China’s GPU Chip Startup Xingyun IC Secures Billions of Yuan in Funding

TMTPOST -- Beijing Xingyun Integrated Circuit Co., Ltd. (Xingyun IC), a company founded by a Tsinghua University graduate, announced on Tuesday that it had secured billions of yuan in angel and angel+ rounds of financing.

Investors in the project include several leading strategic partners and renowned financial institutions, such as Zhipu AI, Ren'ai Group, Casstar, Miracle Plus, ShuiMu Tsinghua Alumni Capital, among others.

Ji Yu, the founder and CEO of Xingyun IC, told AsianFin that the company plans to achieve mass production of its AI assembler products by 2026.

Founded in August 2023, Xingyun IC focuses on developing next-generation, high-performance GPU chips tailored for large model inference applications. The company’s core team consists of experts from Tsinghua University and top global chip firms, making it a strong player in the domestic chip sector.

Ji, 31, is a Tsinghua University graduate with a bachelor’s degree in physics and a PhD in computer architecture. Previously known as one of Huawei's "young genius," he specialized in architecture and AI chips, contributing significantly to the development of Huawei’s Ascend chips.

Yu Hongmin, the co-founder and CTO of Xingyun IC, is an expert in semiconductor development, having previously led the design and production of chips at companies like Baidu, Huawei, and Horizon Robotics. With over a decade of experience, Yu has been instrumental in the development and mass production of numerous chips, overseeing teams of over 100 engineers.

Ji explained that today’s large language model infrastructure resembles the mainframe systems of the 1980s, but the future of AI and computing will be built on “white-box” assembled systems, similar to the evolution seen in the PC and internet industries.

Xingyun IC seeks to create a foundation for the AI industry that mirrors the approach taken by PC manufacturers in the past. By utilizing heterogeneous computing and white-box hardware models, the company aims to revolutionize large model computing systems, reducing the cost and improving the quality of AI models.

Xingyun IC’s goal is to integrate one or two specially designed GPUs into general x86 servers, thus forming a low-cost, high-performance product capable of supporting the memory-intensive needs of AI models. This approach allows for the efficient construction of AI systems using assembled components, a major departure from the costly, proprietary solutions currently dominating the market.

Ji’s vision centers on enabling AI’s “scaling law,” allowing AI models to grow and perform more effectively while reducing the marginal cost of computation. He believes that if AI hardware can be made affordable for consumers, the industry will enter a period of rapid, software-like growth driven by zero marginal costs.

The unique approach of Xingyun IC comes at a time when demand for AI processing power is surging. Domestic GPU chip companies, such as Biren Technology, Enflame Technology, and Moore Threads, have all started IPO tutoring, highlighting the enormous market potential for AI chips.

At the 21st China International Semiconductor Expo held on November 18, the global semiconductor industry association forecasted that AI will drive continuous growth in the semiconductor sector, with the global market expected to reach $1 trillion by 2030, growing at an annual compound rate of 8%.

This booming market for AI chips is expected to provide ample opportunities for companies like Xingyun IC, positioning them as key players in China’s push to dominate AI chip production.