The company officially released its second-generation in-house AI chip, achieving comprehensive breakthroughs in key metrics such as compute performance, energy efficiency, and memory bandwidth compared to the first generation, with an overall performance improvement of 200%.

The new chip uses advanced 7nm process technology, integrating optimized tensor computing units and high-speed interconnect architecture, specifically designed for LLM training and inference scenarios. In actual testing, the speed of training hundred-billion-parameter LLMs increased by 3x, and inference latency was reduced to milliseconds.

The chip has begun mass shipment to customers in finance, government, manufacturing, and other industries, significantly reducing AI computing costs and accelerating intelligent transformation.