When it comes to the hottest concept in 2023,AI, computing powerPresumably it should be on the list.
With the surge of a new wave of artificial intelligence in the world and the AI models of major giants competing to land, it is no exaggeration to say that computing power has become the core competitiveness of the AI era.
Recently, a derivative concept of "computing power" has also come to the "front of the stage", and TA is"Computing power leasing".。Microsoft, Amazon, Google and other leading cloud bigwigs have started the business of computing power leasing, and once a listed company announces a cross-border entry into the field of computing power leasing, it will also be attractedCapital attention and stock price changes。(Reference**: Finance Associated Press, 2023.)08.24）
Let's talk about the topic of computing power leasing today: why should computing power also be "rented"?How big is the "cake" of computing power leasing?
Why should computing power also be "rented"?
Let's answer the first question first: why should computing power be "rented"?
The logic is actually very easy to understand:Computing power is very important, but at the same time it is very expensive。Taking this year's popular AI model as an example, its operation requires a lot of computing power, according to Guotai Junan's calculations, the computing power required to train a GPT-3 model with 174.6 billion parameters is about3640pflop/s-dayThe cost of completing a training session is more than 100 million yuan。(Data**: Finance, 2023.)04.20；Economic Reference Network, 202311.14）
Large companies with deep pockets can have huge computing power through "buying, buying, buying", but most small and medium-sized enterprises have limited financial resources and are shy in their pockets, and they cannot reach computing power in the current situation where the supply of high-performance GPUs exceeds demandSo,"Computing power leasing".Born, this is a kind of passLease computing resources(such as graphics processing CPUs) to provide computing power. Large companies with GPU computing power have started the business of renting computing power, and small and medium-sized companies that urgently need computing power and have relatively limited funds can temporarily meet their own computing power needs through leasing, so as to achieve a "win-win" situation.
How big is the market space for computing power leasing?
While solving the rigid needs of a large number of technology enterprises, computing power leasing has also become a new one"**Track".。Looking ahead, how big is the market space for computing power leasing?Let's sum it up with two key words.
The first key word is:"There is a huge demand for AI computing power."。How many people lease computing power depends largely on how much computing power is in demand. We have introduced the importance of AI many times before, and I believe you can also experience it in your daily life and workAI empowers and transforms all walks of life。With the rise and increasing demand for AI applications such as autonomous driving, smart home, security monitoring, robotics, medical equipment, and smart classrooms, the demand for AI computing power continues to grow. According to the data of the Ministry of Industry and Information Technology, the scale of China's computing core industry will reach in 20221.8 trillion yuan, the total scale of computing power is in the ranksSecond in the world。With the development of artificial intelligence technology, China's computing power demand will maintain rapid growth in the future, creating a broad space for the development of the computing power leasing market. (Data**: Xinhuanet, 2023.)07.18；People's Republic of China**People**, 202310.10；**Times.com, 202305.29）
The second key word is:"Computing power leasing is profitable".。Can hashrate leasing make money?The answer is yes. There are domestic institutions that have made a simple calculation of computing power leasing income, according toComputing power leasing revenue = computing power investment A800 or H800 unit price * corresponding to the computing power of each piece * single P computing power rentDerive the income of a computing power leasing company. Based on the data in September 2023, the gross profit margin of the leased A800 is about 40%, and the net profit margin is aboutaround, while leasing H800 will be more profitable. There are also foreign institutions that estimate that in the cloud computing power leasing service provided by Microsoft, each NVIDIA A100 GPU can earn an average of 1. per hour$5, which is equivalent to earning every month$1000Left, based on the financial model, the gross profit margin generated by Microsoft's GPU computing power leasing business is as high as 42%. Intuitively speaking, the business of computing power leasing is very profitable. (Data**: Soochow**, 2023.)09.27；Finance Associated Press, 202309.20；One Zero Society, 202309.29；IT House, 202309.20）
The third key word is:"The scale of 100 billion can be expected".。AI is still at the beginning of a new cycle, and the computing power leasing industry is naturally in the early stage of developmentTens of billions of large marketThere is great potential for future development. Some institutions are optimistic that the scale of the AI computing power leasing market is expected to exceed that of the ordinary computing power service market, and the scale may reachHundreds of billions of dollars。(Data**: China Economic Weekly, 2023.)11.24；Soochow**, 202309.27）
Computing power leasing has become a new outlet in the market
In the context of the advent of the AI era, computing power has become the era of digital economyCritical productivityIn the case of GPU** shortage and global computing power shortage, the computing power leasing industry has become a new outlet. So how do investors look for opportunities?
At present, there are mainly the following types of companies in the AI computing power leasing market: First, the three major operators and Internet giants such as Alibaba and Tencent are representativesTraditional cloud computing service providers；The second isA central state-owned enterprise with IDC (Internet Data Center) construction and operation capabilities；The third isPrivate enterprises with capabilities related to the construction and operation of IDC (Internet Data Center).；Four areManufacturers who have crossed over into the computing power rental market。(Reference**: Soochow**, 2023.)09.27；This article is not recommended).
Risk Warning: This material is not intended as any legal document, and all information or opinions expressed in the material do not constitute investment, legal, accounting or tax final operation advice, and our company does not make any guarantee for the final operation advice on the content of the information. Under no circumstances shall the Company be liable to any person for any loss arising from the use of any content in this material. China's first operation time is relatively short, and it cannot reflect all stages of development. The market is risky, and investors need to be cautious.