Shanghai State-owned assets lead the investment in an AI "shovel seller".

Recently, Jiliu Technology, a full-stack autonomous AI computing power building contractor, completed nearly 100 million yuan in Series A+ financing. The financing was jointly led by Shanghai Guotou Futeng Capital and existing shareholder Guofang Innovation, with Karamay City Development Fund and Zhangjiang Suixin Fund following as investors. Existing shareholder Zhuoyuan Asia made additional investments for the fourth consecutive round. The financing funds will be mainly used for core technology and product R & D, market expansion, and team building.

According to a reporter from Venture Capital Daily, this is already the company's 7th round of financing since its establishment, and it has received multiple rounds of investment from first - tier funds, industrial capital, and local state - owned assets, with the cumulative financing amount reaching hundreds of millions of yuan.

Jiliu Technology told the reporter from Venture Capital Daily that this round of financing reflects the market's recognition of Jiliu Technology's technical capabilities and engineering implementation strength. It also marks the company's official entry into a new stage centered on large - scale deployment and ecological collaboration from the product prototype stage. Jiliu Technology's revenue reached hundreds of millions of yuan in 2024, and it is expected to maintain a 50% revenue growth rate in 2025.

Focus on the AI Infra Track

Currently, the development of general artificial intelligence has promoted the evolution of computing power infrastructure towards higher performance and stronger stability.

As the gold mine of large models is just starting to be explored, it has also given rise to the AI Infra track. As the "shovel seller", NVIDIA's market value once exceeded 3 trillion US dollars, surpassing Apple to become the world's second - largest company after Microsoft.

AI Infra refers to the intermediate - layer infrastructure that links computing power and applications in the large - model ecosystem, including hardware, software, toolchains, and optimization methods. It is an overall solution.

The process of large - model application implementation is accelerating, and the value potential of AI Infra is being further released.

An investor from a leading institution told the reporter from Venture Capital Daily that in the era when large models are accelerating towards large - scale application implementation, AI Infra (artificial intelligence infrastructure) is crucial for improving the efficiency of model training and inference.

It is worth noting that the core value of AI Infra is more reflected in the inference layer, and the market capacity in this field far exceeds that of the training stage: almost all large - model R & D enterprises, application development companies, and entities in various industries that use large models for upgrading rely on inference support. Training is a phased behavior in the model production stage, while inference runs through the entire cycle of continuous model use, with more persistent and extensive demand.

Under this trend, Jiliu Technology has independently developed the intelligent computing system Galaxy HPAC (High - Performance AI Computing). Based on Galaxy HPAC, the company can provide customers with one - stop services for intelligent computing cluster design, selection, construction, optimization, management, and operation and maintenance, supporting multi - task scenarios such as large - model training, inference, and reinforcement learning.

Jiliu Technology told the reporter from Venture Capital Daily that since its establishment in 2023, Jiliu Technology has achieved leapfrog development from clusters of a hundred cards, a thousand cards to ten - thousand - card clusters, and has implemented multiple long - distance training and inference clusters (50 kilometers, 100 kilometers, 1500 kilometers).

As of now, the company has cumulatively networked, optimized, and delivered a computing power scale of 23 clusters, with more than 66,000 GPUs, more than 4,000 switches, and more than 320,000 optical modules. Its service targets include multiple leading users such as Zhipu AI, SenseTime, operators, data centers, and local state - owned enterprises.

The reporter from Venture Capital Daily noticed that in the AI Infra track where Jiliu Technology is located, there are both startups such as Silicon Flow, Wuwen Xinqiong, Luchen Technology, and Qingmao Intelligence in China, as well as large companies such as Alibaba Cloud, Huawei Cloud, ByteDance Volcengine, and Baidu. Through their respective cloud service platforms, they provide full - stack AI Infra capabilities, covering computing power scheduling, model training, and deployment. Although the above - mentioned companies all serve business scenarios such as large - model training and inference, their products, services, and profit models are different.

Shanghai and Xinjiang State - owned Assets Make Moves

Hu Xiaohe, the CEO of Jiliu Technology, graduated from Tsinghua University. As a visiting scholar at the University of California, Berkeley, he studied under Academician Scott Shenker, the proposer of the SDN network.

If the Tsinghua alumni have occupied half of the Chinese large - model startup landscape, then on the AI Infra track, the "Tsinghua Legion" of teachers and students working together is more like a high - profile bet: from Qingmao Intelligence that emerged from the Media and Information Laboratory, to Qingcheng Jizhi incubated by the High - Performance Computer Research Center, and then to Wuwen Xinqiong nurtured by the NICS - EFC Laboratory - almost all the star companies in AI Infra are from various Tsinghua laboratories; Yuan Jinhui of Silicon Flow and You Yang of Luchen Technology also graduated from Tsinghua.

Xie Wenqi, the co - founder/CFO of Jiliu Technology, graduated from Peking University with a major related to information technology. After graduation, he worked in a PE institution in investment.

After the release of ChatGPT, the two saw business opportunities and decided to start a business. Focusing on the R & D of distributed computing communication systems was the goal set by Hu Xiaohe at the beginning of his entrepreneurship.

Data shows that the global AI infrastructure market is experiencing unprecedented growth, and it is expected that by 2028, the expenditure will exceed 100 billion US dollars.

The reporter from Venture Capital Daily noticed that this round of financing includes state - owned assets from Shanghai and Xinjiang, as well as "Tsinghua - affiliated" artificial intelligence investment institutions.

Shanghai Guotou Futeng Capital was jointly established by Shanghai Guotou Company as the main initiator, together with Lingang Group, SAIC Group, and CATL, with a registered capital of 100 million yuan.

Guofang Innovation was established in 2017. It is an important carrier for the development of the science and technology industry in the Shanghai state - owned assets system in response to the national strategy of the Yangtze River Delta integration. It has cornerstone investors such as Shanghai International Group, Shagang Group, and Shanghai Airport, and manages assets of over 30 billion yuan.

Zhuoyuan Asia is a "Tsinghua - affiliated" semiconductor and artificial intelligence investment institution established in 2016. Its founding partner, Lin Haizhuo, graduated from Tsinghua University and focuses on the fields of artificial intelligence and integrated circuits. It has successively invested in companies such as Baichuan Intelligence, Zhongke Times, Pony.ai, and Fourth Paradigm.

This article is from the WeChat official account "Venture Capital Daily", author: Li Mingming. Republished by 36Kr with permission.