Author|Morning Light
Editor|Chen Qiu
Operations|Chen Xiaoyan
Mirror (ID: DMS-012)
"The AI market opportunity has arrived, and it will definitely bring a wave of market demand and opportunities that will last for more than 10 years, or even greater than in the past 10 years." Lin Yuan, president of Qingyun Technology, said to another mirror.
Founded in 2012 and opened public cloud services in 2013, Tsingyun has developed for nearly 11 years, during which it has experienced three market opportunities - from mobile Internet and digitalization, to the localization of information and innovation, and now to AI and digital intelligence. In line with the development trend of the industry, changes in market demand and customers, Tsingyun's own products and services have also been iterated.
When it comes to AI, since OpenAI launched ChatGPT at the end of last year, it has set off an AI boom at home and abroad. According to a set of data released by research institute PitchBook, the investment circle invested a total of US$1.37 billion (equivalent to about 9.369 billion yuan) in generative AI companies in 2022, almost reaching the total of the past five years.
In 2023, Microsoft's push and Google's entry into the game have continued to push up the popularity of GhatGPT, and domestic technology giants have also successively taken this bus, including Ali, Baidu, Tencent, JD.com, Huawei, Byte, iFLYTEK, 360, SenseTime, etc. have released large models.
In July this year, at the sixth World Artificial Intelligence Conference, more than 30 large models unveiled by major domestic manufacturers, involving large languages, industry, finance, automobiles, medical and many other fields.
Not only that, the boom of AI has also ignited the field of "humanoid robot + AI large model", and technology giants such as Tesla, Google, Microsoft and Samsung Electronics have seized the layout.
In Lin Yuan's view, from the perspective of customer and market demand, in the future, enterprises will invest in AIGC as much as the entire IT investment will become higher and higher. When customer demand is strong, the requirements for the underlying computing power of the enterprise will become higher and higher. "The demand for computing power will be 10 times, 100 times, or even more than before."
Taking NVIDIA's latest earnings report as an example, in the second quarter of this year, data center revenue was $10.32 billion, up 171% year-on-year and 141% month-on-month, more than 29% higher than analysts' expectations of $7.98 billion, and up 14% year-on-year and 18% quarter-on-quarter in the first quarter.
In this regard, NVIDIA said that data center revenue mainly comes from cloud service providers and large consumer Internet companies. The strong demand for NVIDIA's HGX platform, based on Hopper and Ampere architecture GPUs, is mainly driven by the development of generative AI and large-language models.
From the perspective of computing power, a large number of computing centers have sprung up like mushrooms, which is still a stage of centralized construction. However, at present, the pain point of the imbalance between supply and demand of computing power is still difficult to solve, and the report released by the International Data Corporation (IDC) shows that the global data volume is growing by about 60% per year, but the annual growth rate of computing power is only 10%.
At the same time, the cost of computing power is high, especially in the eastern region, and some small and medium-sized enterprises are difficult to bear the high cost when facing AI computing power scheduling.
In order to meet the needs of future users for intelligent computing power, after nearly two years of operation and optimization, Qingyun launched an AI computing power scheduling platform. So what are the advantages of Qingyun's AI computing power scheduling platform? What are the market prospects in this regard? What are the changes to the cloud vendor landscape?
The following is a condensed transcript of the conversation:
Asset-light entry
Q: Qingyun launched the AI computing power scheduling platform, can you talk about the investment in computing power?
Lin Yuan: The reason why the AI computing power scheduling platform was launched is because if Qingyun wants to fight for investment like a large factory, it is impossible, we can't invest much money, but there will be investors to invest, you can think that we are jointly operating the AI research scheduling platform with other partners, that is, we invest in technology and products, and operate the AI computing power cloud in a relatively asset-light manner.
Q: What should I do if the computing power resources are insufficient?
Lin Yuan: The core problem of insufficient use is the supply of GPU chips, and now all manufacturers are facing the same problem. But everyone's solution is different, and the solution of large factories is to buy by money. Qingyun's solution is that at present, many companies have increased investment in computing power and have their own supply channels, and they have demands on our computing power platform, that is, many people are helping us invest in the underlying computing power.
The core of the AI computing power scheduling platform launched by Qingyun is the scheduling platform, which does not prevent us from providing Qingyun's AI computing power cloud to customers together.
Q: Qingyun has investors to undertake the computing power level, so what are the types of investment we currently have?
Lin Yuan: The first category is local state-owned assets and local central enterprises, such as those companies in the energy category, which originally wanted to convert electricity into computing power, and had a very strong will;
The second category is local governments, because local governments need new tools to gather industries and activate GDP, so they are also willing to do this. There are direct investments, and there are also cooperation with central enterprises or banks;
The third category is ecological partners, including companies that make large models and AI chips. They do a lot of computing centers everywhere, but they don't have a platform, they want to sell their cards or integrations, but they need a standard platform and they like to work with neutral vendors. In fact, we are also their channel, and together we provide AI computing power cloud services.
Q: What is the position of Qingyun in the ecological alliance?
Lin Yuan: We are divided into technical ecology and business ecology. In terms of technology ecology, the first layer is the GPU. Because in the technology stack we do, there must be GPUs, which are heterogeneous parts. Including mainstream domestic GPUs, but also foreign GPUs. Of course, we are talking about AI, so at this stage we will only talk about GPUs;
The second category is the model ecology, which needs to have Model as a Service, which has the mainstream commercial model that everyone can see, as well as the open source model. In the middle of the model and the chip, it is actually the computing power scheduling layer. The computing power scheduling layer not only relies on Qingyun, but also many manufacturers are doing model acceleration and model drive, including many Tsinghua series and Wuwen core dome, which is the acceleration of model landing.
In terms of business ecology, it is the people who can get the computing center project, the people who invest in construction, and those partners who can naturally help you sell, he has customers, he has AI end customers, etc.
It's an incremental market, not a stock market
Q: At present, AI computing power is strong, the demand for GPUs is surging, and the price of NVIDIA GPU boards is getting higher and higher. Qingyun has increased its layout in terms of computing power, can you disclose how much revenue it has brought in this regard? What are our expectations in terms of AI revenue?
Lin Yuan: The revenue issue cannot be said in advance, everyone can only look at the public financial report, but basically this trend: AI computing power is now more training-based, and later we hope that it will enter the stage of reasoning. But only in terms of training, the growth rate should still be expected. The GPU-based computing power cloud in 2023 is the same as the CPU cloud in 2013. From 2013 to 2017, basically all companies grew at 100%.
Q: How big is the current market size of overall computing power scheduling? You have predicted the size of the AI computing power market, including everyone's demand for the AI research market, but we are only doing computing power scheduling, what is the market prospect in this regard?
Lin Yuan: From the perspective of revenue, on the one hand, AI computing power scheduling is similar to the original traditional delivery model; On the other hand, there is the AI computing power cloud, which has both a product part and a service part.
With the scale of Qingyun, compared with the past 10 years, the market development space in the next 10 years will be larger, at least 10 times and 100 times the space. And we can do at least more than we could get in the last stage. It will definitely become the main revenue in the future, but it will take time.
Q: What changes will the AI computing power scheduling platform bring to the existing cloud vendor landscape? How does Tsing Cloud position itself in the future?
Lin Yuan: AI will definitely bring changes to the current cloud vendors.
From the technical level, first, the demand of AI for the underlying technology and the mode of cooperation, it is different from the previous stage; Second, since it will bring change, and all people are on the same starting line, the market will bring certain variables, but it cannot be said to be a reshuffle; Third, everyone is moving forward, and many companies are making their own big models. We don't make big models ourselves, but we are deeply engaged in AI computing power scheduling platforms. Therefore, in the face of new AI application needs, we need to accumulate a lot of technology.
In addition, in terms of capital, large factories can invest and build by themselves, but now AI does not lack investors and builders, but lacks operating platforms. To sum up, whether on the technology side or the capital side, AI is the only emerging market, which is an incremental market, not a stock market. Now in the incremental market, everyone is not busy with competition, but on how to better serve customers and solve the problem of computing power.