Infinigence, China's interpreter in the Chip world

As DeepSeek emerged as a global sensation, it has not only reshaped the global AI competition landscape but also marked a crucial step for China in building a self-sufficient AI ecosystem. Yet, behind DeepSeek’s success lies a major challenge: How can China’s computing power fuel this technological revolution? Last week, I talked with Xia Lixue, the co-founder and CEO of Infinigence, to find out the answer.

.

During his Ph.D. studies in electronic engineering at Tsinghua University, Xia saw himself as a scientist. What fascinated him was how AI software and hardware could be seamlessly integrated.

After graduation, he gained invaluable working experience at Alibaba Cloud, where he led strategic projects such as compression and acceleration for Alibaba’s large language models and generative AI model chips.

At the end of 2022, the explosive rise of ChatGPT made him aware of the huge industrial opportunity: large models not only broke down human-machine interaction barriers but also demonstrated the potential to reshape countless industries.

Driven by this vision, Xia embarked on entrepreneurship in 2023, pioneering the market positioning of "computing power operator for the era of large models."

In just two years, Infinigence’s valuation skyrocketed, becoming as one of China’s most representative emerging tech companies in AI computing infrastructure.

What enabled Infinigence to secure nearly 1 billion yuan in funding within two years? Xia explained that investors recognized the company’s ability to solve a critical pain point in China’s AI industry.

AI development has always been driven by three key elements: computing power, algorithms, and data. The Achilles' Heel lies in China’s fragmented computing ecosystem—where various models and different type of chips create an "M×N" complexity, making it difficult for domestic chips to work efficiently, regarded as a bottleneck for China’s large model industry.

To address this, Infinigence built a "Heterogeneous Cloud" that acts as an "interpreter," enabling multiple large models to run efficiently across diverse chips—both in the cloud and on the edge. By "mixing and matching" domestic chips like Huawei’s Ascend and Biren Tech with international counterparts like AMD and NVIDIA, Infinigence achieved a computing utilization rate of up to 97.6%, reducing reliance on any single foreign chip brand.

"The most challenging part wasn’t the technology, but convincing clients that domestic chips could be both practical and powerful, until we solved the heterogeneous computing integration problem with our unique technical edge," Xia said.

At the Global Developers Summit in February this year, Infinigence launched China’s first "Computing Power Ecosystem Supermarket". Corrently, Infinigence’s technology powers computing platforms in Zhejiang, Sichuan, Beijing, and beyond. Notably, companies can now access remote computing resources as seamlessly as local ones.

A dynamic map displayed on a large screen at the company’s entrance shows computing nodes across more than 20 provinces and cities, displaying real-time data on resource utilization, temperature, and power consumption.

Whether China can achieve self-reliance in AI remains a focal point for the market. According to UBS, given the strategic importance of AI and U.S. restrictions, China’s AI computing expenditure is expected to grow from 18 billion USD in 2024 to 90 billion USD by 2029, with the domestic computing ratio surging from 33 percent to 90 percent, which means the market for localized AI computing will explode in the coming years.

AI advancement hinges on three elements—models, chips, and software. Their interplay creates a 'flywheel effect': models leverage the latest chips, while chips adapt to model innovations. Overseas, models, systems, and chips already form a closed-loop ecosystem—that’s partly why NVIDIA remains dominant.

Infinigence connects model developers and chipmakers, to accelerate China’s flywheel in AI industry.

China’s 2025 Government Work Report called for "building an integrated computing power system", underscoring China’s commitment to computing networks.

"This is a national strategy, and exactly what we excel at," Xia said. "We’re not here to catch up. We’re here to redefine the rules of the game."