secondary offering in Hong Kong less than a year later. The company disclosed the plan alongside first-half results showing revenue of 1.736 billion yuan, up 147 per cent year on year. Moore Threads is pitching its KUAE ten-thousand-card … Read more
Jieluo Intelligent broke ground on a 1.2 billion yuan Suzhou base to make 2,000 trackless-navigation AMRs a year, serving about 70 per cent of the world’s top chip makers.
Deep Robotics filed its STAR Market IPO reply, seeking 2.502 billion yuan even though its profit comes from quadruped dogs and its humanoid sold just 4 units in two years.
China’s 3D-stacking chip startups share a label but solve three different problems. A teardown of the logic-recombine, memory-expand and compute-in-storage routes, and why domestic firms bet on the third.
Spot prices for NVIDIA’s B300 have climbed from about 570,000 dollars to more than 1.7 million dollars a unit in China since Lunar New Year. The scramble is reshaping how Chinese buyers source compute, build inference chips and price tokens.
Etched has raised 800 million dollars at a 5 billion dollar valuation with backing from TSMC, Geoffrey Hinton and Fei-Fei Li, plus a 1 billion dollar pre-production order for a chip that runs nothing but Transformers.
Jielo Intelligent breaks ground on a 1.2 billion yuan AMR manufacturing base in Suzhou, targeting semiconductor fabs with 2,000 units annual capacity.
HBM proved the trick HBM has shown that stacking and co-packaging storage media close to the GPU can relieve the data-movement bottleneck in AI compute. Now SK Hynix and SanDisk are applying the same idea to NAND, using HBF to put large-capacity flash next to GPUs and other xPUs in order to relieve the inference … Read more