By continuing to browse our site you agree to our use of cookies, revised Privacy Policy and Terms of Use. You can change your cookie settings through your browser.
CHOOSE YOUR LANGUAGE
CHOOSE YOUR LANGUAGE
互联网新闻信息许可证10120180008
Disinformation report hotline: 010-85061466
VCG
China's first open architecture for AI computing was released on Friday at the 2025 World Smart Industry Expo in Chongqing. The collaborative project was announced by more than 20 companies across the AI industry, including chipmakers and large language model developers.
This initiative provides an open, collaborative model that addresses key challenges faced by China's AI development, like limited high-end computing power, high costs and a fragmented domestic hardware and software ecosystem, creating highly efficient, tightly integrated GPU-centric designs for large-scale AI computing.
The companies described the new approach as the "Android way" for AI, referring to a development pattern in which a foundational platform is made open and accessible to a wide range of companies, much like how Android provides a core operating system that different phone manufacturers can build upon.
Instead of relying on a single company's proprietary technology, this open model encourages diverse players – from chipmakers to software developers – to work together on a standardized platform. This could accelerate technological progress, reduce costs and propel China's vibrant smartphone industry.
The companies plan to release technologies across multiple layers of AI, including components, systems, infrastructure, software and data sets. This move aims to lower the R&D barrier for small and medium-sized enterprises.
Chinese supercomputer manufacturer Sugon took the lead to open up their capabilities of AI storage optimization, design specifications for liquid cooling and the DeepAI software stack.
Li Bin, senior vice president at Sugon, said the company has the responsibility and ability to help build this open ecosystem, adding that they have built more than 20 large-scale computing clusters over the last decade, with over 500,000 accelerator cards of different brands deployed.