Just now, Henan and Zhipu invested in a couple of top students from Tsinghua University and Peking University.
This round of financing was jointly led by Huirong Fund of Henan Investment Group and Xinglian Capital of Zhipu Ecosystem Fund, with Zhuoyuan Capital participating in the follow - on investment.
Currently, as the parameter scale of large models has exceeded one trillion and the training data volume has exceeded 20TB, the shortage of computing power resources and the rising cost have become the core factors restricting the development of general artificial intelligence. At the same time, industry users still face multiple challenges such as prominent computing power bottlenecks, increasing energy - efficiency pressure, insufficient adaptability, and rising demands for autonomy and controllability.
More realistically, the industry's demand for computing power has shifted from "available and easy to use" to "efficient, flexible, green, and secure". How to build an autonomously controllable AI computing power infrastructure and achieve the leap from "available" to "good to use" has become a common challenge faced by the entire Chinese technology industry.
This challenge has also given rise to new business opportunities and tracks. Just recently, Infrawaves, an AI infrastructure manufacturer, announced the completion of nearly 100 million yuan in Series B financing. This round of financing was jointly led by Huirong Fund of Henan Investment Group and Xinglian Capital of Zhipu Ecosystem Fund, with Zhuoyuan Capital participating in the follow - on investment. Among them, Xinglian Capital made additional over - investment in both Series A and this round, and Zhuoyuan Capital has supported the company for six consecutive rounds.
Elite Graduates from Tsinghua and Peking Universities Join Hands to Start a Business, Moving from the Laboratory to the AI Infra Track
The story of Infrawaves is quite simple. The original intention was that a young man wanted to take control of his own future.
"Actually, I've always wanted to start a business and do something meaningful on my own," recalled Hu Xiaohe, the founder of Infrawaves.
As early as in the wave of mass entrepreneurship and innovation in the mobile Internet era, he worked on a project with Cheng Hao, the founder of Zhaoxi Calendar and his senior from the Department of Automation at Tsinghua University. At that time, he was one of the early members of the Future Intelligent Robot Team at Tsinghua University and grew up in the entrepreneurial atmosphere at Tsinghua University along with Yu Hao, the current founder of Dreame.
In his junior year, after having some internship experience, Hu Xiaohe quickly realized that "without core capabilities, one can only do operations or write code". So, he consulted a tutor with overseas experience in network security entrepreneurship. The tutor advised him to pursue a doctoral degree first, saying, "At least become an expert in one field and regard pursuing a doctorate as a complete entrepreneurial training."
During his doctoral studies and post - doctoral research, Hu Xiaohe delved deeply into the fields of cloud computing and high - performance networks. He not only studied high - performance network systems under the guidance of Researcher Li Jun but also studied with Academician Scott Shenker, the proposer of SDN networks, at the University of California, Berkeley.
In 2020, he joined the SenseTime Research Institute for an internship and further identified the pain points in the industry. After a short internship, by 2022, although Chat - GPT had not yet exploded, it was his last year as a post - doctoral researcher, and he began to seriously look for a business direction.
"In the first half of 2022, we were looking at the applications of high - performance networks in the fields of AI and semiconductors." In the second half of the year, his project was admitted by Miracle Plus and received the first start - up capital. In early 2023, he officially started his business, just in time for the explosion of large AI models, hitting the rhythm of the industry.
Xie Wenqi, another founder and CFO of Infrawaves, was Hu Xiaohe's roommate during their time at Hengshui High School, and they have excellent chemistry. Different from Hu Xiaohe, who has been deeply involved in academic research, Xie Wenqi graduated from Peking University with a major related to information technology for his undergraduate degree and continued his postgraduate studies at Tsinghua University. After graduation, he worked in a PE institution, making cumulative investments of hundreds of millions of yuan.
In Hu Xiaohe's view, a new - generation AI company always needs a strong CFO. However, his work is not just about looking at accounts and raising funds. More importantly, he can be deeply involved in business implementation. After all, 30% - 40% of the computing power business has a financial nature. So, it can be seen that Xie Wenqi often goes to the site to supervise the construction of clusters.
In addition to these two core figures, most members of the team come from Tsinghua University's laboratories, and some have experience in cooperative projects with Alibaba and Huawei, combining in - depth technical knowledge with engineering implementation capabilities.
Build the "Next - Generation AI Supercomputer", Achieving Annual Revenue of 500 Million in Two and a Half Years
Different from the scenario of general AI companies that need to burn a lot of money at the beginning, Infrawaves had commercial revenue in the year of its establishment.
"Our technology is a perfect match for AI infrastructure," Hu Xiaohe believes. "AI Infra is essentially a combination of supercomputing and cloud computing."
They had done a lot of work in the fields of cloud computing and high - performance networks during their postgraduate and post - doctoral studies, which could be directly transferred. For example, before starting the business, they had launched the first operator - level Tbps programmable network product in China.
During the entrepreneurial process, Hu Xiaohe deeply realized that "the value of technology needs to be enhanced through a snowball effect". So, from a cluster of a hundred GPUs, a thousand GPUs to ten thousand GPUs, and then to productization and business operation, every step of Infrawaves is like "jumping up a step" - there is a sense of achievement if you succeed, but it's a bottomless abyss if you fail. However, they persisted in continuously breaking through their capabilities and expanding their influence.
In 2023, they signed their first client, SenseTime, starting with technical services and gradually completing the integrated software delivery. Subsequently, they cooperated with 21Vianet, Zhipu, etc., and the business volume increased rapidly.
How to understand Infrawaves' business? Xie Wenqi, the co - founder and CFO, once gave an image answer. He believes that different from real - estate developers focusing on computing power platform operation and decorators focusing on system optimization, Infrawaves is a builder of computing power clusters, focusing on building a high - performance, open - source, and autonomously controllable intelligent computing system covering computing and communication software and hardware.
"We don't make chips, nor do we optimize operators." Infrawaves focuses on three major modules: communication, scheduling, and monitoring, with the goal of building the "next - generation AI supercomputer" - transforming GPU cards into available computing power and constructing a high - performance, scalable, and domestically produced intelligent computing cluster.
Now, Infrawaves has developed a full - stack self - developed high - performance open - source intelligent computing system, Galaxy HPAC (High - Performance AI Computing). The system mainly includes three core products: the high - performance integrated AI network (Mercury - X), the artificial intelligence platform (Venus - AICloud), and the computing power construction and maintenance platform (Venus - AIDOC).
Hu Xiaohe said that the core advantages of Infrawaves lie in cost - effectiveness and openness, and the essence is to achieve "decoupling". Through Infrawaves' self - developed RoCE networking solution, the customer's cost can be reduced by 40%, and the delivery cycle can be shortened from several months to several weeks.
In terms of commercialization, the founding team of Infrawaves also has their own understanding. "Pure software can't be sold at a high price in China," Hu Xiaohe said frankly. The real value carriers that can be recognized are hardware and services - hardware can be priced, and services can be audited. So, they defined the cluster as an infrastructure - level product to avoid the difficulties of monetizing pure technology.
Based on this, Infrawaves' business is mainly divided into two categories:
Large - scale customized clusters: Serve leading customers in the Internet, large - model, and autonomous driving industries, with no more than 30 customers nationwide;
Integrated edge clusters: Standardized software and hardware products, targeting local governments, industrial parks, and information technology application innovation scenarios, and promoted on a large scale through channels.
At the same time, in response to the core requirements of high utilization, high stability, and full - stack domestic production of intelligent computing clusters, they continuously research and develop domestic high - throughput unified communication systems: They launched a fully domestic 102.4T near - optical packaging white - box switch, aligning with the world's highest performance through a combination of multiple technologies. Through technological innovation, the end - to - end computing power utilization of the cluster is increased by 5%, achieving 100% rapid fault diagnosis and automated operation and maintenance.
Now, Infrawaves has delivered a total of 30 clusters and over 76,000 GPUs. The customers include multiple users such as Zhipu AI, SenseTime, operators, data centers, and local state - owned enterprises. It is reported that Infrawaves' revenue in 2025 reached 500 million yuan, and the delivery team only has a dozen people, mainly relying on automated software to improve labor efficiency.
Difficult Early - Stage Financing, Continuous Investment from State - Owned Capital Later
Although it seems easy to talk about now, in Hu Xiaohe's view, "Entrepreneurship is a high - risk endeavor, and early - stage financing is very difficult."
In 2023, when the AI boom was just starting and the capital market seemed lively, Infrawaves' financing was not smooth. Institutions such as Sequoia China Seed Fund and CDH Investments had advanced to the investment decision - making meeting but finally did not invest. The team made an Excel sheet of thirty or forty institutions and followed up with each one, but with little effect. Finally, they only received support from Miracle Plus and the Tsinghua Alumni Fund.
Since then, state - owned capital has participated in almost every round of financing. Government funds from Beijing, Shanghai, Xinjiang, Henan, and other places have successively invested in the company. As of now, Infrawaves has completed multiple rounds of financing. When talking about why the company was favored by state - owned capital at an early stage, Hu Xiaohe told China Venture Capital Daily, "We don't just raise money for the sake of it, but are driven by business needs."
On the one hand, the computing power cluster solution can solve practical problems and drive regional industrial development, meeting the needs of local governments. On the other hand, AI computing power has obvious regional characteristics, and the support of local governments is helpful for business implementation.
He also said that in the future, a large number of edge intelligent computing clusters will be built in hundreds of city - level industrial parks across the country, and the demand for information technology application innovation is huge. The state - owned capital background makes business implementation smoother.
For example, Infrawaves has already launched two clusters in Xinjiang, and cooperation with Henan has also started. With the support of local state - owned capital and policies, it is quickly entering the computing power infrastructure market in various regions.
Under such circumstances and needs, Infrawaves has taken every step steadily. In terms of the financing rhythm, they adhere to a step - by - step approach, striving to keep the equity structure stable and not over - diluting the shares for unnecessary high valuations.
Now, in the financing market, Infrawaves has finally achieved success after going through difficulties. Hu Xiaohe said that different from the early stage of widely seeking investments, at this stage, Infrawaves' financing is more targeted. In the future, as the regional demand for computing power continues to be released, it will bring incremental space for Infrawaves. It will continue to focus on the dual - wheel drive of large - scale customized clusters and edge standardized clusters, strengthening its capabilities in automation, domestic production, and intelligence. At the ecological level, it will also further deepen cooperation with local governments and industrial capital.
In terms of judging the market pattern, Hu Xiaohe said bluntly, "The entrepreneurial window for this track has closed, and perhaps only one or two entrepreneurial companies can succeed in the end. Our real benchmark is NVIDIA's solutions in the CGPU field, and our goal is to become a key player in the field of AI computing power clusters."
This article is from the WeChat official account "Dongshisi Tiao Capital" (ID: DsstCapital), written by Zhang Xue and published by 36Kr with authorization.