Ganming, the policy, from the concave templeQuotation Report Public Number QbitAI
Another year, the AICC conference, the China AI calculation report was released.
As the basis for the development of artificial intelligence, through the calculation of power, we can see the direction of the flow of AI tide.
Compared with previous years, the biggest change is the ranking of AI computing power in various cities in China.
Hangzhou surrendered to the top spot, and Beijing became the strongest city in the field of artificial intelligence in China (Beijing is not the first for many years).
Hefei fell out of the first echelon, Suzhou, Nanjing, and Xi'an were among the top ten for the first time, crowding out three cities.
Behind this, it is also the increasingly fierce competition in China's cities around artificial intelligence.
This report, from IDC and Inspur, is called"2019-2020 China Artificial Intelligence Computing Development Assessment Report".
The development of China's AI computing power was evaluated from the perspectives of geography, industry, chip, technology, etc., and the flow of computing power (chips, technology) and the flow direction of computing power (application) were sorted out.
Its role is not limited to glimpsing the direction of AI development, but also as a reference for cities and industries when choosing employment.
China's strongest AI city: North, Hangzhou, Shenzhen, Shang, Guang
According to the report, Top10, the strongest artificial intelligence city in China in 2019, is as follows:
(first echelon) Beijing, Hangzhou, Shenzhen, Shanghai, Guangzhou; (second echelon) Hefei, Suzhou, Chongqing, Nanjing, Xi'an.
Why are they?
Policy support, macroeconomics, large-scale technology companies, talent pools, and entrepreneurial ecology are key indicators.
Beijing (Baidu, Bytes, etc.), Hangzhou (Ali, NetEase, etc.) and Shenzhen (Huawei, Tencent, etc.) basically divided up China's large technology giants and emerging AI unicorn companies.
Moreover, North, Hangzhou, Shenzhen, Shangdong and Guangzhou are also the best economic developments in China's cities. A large number of top universities and talents and entrepreneurial ecology have also been perfected.
More importantly, local governments have also given corresponding support policies to promote the development of artificial intelligence technology.
Beijing is the city with the most licenses for auto-driving road test licenses in China; Hangzhou has given a large number of policies in attracting talents and companies to settle down; Shenzhen is a real money and silver subsidy for talents and enterprises.
The cities in the second echelon have also invested a lot of resources in these areas and have taken a comparative advantage:
They have been encouraged by the government, guided by policies, and established high-tech zones to provide a good development environment for the development of artificial intelligence, such as the industrial park in Suzhou and the western innovation port in Xi'an;
The cultivation of talents in universities, such as the University of Science and Technology of Hefei in Hefei, Nanjing University in Nanjing, and Xi'an Jiaotong University in Xi'an;
Advance with leading enterprises in the artificial intelligence industry chain, such as: Kefei Xunfei of Hefei;
The injection of a large amount of funds provides good support for the development of artificial intelligence. For example, Chongqing has signed several artificial intelligence projects, and it is expected to invest more than 5 trillion yuan in the next three years.
The degree of competition can also be seen from the changes in city rankings:
Compared with 2018, there has been a very big change: Beijing surpassed Hangzhou to become the first, Guangzhou entered the first echelon, and Suzhou, Nanjing and Xi'an were among the top ten.
In response to the exchange of these terms and the appearance of the new city, the report also gives an explanation:
First of all, Beijing, with the rapid development of Internet industry such as bytebeat, Baidu, and the largest number of artificial intelligence start-ups (nearly 500) and talent reserves in the country, surpassed Hangzhou to rank first;
Secondly, Guangzhou, with leading GDP growth, the government has increased a large amount of investment in artificial intelligence, coupled with the presence of a large number of industry-leading companies, making it among the top five;
Finally, Suzhou, Nanjing and Xi'an, with the construction of the government's science and technology industrial park, the gathering of talents and capital, and the leading enterprises, entered the top ten for the first time.
Of course, computing power and strength are just a holistic perspective. The power of computing must be transformed into productivity and driven by industry applications to reflect its value.
Calculate power into productivity: support for chips, servers, and AIaaS
Massive data is being generated all the time, and the speed and method of data storage are improving. In 2018, the total amount of data created globally was 32.6ZB, and the report predicts that by 2025, this number will increase to 175.2ZB.
The algorithm has evolved over the decades and has been rapidly developed and optimized since the advent of deep learning and accelerated computing. But the growing AI model poses a huge challenge to computing power.
Take the MegatronLM language model recently released by NVIDIA, which contains nearly 10 billion parameters.
The development of computing power is inseparable from the support of chips, servers, cloud computing and software.
At this stage, AI training chips occupy a larger market. The current AI chips are mainly divided into GPU, FPGA, and ASIC. The mainstream chips are still GPUs. Among them, NVIDIA and AMD are two prominent manufacturers. Domestic AI chips are mainly concentrated in the ASIC field.
IDC expects that the market for artificial intelligence chips will maintain rapid growth, with a compound growth rate of 53.0% over the next five years. As AI gradually becomes more and more, the market share of AI reasoning will exceed the training market by 2022. With the rapid growth of edge and end-side demand, the artificial intelligence chip market will usher in diversified development.
At present, the main domestic AI chip is ASIC, and suppliers include Cambrian, Horizon, Huawei, etc.
The AI market broke the development of the server and entered the fast lane. At this stage, the artificial intelligence server uses a heterogeneous architecture to accelerate the calculation. The combination of the chips is CPU+GPU, CPU+FPGA, CPU+ASIC and so on. Although AI servers can take many different forms, the CPU+GPU architecture is widely used on the market today.
Compared to traditional servers, the difference in AI servers is mainly reflected in the larger capacity of memory and the interconnect protocol optimized for AI.
IDC estimates that China's artificial intelligence infrastructure market will be approximately $1.9 billion in 2018, and will reach $8.3 billion in 2023, with a compound annual growth rate of 33.8% over the next five years.
Among them, the server market accounted for more than 85% of the entire hardware market. In 2018, GPU servers continued to grow at a high rate, with sales up 131.2% year-on-year, still the mainstream of AI servers.
According to the IDC report, China's artificial intelligence has gradually entered a stage of large-scale application, and the process of industrial AI is accelerating. From the perspective of suppliers, Chinese domestic suppliers account for most of the domestic market share. In 2018, the top three suppliers of China's GPU server market share are Inspur, Huawei and Shuguang.
The wave accounted for more than 50%. With an early entry into the field of artificial intelligence, Inspur collaborates with leading Internet companies through the JDM model. In the Chinese Internet industry, Inspur GPU server market share exceeds 60%, and continues to penetrate the traditional industry.
Software and cloud services
In recent years, companies have moved from traditionally procured hardware and software to deployment on public clouds. From the perspective of AI's capabilities, enterprises have gradually begun to purchase computing power such as cloud GPUs and FPGAs from public cloud service providers, as well as AI capable AIaaS services.
The convergence of AI clouds is an inevitable trend. AI will enable enterprises to easily acquire AI capabilities on the cloud in the form of public cloud services, thus effectively accessing and using AI technology.
IDC expects that AIAaS will be one of the most important drivers for the development of the cloud computing market. In the next five years, the annual compound growth rate of the AIIaS market will be 66.0%.
Software platform vendors in the AI ecosystem are an indispensable class of participants. The software framework market is gradually showing its duality and developing towards standardization.
TensorFlow continues to dominate with its performance and ecological advantages, and PyTorch has the potential for growth with its flexibility and enhanced performance. Baidu's deep learning open source platform PaddlePaddle is the representative of the domestic independent development software framework. The future competitive landscape is even more intense.
With the increase of computing power, more and more enterprises and open source organizations are involved in the research and development of artificial intelligence open source software, and new software platforms are entering the market.
The flow of computing power: Internet, government and finance are dominant
With the development of computing power, AI application scenarios are becoming more and more extensive. IDC has combed and explained the application scenarios of key industries. As shown in the chart below, the vertical axis is the size of the market and the potential for future development. The horizontal axis is the timeline for predicting the maturity of the solution and its widespread application.
AI has evolved from the laboratory stage to the industry, and technology companies have introduced artificial intelligence products such as smart speakers. In 2019, the market for smart speakers was growing rapidly.
According to IDC, shipments of smart speakers in the first quarter of 2019 reached 11.22 million units, an increase of 787.2% year-on-year. The household penetration rate has been comparable to that of PCs and smart TVs, and there is huge room for future development.
As the application scene matures, artificial intelligence is gradually infiltrating into all walks of life. At present, China's more mature application scenarios include biometrics, fraud analysis and investigation, intelligent customer service, and public safety.
IDC predicts that the artificial intelligence market will grow at a compound annual growth rate of 44.9% in the next five years, and the overall scale will reach US$17.5 billion. Internet, government and finance remain the dominant market.
For the short-term and predictable application scenarios of artificial intelligence in the future, IDC believes that there are manufacturing, retail, and telecommunications industries. Scenes that are expected to be widely used after 2025 include autopilot, intelligent diagnosis, adaptive learning, and more.
Based on continuous research and the latest research, IDC has obtained the application penetration and computing investment distribution of China's artificial intelligence industry:
In the first half of 2019, the TOP4 ranked according to the AI industry application penetration ranking was consistent with last year's rankings for the Internet, government, finance and manufacturing, while telecommunications exceeded the service ranks fifth.
In the first half of 2019, the TOP5 industry ranked according to the AI computing power ranking was also consistent with last year's ranking, followed by the Internet, government, finance, manufacturing and services.
Among them, the artificial intelligence technology enterprises represented by Keda Xunfei, Shangtang, Defiance, Yitu, Cambrian, and Fourth Paradigm in the service industry have accelerated their investment in artificial intelligence infrastructure and gradually formed in the artificial intelligence industry. Its unique core competitive advantage.
Insights in AI computing power: AIaaS and edge AI will usher in rapid development
Overall, this report has the following main points:
1. The competition among various cities in China has become increasingly fierce. Compared with 2018, Beijing surpassed Hangzhou to become the first. Guangzhou entered the first echelon, and Suzhou, Nanjing and Xi'an ranked among the top ten.
2. As AI gradually becomes more and more, the market share of AI reasoning will exceed the training market by 2022. With the rapid growth of edge and end demand, the artificial intelligence chip market will usher in diversified development.
3. AIAaS will become one of the most important driving forces for the development of the cloud computing market. In the next five years, the annual compound growth rate of the AIIaS market will be 66.0%.
4. The compound growth rate of artificial intelligence market will reach 44.9% in the next five years, and the overall scale will reach US$17.5 billion. Internet, government and finance are still the dominant market.
5, artificial intelligence in the short-term can be foreseeable application scenarios, I have manufacturing, retail, telecommunications industries. Scenes that are expected to be widely used after 2025 include autopilot, intelligent diagnosis, adaptive learning, and more.
How to deal with these trends? IDC gave recommendations in the report:
1. The computing power will become the core competitiveness.
Solution providers must be able to provide sufficient computing power and should be able to develop targeted software platforms for different applications.
2. AIAaS and Edge AI are ushing in rapid development.
Solution providers need to provide complete solutions including cloud training, cloud inference, and edge-side reasoning.
3. Ecological and standardization construction is of paramount importance.
Suppliers should be more active in promoting industry standardization development, building an industrial ecosystem, and providing an integrated platform for more upstream and downstream suppliers.
4. Focus on segmentation and provide personalized solutions.
In particular, start-ups should have clear focus areas and market segments, establish their own core competitiveness, and provide more personalized solutions for industry users.
Report original download address: