"Co-Ji Technology" Completes Multi-Million Yuan Financing for Developing the Grid-Type Computing Power Dispatching Network | 36Kr Exclusive

Author | Qiu Xiaofen

Editor | Su Jianxun

36Kr has learned that "Gongji Technology" has completed an angel round of financing of tens of millions of yuan, led by Shuimu Ventures, with the seed round investor Miracle Works Forum continuing to increase its investment, and Xinjin Venture Capital jointly investing. The FA for this round of financing is exclusively provided by Yuanhe, and the funds will be mainly used for product research and development.

It is understood that "Gongji Technology" is mainly committed to solving the problem of difficult access to computing power resources in the elastic temporary computing scenario, and has constructed a power grid-like computing power scheduling network to use low-cost idle computing power to serve high-value flexible needs.

Trend of Computing Power Demand and Supply

Currently, stimulated by the downstream AI, the upstream intelligent computing centers are continuously being constructed. Industry data shows that as of the first half of this year, China's computing power scale has reached 246 EFLOPS, and the year-on-year growth rate of intelligent computing power exceeds 65%. Among them, there are more than 13,000 computing power application projects in multiple fields such as industry, education, medical care, and energy.

However, the actual operation of the current intelligent computing centers is facing huge pain points. Wang Peng, the marketing partner of "Gongji Technology", told 36Kr that the main contradiction is that the elasticity and low price of computing power cannot be achieved simultaneously.

On the one hand, from the demand side, in the continuously rising computing power demand, the demand for large model inference is huge. But the use scenarios of inference are relatively fragmented and scattered, and the inference demand of 2C applications will have obvious peaks and troughs in time due to the work and rest patterns of users, resulting in insufficient computing power at some times (users queuing) and waste of resources at some times - and according to actual measurements, such elastic computing power demand accounts for more than 10% of the total computing power demand.

According to the prediction of the Ministry of Industry and Information Technology, it is expected that China's computing power market will reach a scale of over one trillion yuan by 2030, and the scale of the elastic computing power market will exceed 100 billion yuan.

On the other hand, from the supply side, in order to avoid the risk of lease interruption and termination, the currently newly built computing power centers mainly provide services to the outside in the form of entire lease. The prices of a small number of elastic and flexible services are often several times that of the usual ones. Although some small manufacturers can provide relatively cheap elastic computing power services, they are actually still in the mode of entire lease on time. Limited by the computing power cost structure, it is difficult to sustain commercially. From a business logic perspective, the centralized self-built computing power centers are not suitable for meeting short-term, fluctuating, and fragmented computing needs.

At the same time, 67% of the global computing power resources are wasted due to idleness every day, and there are 570 million personal computers in China.

Take Internet cafes as an example. Among the 300,000 Internet cafe computers that "Gongji Technology" has currently reached, each computer is used on average for no more than 3 hours per day, and 15% of them are 4090; Take the Steam platform as an example, the total computing power of Chinese game players' graphics cards on its platform is even higher than the total of the intelligent computing center in the Yangtze River Delta.

And such decentralized and idle computing power resources are very suitable for meeting temporary and short-term computing needs - such resources do not need to worry about overstocking and lease interruption, and it is precisely based on this that it is possible to achieve both elasticity and low price.

In order to truly and stably provide computing power services to customers with decentralized, fluctuating, and idle computing power resources, "Gongji Technology" has developed "Power Grid-like Computing Power Scheduling Network" in order to better match computing power resource providers and computing power demanders, and try to solve the problem of the impossible triangle of computing power elasticity, low price, and stability.

It is reported that in terms of computing power supply, "Gongji Technology" has built SLM Task Scheduling and Distribution Platform, which is the only one in the world in both idle-time invocation and one-click access. The platform has integrated idle resources from a large number of personal devices, thousands of Internet cafes across the country, and medium and large clusters.

These decentralized computing power resources have been effectively utilized under the support of "Gongji Technology"'s efficient and accurate prediction and scheduling algorithms. And after nearly a hundred iterations of the product version, this platform is compatible with different network environments and different operating systems, allowing suppliers to one-click access and dynamically exit, ensuring the stability of computing power services without affecting the original business and usage habits.

"Gongji Technology" CEO Fu Zhi told 36Kr that the core capability of the "Calculated Platform" to truly solve the impossible triangle problem lies in the implementation of the energy engineering method in the computing field, achieving computing power load prediction, and achieving a perfect matching and scheduling of supply and demand, thereby significantly improving the overall resource utilization rate. Moreover, through the core technology, the idle or unused resources have realized their value, "achieving a win-win situation for computing power suppliers, demanders, and 'Gongji Technology'."

Currently, "Gongji Technology" has already had many commercial cases landed, and the cooperating AI manufacturers include BibiGPT, Bai Xiong Shu Zhi, Cheng Li Yue Guang, etc. It is reported that the above companies have achieved large-scale revenue and the income is growing rapidly.

For example, with the technical support of "Gongji Technology", BibiGPT has met the needs of millions of users through intelligent summary services. The elastic computing power service of Gongji Technology has effectively responded to the fluctuations in user demand, ensuring the continuity of services and the optimization of costs.

In addition, "Gongji Technology" also based on the ability of decentralized computing power scheduling, first proposed the "Computing Instead of Storage" computing and power coordination implementation plan, replacing power transmission with task distribution, effectively promoting the local consumption of new energy in the microgrid scenario, and significantly reducing the marginal cost of computing. Currently, the team has signed a contract to build the Qinghai Provincial Intelligent Green Computing Power Scheduling Platform.

In terms of the team, "Gongji Technology" CEO Fu Zhi is an undergraduate of Tsinghua University in the class of 2018, a direct doctoral student in the class of 2022, under the tutelage of Academician Jiang Yi, won the Tsinghua University Future Scholar Scholarship (the highest in the school), was selected as a Tsinghua Meiling Science and Technology Innovator, and has had consecutive entrepreneurial experiences.

Recently, this project also won the global runner-up of the China International College Students Innovation Competition and the champion of the graduate student entrepreneurship group. In addition, Fu Zhi said that the company's next round of financing has been initiated, and the expected financing amount is more than 50 million yuan.

Investor Evaluations

Mao Shengbo, Partner of Miracle Works Forum:

Gongji Technology, with the vision of "dynamically combining computing power and energy to build a bridge between information and energy", accurately grasps the pulse of technological development. Its team has successfully scheduled tens of thousands of idle graphics cards, providing efficient, elastic and low-cost computing power support for AIGC enterprises and scientific research institutions. This not only pioneers a new business model of computing power sharing, but also explores a cutting-edge path for the optimal utilization of resources. I am full of confidence in the future of Gongji Technology and firmly believe that they will lead the development of inclusive computing power and make important breakthroughs in promoting the efficient allocation of global computing power resources.

Yan Siyu, Executive Director of Shuimu Ventures:

The Gongji Technology team, with its vitality and innovation, accurately grasps the pulse of shared computing power innovation, injecting new vitality into the industry. Their persistent pursuit of technical details provides a practical and feasible computing power support solution with elasticity and low cost for the AIGC and scientific research fields. I am very optimistic about this team and believe that they will promote Gongji Technology to achieve greater breakthroughs in the field of computing power services.

end