Behind the global computing power map, a fierce competition for computing power

Source: Digital Intelligence Frontline

Image source: Generated by Unbounded AI tool

This year, one of the most talked about stories in the field of computing power is——

On May 29, Nvidia founder Jensen Huang just wore an iconic leather jacket and announced to the world, “We have reached the tipping point of generative AI. From then on, every corner of the world will have computing needs.” One day later, Nvidia’s market value exceeded one trillion US dollars, and it became the “water seller” behind the big model boom.

Of course, not only Nvidia, but also in the computing power industry chain, including Inspur Information, Cambrian, ZTE, etc., one counts as one, and the stock price has nearly doubled from the beginning of the year. No one doubts that the big model will bring an unprecedented push to the computing power market.

But in fact, long before the big model, the computing power market has been booming for many years, and it is regarded as as critical as water in the agricultural age and electricity in the industrial age. After all, the pull of computing power to the economy is not so great. Recently, the “2022-2023 Global Computing Power Index Evaluation Report” jointly released by IDC, Inspur Information, and the Global Industry Research Institute of Tsinghua University shows that every time the computing power index of a country increases by 1 point, the digital economy will grow by 3.6‰, and the GDP will increase by 1.7‰**, and this trend is expected to continue from 2023 to 2026. Another more direct correspondence is that every dollar invested in IT spending can drive $15 of digital economy output and $29 of GDP output.

This is the third time these companies/institutions have released the global computing power index report. In addition to the basic pattern of global computing power and the steady growth effect of computing power on the economy, some new trends have also attracted industry attention.

China and the United States have led the way for three years, and the chasers are fierce

In the global computing power market, the United States and China are still the most powerful.

Since the release of the first report in 2020, the United States and China have led the way for three consecutive years** in the global computing power rankings**. In the case of a full score of 100, in 2022, as the only two countries with a score of 60, the United States has increased its score by 5 points to 82** by relying on the substantial increase in computing power investment of ultra-large-scale Internet giants**. However, China has been hindered by repeated epidemics, and its computing power investment has slowed down, but the overall growth rate is still higher than GDP, and the ** score increased by 1 point to 71 points**.

As for the other dozen or so countries on the list, they are still in the echelon of chasers (40 to 60 points) and starters (below 40 points), and there is still a distance from the “passing line” of 60 points. For example, the difference between Japan and China in third place is 13 points, which is 58 points.

For countries in different echelons, the gap in scores also means a huge gap in the economic promotion effect of increasing investment in computing power. “The two are not a slanted line, but a parabola that accelerates upward.” Zhou Zhengang, vice president of IDC China, introduced that when the national computing power index reaches above 40 points and above 60 points, the index increases by 1 point, and the impetus to GDP growth will increase to 1.3 times and 3 times when the index is below 40 points respectively.

However, there are various signs that the competition among the three echelons is becoming increasingly fierce. As of 2022, among the 15 countries shortlisted, the starter camp is shrinking. For example, India has risen to eighth by virtue of its investment in computing power and emerging technologies, from starters to chasers.

Singapore and Ireland were included in the evaluation of countries for the first time, which mainly depends on the investment in cloud computing and data centers. For example, Singapore directly ranked fifth with 55 points this time. Behind this is that in the past few years, many cloud vendors around the world have set up their data centers in Singapore to complete the radiation to the entire Southeast Asia.

The catch-up countries, mainly European countries, are narrowing the gap with China and the United States in terms of AI computing capabilities**. For example, the artificial intelligence server markets in Germany, Britain, France, Japan, and South Korea all grew by more than 40% year-on-year.

Why is it said that China’s computing power is turning to high-quality development

Observing the characteristics of China’s computing power market, Li Donghong, vice president of Tsinghua University’s Global Industry Research Institute, told Data Intelligence Frontline that they found in the survey that there was a prominent change this year-several core indicators have undergone significant changes. Although the growth rate of China’s general server market has dropped from double digits in previous years to single digits, computing efficiency and application levels have improved significantly. This means that China’s computing power industry has moved from a period of rapid growth to a new stage of high-quality development.

For example, in the application level sub-item, China’s score increased from 70 to 72. This sub-item mainly involves investment in software, hardware and services related to the five emerging technologies of big data, artificial intelligence, Internet of Things, blockchain and robotics. Especially in the Internet of Things and Robotics, China scored more than the United States and ranked first. On the contrary, artificial intelligence and big data still need to be supplemented.

Computing power efficiency is also a key indicator, which means that under the same computing power, more efficient ones can do more things. The improvement of computing efficiency is generally contributed by cloud computing, new technology penetration rate and intensive data center construction**.

China can see significant improvement in these factors. For example, cloud computing, Li Donghong said that the scale of China’s cloud computing market will reach 455.24 billion yuan in 2022, a year-on-year increase of 33.5%. In the past, enterprises often used an extra server for each application, but now everyone is using cloud technology or other scheduling technologies to make the use of the entire server more efficient.

From the perspective of industry insiders, green development is indispensable in the high-quality development of computing power. In 2021, China’s average PUE (data center energy efficiency index) is 1.55, but by the end of 2022, a total of 153 national green data centers have been built, and the PUE of large and larger data centers planned to be built will drop to 1.30. The liquid-cooled server market is driven, with a year-on-year growth rate of 305.2% in 2022.

Due to the increasingly complex application scenarios, China’s innovation in multi-computing power is accelerating. Arm-based servers are a branch that cannot be ignored. In 2022, the Chinese market will grow by 138% year-on-year. In order to meet the needs of AI workloads, more and more servers use acceleration cards such as GPUs, FPGAs, and ASICs. According to IDC statistics, in 2022, China’s acceleration server market will increase by US$4.40 billion compared with 2019, and half of the increase in the server market will come from acceleration servers. “Future computing power must be diversified.” Zhang Dong, vice president of Inspur Information, said. The industry has seen that Inspur Information has created a variety of computing power platforms for different applications such as “cloud computing, big data, edge computing, artificial intelligence, AI4S (Artificial Intelligence for Science, that is, artificial intelligence for science)”.

Zhang Dong told Data Intelligence Frontline that when building computing power, enterprises must balance software and hardware. They not only need to buy hardware, but also need to invest in software, such as basic software such as operating systems, virtualization, and cloud, so as to fully utilize the equipment.

Manufacturing computing power surpasses the financial industry

In 2022, the biggest “dark horse” on the industry computing power rankings is the manufacturing industry.

This year, the computing power level of the manufacturing industry increased by 3 points in one fell swoop. With a score of 62, the financial industry surpassed 61 points and became the industry with the second highest level of computing power development after the Internet. In terms of growth, the only ones that can match it are the government and the education industry. Both of them will become the industries with the fastest growth in computing power in 2022 with an added value of 7 points.

It is understandable that the Internet industry continues to sit firmly at the top. After all, computing power itself is an important production tool for its income generation. However, it is somewhat surprising that the financial industry, which has always been more advanced in digital technology, was overtaken by the manufacturing industry. Why did the manufacturing industry achieve a “counterattack”?

People in the industry generally believe that this is not unrelated to the background that all countries are vying for the commanding heights of manufacturing powers and the continuous deepening of the digital transformation of the manufacturing industry.

In recent years, developed countries have implemented “re-industrialization” strategies one after another to encourage the return of domestic manufacturing enterprises. China also proposed for the first time in the “14th Five-Year Plan” the goal of “keeping the proportion of manufacturing industry basically stable”. At present, many provinces and cities have taken action. For example, Guangdong proposed that the added value of manufacturing industry should account for more than 35% of the regional GDP by 2027, which is a step further than the goal of “accounting for more than 30% by 2025” proposed two years ago.

At the same time, the digitalization of the manufacturing industry is moving from the previous operation and management end to core scenarios such as R&D and manufacturing processes. The “Report” also pointed out that in 2023, the digitalization of enterprises will usher in an inflection point, entering the era of digital business from the era of digital transformation. This stage will have several key features, such as being supported and promoted by the CEO and business executives, using digital technology for business competition and innovation. **The core is to give full play to the value of data elements and realize the commercialization of data. **

Specific to the sub-sectors of the manufacturing industry, especially new energy vehicles, semiconductors and other industries with “fast development and rapid market changes” are the most active. A typical example is Geely Automobile. In July 2022, Geely built the first multi-computing intelligent computing center in the domestic automobile industry in Huzhou, Zhejiang Province - Xingrui Cloud Intelligent Computing Center, which was officially opened in February this year. Geely’s overall R&D efficiency has increased by 20%.

Sany Heavy Industry, another leading equipment manufacturing company, with the assistance of Inspur Information, has created an intelligent computing solution for global factories and R&D centers, opened up the AI business process integrating cloud, edge, and end-side integration, and can support more than 75 automated production systems and connect tens of thousands of edge devices.

Enjie, the world’s largest wet-process lithium battery separator manufacturer, also uses the Diana intelligent manufacturing all-in-one machine jointly developed by Inspur Information and Zhanwan Technology, based on edge computing and industrial Internet of Things technology, to collect and process data such as equipment operation, production line production, and product quality in real time, improve the quality and efficiency of production line production, and assist business decision-making.

Computing power has brought a good input-output ratio to the manufacturing industry. The “Report” shows that among the top 30 manufacturing companies in the world, every dollar invested in IT can drive a revenue output of $45 and a profit output of $6, which is much higher than that of other industries. A more intuitive data is that, benefiting from the deepening of digital transformation, in 2021, the added value of my country’s manufacturing industry will account for 27.4% of GDP, which is the first recovery after the continuous decline in the proportion of manufacturing industry in the past ten years**, and will continue to grow positively to 27.7% in 2022.

Du Yanze, research manager of IDC China, believes that the key to the digital transformation of the manufacturing industry in the future can be summed up in two words-remedial lessons and integration.

Supplementary lessons mean that the manufacturing industry will continue to narrow the gap in digital maturity with the Internet, finance and other industries. For small and medium-sized manufacturing enterprises, the focus is on the popularization of digital applications; for large and group-type manufacturing enterprises, it is necessary to switch from distributed applications to centralized computing centers and integrated applications.

One of the keys to integration is IT/OT integration, and the trend is that IT (information technology) continues to penetrate into OT (operational technology). This has also led to the rise of cloud-based industrial software, edge computing, design simulation, autonomous driving simulation, global digital twins, and AI visual quality inspection, which puts a strong demand on computing power.

Zhou Zhengang said that the current increase in computing power in the manufacturing industry is still quantitative, and it is expected to achieve qualitative growth in the future. The key here is that more and more manufacturing industries are digitized, not only serving internal employees and production lines, but also serving more customers like the Internet and telecommunications industries.

How to solve the “computing shortage” of large models

This year, “computing power” and “big model” are tied together to a certain extent. The industry often mentions the “computing shortage” caused by large models. There is such a metaphor in the “Report”: It takes 14.8 days to train the GPT-3 large model on 1,000 Nvidia V100 GPUs. Under the condition that the PUE of the data center is 1.1, the total energy consumption will reach 1287MWh. Calculated based on the per capita living electricity level in China in 2021, the power consumption of a single large model training is equivalent to the total living electricity consumption of a Chinese for 4 years.

In fact, although the popularity of large models only started at the beginning of this year, Zhou Zhengang told Data Intelligence Frontline that many users and suppliers have reported that their investment in artificial intelligence, especially AIGC, has increased significantly as early as last year.

This is also what he believes is the most noteworthy point in this year’s latest “Report”. The “Report” shows that the global generative AI computing market will grow from US$820 million in 2022 to US$10.99 billion in 2026, and its share of the overall AI computing market will increase from 4.2% to 31.7%.

Zhang Dong revealed that the average demand for artificial intelligence servers by enterprises this year is more than 5 times that of the previous ones, and some users even proposed more than 10 times the willingness to purchase.

Currently, AIGC technology is being explored and implemented in various fields such as the Internet, finance, education, medical care and manufacturing.

In the Internet industry, such as developing games, generating self-media content, e-commerce personalized recommendations, intelligent customer service and automated marketing; in the financial industry, AIGC can assist analysts to capture, analyze data, and generate reports; in the education industry, it can convert 2D teaching materials into 3D teaching models and synthesize virtual teachers; in the medical industry, it can provide potential candidate drugs, and conduct more accurate design and optimization; in the industrial field, AIGC provides auxiliary functions in CAD design…

Before the big model knocks on the door of the next industrial revolution, computing power is still a problem that needs to be solved by all parties. How to solve the problem of gradually unbalanced computing power supply and demand, ** Zhang Dong summarized four key points: diversification, systematization, infrastructure and ecologicalization**.

Zhang Dong believes that the paradigm of advanced computing should be an application-oriented model with system design as the core, and establish a computing development paradigm for the integration of multiple heterogeneous computing power, software and hardware collaborative design and optimization. At the same time, computing power can be provided to thousands of industries as conveniently as water and electricity, solving the problem of “affordable and well-used” for everyone**. When the computing power is applied to the industry, it needs to be supported by an ecology. In this process, the upstream and downstream of the industrial chain can carry out collaborative innovation.

Under such thinking, Inspur Information supports the digitalization of many industries and enterprises, as well as the construction of large-scale models and industry large-scale models, and accelerates the implementation of computing power in various industries through the creation of Yuannao ecology.

In response to the demand for computing power including large models, the “Report” puts forward targeted action suggestions. For example, countries should increase investment in computing power infrastructure at the national level and actively explore integrated computing power services; enterprises should actively promote the in-depth application of AI in business scenarios, practice the principle of digital priority, actively promote green computing, and consider increasing investment in liquid-cooled servers.