CATL has invested in a post-2000s talent from Tsinghua University.

Post-2000s entrepreneurs have stepped onto the center stage.

PE Daily has learned that Gongji Technology recently completed a new round of financing of several hundred million yuan. The investor lineup covers national-level funds, industrial capital and multiple market-oriented institutions, including China Internet Investment Fund, Ningbo Wending and Puquan Capital affiliated to CATL, Yuanhe Puhua, Saime Technology, Deshi Investment, Qiheng Co-Creation, etc. Existing shareholders Chunhua Venture Capital, Lihe Zhongke and Xinjin Venture Capital continued to increase their investment.

Founder Fu Zhi was born in 2000. He graduated from Tsinghua University with a bachelor's degree, and founded Gongji Technology during his doctoral study at Tsinghua University. The company entered the computing power business starting from distributed computing power scheduling: on one end, it accesses GPU resources scattered in different regions and different clusters, and on the other end, it provides on-demand callable computing power for model training, inference and AI applications.

Fu Zhi compares this business to building a "power grid". Computer rooms across the country are like power plants, and Gongji is responsible for connecting and scheduling the originally fragmented resources, so that the computing power can be "used by the owner in peak hours and connected to the shared grid in off-peak hours". Since its establishment three years ago, Gongji has served more than 10,000 enterprises and teams. According to the company's disclosure, its annual revenue growth rate exceeds 800%; in 2026, quarterly revenue has doubled consecutively, with ARR exceeding 200 million yuan.

With the business exploding, business cooperation opportunities and investors coming to the company have increased significantly. After years of entrepreneurship, Fu Zhi has always held a firm belief: "If something benefits the whole world, we must pursue it." A story about the post-2000 generation breaking through in the computing power track thus unfolds.

A Post-2000s Tsinghua PhD

Breaking Into the Distributed Computing Power Track

Fu Zhi's entrepreneurial idea originated from several biographies of entrepreneurs in his mother's small shop. At that time, he was still studying in Tongren, Guizhou. His firmness beyond his peers had already begun to show. When his score dropped sharply in the final mock exam before the college entrance examination, he felt relieved instead. The bad result was like a fault list: since the loopholes had been exposed, he could start to fix them. Two months later, he received the admission letter from Tsinghua University.

After entering Tsinghua University, entrepreneurship quickly turned from an idea into action. In his junior year, Fu Zhi and several classmates decided to develop a low-code game engine to help ordinary people with game inspiration make up for their development capabilities and produce games. It coincided with the metaverse boom, the valuation of the project rose rapidly, and it received 10 million yuan in financing, even turning down the invitation of well-known investors.

No one expected that the capital market would cool down the next year, and the project was suspended. Fu Zhi calmed down to review and realized that although the product lowered the development threshold, what was really scarce was high-quality ideas that could be monetized. Creators could not create commercial value by making games, so the platform could not achieve a real commercial closed loop. "We can only do the right thing well, and it is impossible to make the wrong thing right."

Later, Fu Zhi shifted his focus back to campus. During his doctoral study, he studied under academician Jiang Yi, and his research topic focused on distributed energy: how to connect scattered photovoltaic, energy storage and electric vehicle batteries into the same system for unified scheduling across different times and locations. This idea of "peak shaving and valley filling" in the energy field was later applied by him to computing power.

The new starting point was hidden in an awkward experience of scientific research. In order to train the neural network model, Fu Zhi needed temporary computing power. The solutions provided by cloud vendors were not suitable for short-term, small-scale demands, and it was not cost-effective to purchase equipment by himself. Finally, with several personal computers pieced together from various sources, Fu Zhi completed this scientific research task. After that, he built a mutual aid platform on campus to match computing tasks with idle computers, and the number of participants soon reached hundreds.

This time, Fu Zhi did not rush to start a business. He spent more time observing, thinking, polishing the MVP, and verifying whether scattered equipment could provide stable services for fragmented demands. Within five months, the platform undertook hundreds of paid tasks, and nearly 10,000 people spontaneously applied to connect their devices to the platform. In 2023, he reached a consensus with Tsinghua alumni including Xu Zhongziheng, and Gongji Technology was officially established.

At that time, when the computing power industry was constantly chasing the construction of larger clusters, Fu Zhi looked at the other end: there were a large number of underutilized computing power resources in the already built computer rooms. Therefore, Gongji did not choose the path of building its own data center, but thought about how to connect the idle computing power of computer rooms across the country to the same platform to support diverse and fragmented computing power demands.

Inside Gongji, this business model is summarized as the "Ziheng Curve": the shorter the computing power lease term on the demand side and the higher the suddenness, the higher the premium per unit time; the more fragmented the idle equipment on the supply side, the more difficult it is to be bought out by traditional large customers, and the lower the willingness of asset holders to set prices. The difficulty lies in delivering stable services with fluctuating supply.

For Fu Zhi, that right thing is becoming concrete: building a distributed computing power network, so that scattered computing power can flow freely and be matched accurately.

Going Deep Into the Industry

The Financing Story

"You are too young."

In the early stage of financing, Fu Zhi often heard this sentence. He was not afraid of being questioned, and even enjoyed the confrontation of views: whether the technology can be realized and whether the cost is lower can be calculated item by item, but age is the only thing that cannot be debated.

At that time, the demand for large model training exploded, and almost all the capital in the computing power track poured into large-scale training clusters. The computing power business familiar to investors was still an asset-heavy path: building data centers, purchasing GPUs, and then leasing computing power to large customers for a long time. Few people paid attention to Gongji's plan to do asset-light distributed computing power scheduling, and doubts also fell on the team: a few post-2000s, can they make such a business a success?

A few months after the company was established, the financing did not progress for a long time. Although the team had insight into the demand for computing power scheduling and completed preliminary verification, it still needed R&D investment to polish the technical solution into a product that can be delivered in a standardized way. Fu Zhi used his savings to keep the team running, and the partners even borrowed money to buy servers. In this way, they gritted their teeth and survived the difficult stage from prototype to commercialization. Until August 2023, Qiji Chuangtan extended an olive branch. Fu Zhi recalled that Gongji's first customer also came from the Qiji alumni network.

Xinjin Venture Capital participated in Fu Zhi's earlier entrepreneurial journey. This institution invested in his first entrepreneurial project, and when it came to Gongji, it appeared in the shareholder list again, and helped the team connect to the Qinghai Province computing power scheduling project.

This is a leap forward to go deep into the industry. Undertaking national-level and large-scale computing power platform projects has pushed this young team to more complex engineering sites. Gongji's solution is to introduce external forces, build a team of senior industry expert consultants, provide professional support for technical solutions, engineering implementation, compliance standards and other aspects, and incorporate external resources into the project system through a clear cooperation mechanism. Youth means stronger learning ability and the execution to take on tough tasks, coupled with the ability to mobilize top external resources, the team quickly makes up for the boundary of experience.

Since then, relying on distributed scheduling solutions and engineering collaboration mechanisms, Gongji has successively participated in computing power projects in Hebei, the Yangtze River Delta, the Guangdong-Hong Kong-Macao Greater Bay Area and other regions, and deeply participated in the construction of the national integrated computing power network.

In the view of Xinjin Venture Capital, Gongji's performance in solving problems in real business has already given the answer. They described this team as: "Very young, but capable of dealing with real, difficult and important problems."

With the gradual implementation and expansion of the business, the financing rhythm has gradually become clear. In October 2024, Gongji completed an angel round of tens of millions of yuan, with investors including Shuimu Venture Capital, Qiji Chuangtan and Xinjin Venture Capital; in December 2025, the company completed a Pre-A round of financing of nearly 100 million yuan, with investors including Chunhua Venture Capital, Beyond Era, Lihe Zhongke and Alpha Startup Club.

In October this year, Gongji Technology completed financing of several hundred million yuan, with investors including China Internet Investment Fund, Ningbo Wending and Puquan Capital affiliated to CATL, Yuanhe Puhua, Saime Technology, Deshi Investment, Qiheng Co-Creation, Chunhua Venture Capital, Lihe Zhongke, Xinjin Venture Capital, etc. So far, Gongji's shareholder camp has further expanded to national-level funds, industrial capital and multiple market-oriented institutions.

PE Daily sorted out and found that there is an earlier clue for this investment.

As early as 2024, Gongji stood out from 5.12 million participating teams and won the runner-up of the China International College Students' Innovation Competition. Zeng Yuqun was one of the judges of that year's competition.

In the early stage of entrepreneurship, when Fu Zhi met with investors, the topic extended from the competition field to where the future computing power demand would flow. Large model training still relies on centralized clusters. With the popularization of AI applications, huge inference demands will be scattered in different regions and nodes. The two sides' judgments on distributed computing power also converged here.

Having idealistic sincerity and sophisticated behavior is not contradictory in him.

Fu Zhi used Lu Qi's words to understand entrepreneurship: willpower is related to ideal and inspiration, and ability solves problems one by one; mental strength allows the two to go through a long period of uncertainty.

Earlier this year, this mental strength encountered a new test. When the computing power market suddenly heated up, and voices of financing, valuation and listing poured in at the same time, Fu Zhi once hesitated whether to follow the market rhythm to speed up the pace. Investors advised him to follow his own rhythm: "All those things will return to the essence in the end."

After the noise passes, the problem will eventually return to the essence: whether this distributed computing power network can continuously undertake the evolving demands under the wave of the artificial intelligence industry in rounds of global scientific and technological revolutions.

How a Distributed Computing Power Network is Deployed

The value of distributed computing power first appears at the moment when traffic suddenly surges.

Fu Zhi has an almost instinctive sensitivity to this. As early as 2024, when he first met the technical person in charge of a leading domestic AI creation community, he predicted that the other party would definitely encounter computing power queuing in the future. A year later, the other party came to him with the problem as expected. After testing for a period of time, Gongji's existing solution failed to fully meet the other party's needs. The team was a little frustrated, and even advised the other party to give up, but the other party said: there is no other alternative.

The team continued to look for solutions, and finally got inspiration and technical guidance from a professor in the Department of Mechanics of Tsinghua University, designed a scheduling solution adapted to its business characteristics: during the peak period when the community traffic suddenly surges in the daytime, schedule computing power resources on demand to undertake the instantaneous explosive inference tasks, and release the computing power during the off-peak period at night. The solution was successfully implemented, and the other party did not need to lease equipment for a long time for the peak hours, so the computing power cost dropped by more than half. This has also become a long-term large-scale practice of Gongji's distributed computing power scheduling technology in high-fluctuation inference scenarios, enriching the service implementation experience for tidal inference scenarios.

An even more extreme challenge occurred when serving Remy, a 3D generation AI application for creators. At that time, Remy became popular quickly after being launched following the Hongmeng system launch event. Within 48 hours, the number of users of this 3D AI application exceeded 500,000. After the business exploded, the computing power demand quickly increased from the level of hundreds of cards to nearly 2,000 cards. Gongji completed resource scheduling and launch in 96 hours; after the traffic receded, the resources were released synchronously to serve other computing power demands, so customers did not have to bear idle costs for the peak demand.

During the business review, the team summarized it into two typical types of elastic computing power demands: one type is cyclical customers with regular fluctuating traffic, and the other type is event-driven burst customers with instantaneous peak surge. These two types of demands for computing power elasticity are widespread in the AI industry. For more than three years, Gongji has served more than 10,000 enterprises and teams. Partners include leading AI companies such as Zhipu, FaceMind and VAST, as well as large enterprises and scientific research institutions such as Tencent, China Mobile and Tsinghua University.

In order to continuously and stably undertake computing power demands, Gongji has established a complete scheduling mechanism: real-time monitoring of changes in supply and demand, grading according to the computing power and stability of nodes, and then matching resources according to the requirements of tasks for latency and scale. Before the peak arrives, the system reserves redundant computing power in advance; once a node goes down, the service is quickly switched to other clusters, and further realizes "self-use during peak hours, grid connection during off-peak hours" of resources on the customer side.

The most tricky thing is the state continuity during the switching process — the inference context must be continued accordingly. Gongji relies on distributed storage to synchronize the context between remote clusters, so that inference tasks can run in relay in different regions, even under different chip architectures. The SLA of this asset-light scheduling platform has exceeded 99.9%.

In Gongji's judgment, in the CPU era, cloud services solved the elasticity problem; but in the GPU era, real cloud services have not yet taken shape. Centralized clusters are still more suitable for centralized training of large models, while tidal inference tasks need to be called and consumed at edge computer rooms at any time like water flow.

Distributed computing power will not only stay in revitalizing idle stock computing power. Fu Zhi believes that with the growth of inference demand, more computing power nodes will be distributed in cities, close to users, and carry application traffic nearby. That is to say, computing power is becoming a new type of infrastructure in the AI era, which means that whoever can take the lead in connecting enough nodes, complete the adaptation of different devices, and maintain stable services in fluctuations, is likely to seize the entrance of this new type of infrastructure.

At present, Gongji's computing power network has accessed more than 700,000 devices from individuals and Internet cafes, and more than 50 data centers. The team's next stage goal is to expand to thousands of nodes. As for how to deploy this network, Fu Zhi once confessed in a TV program that the most urgent thing at present is team building.

Looking for peers who can take charge of their own work alone is the only way for many companies to grow. Fu Zhi values doing practical things, and also cares about where this ability is ultimately directed. "Do not pursue things that benefit only yourself, but must pursue things that benefit the whole world" is his self-motivation and the underlying principle. Leading a good team is another level of practice. In the early stage of entrepreneurship, he compared himself to Sun Wukong, used to put out fires everywhere on the front line, but now he can better understand Tang Seng: there is no need to do everything personally, the most important thing is to anchor the long-term direction and lead the team to move forward.

For the definite end point of the computing power network, Fu Zhi is used to using the energy system as a metaphor: large data centers are like centralized power plants, and small computing power nodes are like scattered photovoltaics and energy storage. Relying on unified intelligent scheduling, large and small nodes can collaborate and link, and finally converge into an elastic distributed computing power network.

A truly mature infrastructure will eventually hide its own huge scale and complexity, and quietly and reliably support the free growth of the world. Just like when you turn on a lamp, few people think of the power plant far away. The underlying value is hidden in the silent supply.

This article is from the WeChat official account "PE Daily" (ID: pedaily2012),