US, China, EU: Who Will Win the AI Race? An In-depth Analysis Across Six Dimensions

Artificial intelligence is a fundamental technology that can be used to help countries improve their competitiveness, productivity, protect national security, and help solve many social problems. Currently, countries around the world are competing to develop artificial intelligence technology.

 

This article, translated from a report released by the Center for Data Innovation (CDI), a well-known think tank in Washington, D.C., compares the AI technologies of China, the United States, and the European Union. Who Is Winning the AI Race: China, the EU or the United States?   

 

The report compares the development of artificial intelligence in China, the European Union, and the United States across six dimensions: talent, research, development level, application rate, data, and hardware. The report points out that although China has made great progress in AI technology, the United States still holds a clear lead, while the European Union lags behind in many indicators. However, with China's rapid development, this situation may change in the next few years. However, when considering per capita indicators, the United States' lead becomes even greater, while China ranks third, below the EU. The report also provides a series of policy recommendations to help each country or region improve its AI capabilities.

 

Author: Daniel Castro，Michael McLaughlin，Eline Chivot

Source: Zhidongxi, Chayan Guanshu

 

       In the last digital innovation revolution, the United States reaped enormous economic benefits, becoming home to some of the world's most successful technology companies and spawning tech giants such as Amazon, Apple, Facebook, Google, Intel, and Microsoft. At the same time, many countries and regions around the world, including the European Union, paid a corresponding economic price. Recognizing that missing the next wave of innovation, such as artificial intelligence, could pose similar problems, many nations are taking action to ensure their success in the next global digital economic transformation.

 

       China, the European Union, and the United States are now competing as rivals for global AI leadership. China has clearly indicated its ambition to achieve AI dominance in many ways. The EU's coordinated plan on artificial intelligence also shows its "hope that Europe will become the world's leading region in the development and deployment of top-notch, safe artificial intelligence." The outcome of this competition will affect the future economic output and competitiveness of China, the US, and the EU, as well as military advantages.

 

Key Findings

 

       Overall, the United States currently leads in artificial intelligence, China is rapidly catching up, and the European Union lags behind both. The United States ranks first in four of the six categories (talent, research, development level, and hardware), China has two (application rate and data), and the European Union has none. In the scoring given in this report, the United States scored a total of 44.2 points, followed by China with 32.3 points, and the European Union with 23.5 points.

 

Reasons for US Leadership

 

First, it has the most AI startups, and its AI startup ecosystem has received the most private equity and venture capital funding. Second, it leads the development of traditional semiconductors and the computer chips that power AI systems. Third, although it produces fewer AI academic papers than the EU or China, it has the highest quality papers. Finally, although the total number of AI talents in the United States is lower than that of the EU, its talent pool is more "small but refined".

 

China is rapidly closing the gap with the United States

 

It possesses more data than the EU and the US, which is very important because many of today's AI systems use large datasets to accurately train their models. In terms of venture capital and private equity financing, Chinese AI startups received more funding in 2017 than US startups. However, China lags significantly behind the United States and the EU in high-quality AI. As of 2017, several EU member states, including Italy, ranked among the top 10% internationally in international AI researchers. Despite this, China has made significant progress in most indicators, and it clearly surpasses the EU in funding and AI application rates.

 

While lagging behind in many indicators, the EU also possesses strong competitiveness. It has the most researchers and produces the most research results. However, there is a disconnect between the number of EU AI talents and its commercial AI application rate and funding. For example, US and Chinese AI startups received more venture capital and private equity funds in 2017 alone than the EU received in the three years from 2016 to 2018.

 

▲Absolute Indicator Ranking

 

To understand each region's AI advantages relative to its population size, the report also calculated scores based on the average workforce. The United States leads (58.2 points), the EU ranks second (24.3 points), and China ranks third (17.5 points).

 

As the report shows, China, the EU, and the US all have room for improvement. For example, China should expand its capacity to teach AI-related disciplines in universities, encourage research quality rather than quantity, and foster a stronger open data culture. Meanwhile, the EU should focus on policies that incentivize talent to remain in the EU, help translate research results into commercial applications, encourage the development of larger companies that can better compete in the global market, and reform regulations to better utilize data for the benefit of AI research. For the United States, the focus should be on policies to increase the domestic talent base, encourage immigration of foreign talent, and increase R&D incentives.

 

Is the AI Competition a Zero-Sum Game?

 

Many believe that nations do not compete in innovation. In this view, there are only winners, not losers. But in reality, there are both winners and losers in the global AI race. Countries that fail to develop successful AI products or services will risk losing global market share. As Andrew Moore, former head of Carnegie Mellon University's computer science department and current head of Google Cloud AI, said, the AI race will determine "who will be the Google, Amazon, and Apple of 2030." Nations that underinvest in AI R&D, especially in military applications, will put their national security at risk. Therefore, countries that lag behind in the AI race may suffer economic losses and affect national security, thereby weakening their geopolitical influence.

 

However, in some areas, the AI race is not a zero-sum game. Advances in AI science, particularly in universities, can be disseminated worldwide. Much of AI R&D, especially research focused on health, the environment, and education, can benefit all countries. For example, the development of AI systems that can identify diseases or generate new therapies faster and more accurately than doctors could benefit the entire world. This year, Chinese and American researchers collaborated to develop an AI system that can accurately diagnose common childhood diseases. This system achieved over 90% accuracy in diagnosing asthma and 87% accuracy in diagnosing gastrointestinal diseases. The researchers also trained the system using electronic medical records from 600,000 Chinese patients. In addition, because much of AI research is open, researchers worldwide can quickly learn from the achievements of others abroad.

 

Talent Situation

 

Talent is The Key to AI Research

 

As David Wipf, chief researcher at Microsoft Research Beijing, said, "The future of AI will be a battle for data and talent." Talent not only determines a company's ability to deploy and adopt AI, but also its cost. Given the increasing demand for AI talent in numerous industries such as transportation, finance, and manufacturing, the current talent shortage will likely worsen in the future.

 

The Chinese, EU, and US governments have announced or implemented numerous talent initiatives.

 

For example, in 2018, China's Ministry of Education announced a plan to promote AI education. In response, several top Chinese universities opened new AI departments and programs. The UK government announced that it would fund up to 1,000 students with up to £115 million ($129 million) to help them obtain AI doctoral degrees. President Trump issued an executive order that would expand scholarships, training programs, and funding for university professors conducting AI research.

 

The report analyzed the number of AI researchers, the number of top AI researchers, and the degree level of AI researchers to assess the talent and talent cultivation situation in China, the EU, and the United States. The latest data shows that the United States leads in AI talent (6.7 points), followed by the EU (6.2) and China (2.1). Considering per capita indicators, the United States (8.4 points) also leads the EU (5.8 points) and China (0.9 points).

 

Total Number of AI Researchers: This section defines AI researchers as individuals who published journal articles or obtained AI-related patents between 2007 and 2017. The EU is estimated to have 43,064 researchers, ahead of the US (28,536) and China (18,232). Indeed, the total number of AI researchers from Germany (9,441), the UK (7,998), France (6,395), Spain (4,942), and Italy (4,740) exceeded that of US researchers. Considering per capita, the US (173 researchers per million workers) leads the EU (173) and China (23).

 

▲2017 Number of AI Researchers

 

Top Number of AI Researchers (Based on H-index): Quality may be more important than quantity.

 

One of the metrics for evaluating the quality of researchers is the H-index, which measures a researcher's productivity and impact. The report counted the number of top 10% international AI researchers based on the h-index ranking. By 2017, the EU had approximately 5,787 high H-index researchers, ahead of the US (5,158) and China (977). The UK (1,177), Germany (1,119), France (1,056), Italy (987), and Spain (772) totaled 5,111 high H-index personnel. Although data from the other 23 EU countries are unavailable, it's evident that the remaining countries possess sufficient top AI talent to nearly offset the difference of less than 100 between the US and EU. When considering the workforce size, the US (31 researchers per million workers) leads the EU (23) and China (1).

 

▲2017 Number of High H-index AI Researchers

 

Top Number of AI Researchers (Based on Academic Conferences): A second measure of quality is the number of authors who published papers at top AI academic conferences. AI startup Element AI tracked 21 AI conferences in 2018. The US (10,295 researchers) leads the EU (4,840) and China (2,525). Considering per capita indicators, the US (62 researchers per million workers) also leads the EU (19) and China (3).

 

▲2017 Number of AI Researchers at Top Academic Conferences

 

Education Top AI Researchers (Academic Conferences): Cultivating AI talent is also crucial.

 

The report counted where researchers who published at 21 major academic conferences in 2018 obtained their doctoral degrees. The United States (44%) has more than the combined total of the EU (estimated 21%) and China (11%). This largely provides the US with an advantage in AI talent, given that 79% of students received doctoral degrees. A significant portion of those who obtained mathematics or computer science degrees in the US intend to remain in the US.

 

▲Location where AI Researchers Obtained Doctoral Degrees

 

Data shows that although the EU has a large number of AI talents, its top businesses have fewer talents than US companies. This, coupled with a lack of venture capital and private equity financing, may hinder its ability to develop top AI companies. For example, based on AI papers and patent records, half of the 20 companies with the most AI talent in 2017 were located in the US. These ten US companies together employed 1,623 AI workers. In contrast, the EU only had six such companies, with a total of 522 AI workers. The only Chinese company to make the top 20 was Huawei, with 73 employees. Similarly, based on the H-index, in 2017, the EU had 85 among the top 20 companies with the most AI researchers, while the US had 232. China had 7 high H-index researchers.

 

Another concern for both the EU and China is that the US still attracts more AI talent from other countries compared to Europe and China.

 

For example, in Between 1998 and 2017, 1,283 foreign AI academic researchers came to the US from abroad. Europe and China attracted 834 and 58 such researchers, respectively. Furthermore, data collected by Elsevier shows that between 1998 and 2017, the US attracted more foreign academic researchers (318 AI researchers) than left the US (166).

 

Compared to China, the EU has many advantages in AI talent. For example, in 2017, the UK (1,177), Germany (1,119), France (1,056), and Italy (987) each had more high H-index personnel than China (977). However, China's lack of top AI talent may be due to its relatively recent interest in AI, with only 25% of Chinese AI researchers having more than ten years of experience, compared to 50% in the US. Furthermore, there are many ways for China to reduce the lack of talent, and the importance of talent shortage may decrease.

 

China is investing in AI education

 

In 2017, China's State Council issued a plan encouraging universities to establish AI disciplines. In 2018, the Ministry of Education launched several initiatives aimed at promoting education, including the development of 50 AI research centers, world-class online courses, and the training of more than 500 faculty members and 5,000 students over five years. Since 2016, three top universities in China (Tsinghua University, University of Science and Technology of China, and Shanghai Jiao Tong University) have significantly increased the number of students enrolled in AI and machine learning courses. For example, between 2016 and 2018, the number of AI and machine learning places at the University of Science and Technology of China increased from 1,745 to 3,286. Secondly, Chinese researchers can quickly replicate advanced algorithms developed in other countries because AI researchers often detail the architecture of their AI models, and how to implement and train it, on publicly available preprint websites. Moreover, Chinese researchers translate English AI publications significantly more frequently and faster than Western countries translate Chinese papers, resulting in information asymmetry. Third, Kai-Fu Lee believes that China's lack of high-end talent is not the main obstacle to its leading the development of artificial intelligence, noting that "the current era is doing very well in the commercialization of AI. He believes that major breakthroughs in fields such as deep learning occur once every few decades, and AI has entered a new era that suits China's national conditions: possessing a large number of highly skilled, although not necessarily the best, AI researchers and practitioners." Data will be the decisive factor in determining the functionality of AI systems.

 

 

Science Research

 

All countries need continuous innovation to maintain their position

 

Over the past decade, algorithmic innovation and greater computing power have improved the capabilities of AI systems and significantly reduced the time required to train them. However, artificial intelligence is far from a mature technology. More research and further advancements are needed.

 

This section will analyze China, the European Union, and the United States in terms of the quantity and quality of AI academic papers and commercial R&D funding.

 

Number of AI papers: In 2017, China published 15,199 AI papers, the EU 14,776, and the US 10,287. Historically, however, the EU produced the most AI papers. For example, from 1998 to 2017, EU researchers wrote nearly 164,000 AI papers, while Chinese and American authors produced 135,000 and 107,000 respectively. However, calculating per capita metrics, the US published 63 AI papers per million workers in 2017, leading the EU (59) and China (19).

 

▲Number of 2017 AI Papers

 

Paper Quality

 

Although, as the Allen Institute for Artificial Intelligence wrote, "All papers are created equal." But in reality, the highest quality research is conducted in the United States. In 2016, the US had a field-weighted citation impact (FWCI) benchmark of 1.83, meaning that researchers cited papers by US authors 83% more often than the global average. In contrast, the EU and China had FWCIs of 1.20 and 0.94 respectively, indicating that Chinese authors were cited less frequently than the average global AI expert. However, China's FWCI has increased year on year since 2012.

 

▲Field-Weighted Citation Impact

 

Top R&D Spending Tech Companies

 

Another way to gauge a region's research capabilities is to count its spending on R&D. It is difficult to know how many companies specifically spend on AI research, but the overall R&D spending of tech companies (many of which are developing AI services) can substitute for AI R&D spending. This indicator examined the top 100 tech companies by R&D spending in 2018. The US (62 companies) leads the EU (13) and China (12). Per 10 million workers, the US also leads the EU and China.

 

▲Top 100 Tech Companies by R&D Spending in 2018

 

Global Ranking of Total R&D Spending by Tech Companies Top 2500

 

In 2018, there were 268 technology companies among the world's top 2500 companies in terms of R&D spending. The report counted the R&D spending of these 268 companies by region. The US (€69 billion, $77 billion) leads China (€10 billion, $12 billion) and the EU (€9 billion, $11 billion). Per capita, the US ($470 per worker) leads the EU ($42) and China ($15).

 

▲Global Top 2500 Ranking of R&D Spending by Tech Companies in 2018

 

Analysis of the data shows that the US is leading in AI research, both because of its massive spending on R&D and the commitment of its elite research institutions. Nonetheless, China's catching up to the US and the EU is not only because it is conducting more research but also because it has started to conduct high-quality research.

 

The US's lead in research is partly due to its elite organizations

 

For example, the top five tech companies for R&D are all US companies. Another way of evaluating the quality of research in a country is to count the organizations that publish the most influential AI papers. The US also leads in this indicator. Carnegie Mellon University, MIT, Microsoft, IBM, and Stanford University were the US organizations that published the most AI papers between 2013 and 2017. The combined FWCI of these five organizations is 4.0, significantly higher than the combined FWCI of the top five organizations in the EU (1.9) and China (1.4).

 

Although the average research quality of top EU organizations is higher than that of China, the EU's paper output and quality have been relatively stagnant

 

Since 1998, the EU's FWCI has only grown by 11%, compared to 24% for the US and 154% for China. If China maintains this growth rate, China's FWCI in 2018 might have already surpassed the EU (data is only available up to 2016). Furthermore, the five countries driving AI research primarily in the EU - the UK, Germany, France, Spain, and Italy - have actually seen a contraction in their annual AI publication output since 2014.

 

The EU's stagnation is coupled with China's rise. While the US and EU FWCIs in 2009 were almost identical to those in 2016 (1.82 and 1.83 for the US, and 1.21 and 1.20 for the EU), China's FWCI increased from 0.59 to 0.94 during the same period. China's FWCI is rapidly approaching or exceeding the global average of 1.00.

 

China also doesn't need to match the US in More research is needed on FWCI because it has generated a large amount of research. For example, a recent analysis of AI papers by the Allen Institute for Artificial Intelligence (AI2) found that the US share of the top 10% most cited AI papers fell from 47% in 1982 to 29% in 2018. From approximately 0% in 1982 to 26.5%. AI2's research indicates that China's paper output will surpass that of the US by 2020 and 2025, ranking first and second in the top 10% and 1% of all AI research papers, respectively. Although China's citation count may be inflated due to self-citation, the quality of China's research has absolutely improved relative to the US and the EU.

 

 

发 展

 

healthy AI systems

 

To fully reap the benefits of AI, countries must have healthy AI ecosystems to guide the development of innovative AI technologies and companies. For example, countries must have sufficient venture capital and private equity funding to meet the funding needs of inventors to develop and sell their products or services. This section analyzes AI venture capital and private equity funding, the number of AI companies, mergers and acquisitions, and patent data. The latest data shows that the US (14.9 points) leads the EU (5.3 points) and China (4.8 points). In per capita terms, the US (19 points) leads the EU (4.5) and China (1.4).

 

Total Venture Capital and Private Equity Investment ( 2017-2018): Tracking private funding is a good way to measure a country's ability to develop AI companies. Statistics on venture capital and private equity financing for AI companies between 2017 and 2018. The US (estimated at $16.9 billion) leads, followed by China (estimated at $13.5 billion) and the EU (estimated at $2.8 billion). Per capita, the US still leads.

 

▲Total Venture Capital and Private Equity Investment (2017-2018)

 

Number of Venture Capital and Private Equity Financing Transactions ( 2017-2018)

 

Artificial intelligence venture capital and private equity financing can be concentrated in a few large transactions, but the total number of venture capital and private equity financing transactions can also be tracked. In 2017-2018, US AI companies received the most investment (1,270 transactions), exceeding the EU (660) and China (390). In per capita terms, the US (8 transactions) leads the EU (3) and China (0.5).

 

▲Number of Venture Capital and Private Equity Financing Transactions (2017-2018)

 

Number of AI Startups (2017)

 

Similar to other technology-based startups, AI startups can be a major driver of a country's economic growth and competitiveness. The US had 1,393 AI startups in 2017, exceeding the EU (726 startups) and China (383 startups). In per capita terms, the US ranks first (8), followed by Europe (3) and China (0.5).

 

▲Number of AI Startups (2017)

 

High Citation Number of AI Patents (1960-2018)

 

Patents are a means of innovation. However, it is difficult to measure innovation using patents, partly because national standards for granting patents differ. This report focuses primarily on Patent Cooperation Treaty ( PCT) patent applications and highly cited patents. From 1960 to 2018, US patent applicants filed 28,031 highly cited patents with the USPTO, significantly exceeding the number of highly cited patent families filed with offices in the EU (2,985) and China (691). While this indicator shows where applicants filed patents, not their location, most applicants typically file patents first in their country of residence.

 

▲Number of Highly Cited AI Patents (1960-2018)

 

The US leads in all AI development indicators, suggesting that it is better positioned to continue developing leading global AI companies compared to China and the EU. Patent and acquisition data also show that the US has already taken the lead in developing world-class AI companies. However, partly due to its strong venture capital and private equity ecosystem, China is catching up with the EU and the US. Conversely, while currently ranking slightly higher than China in AI development, the EU may lack the funding to challenge the US's position.

 

US companies have excelled in patents and dominance in AI acquisitions. For example, in 8 of 15 machine learning subcategories, Microsoft and IBM filed more patents than any other institution, including supervised learning and reinforcement learning. The Chinese Academy of Sciences filed the most patents in deep learning, while Siemens (Germany) filed the most patents in neural networks. Nevertheless, US companies lead patent applications in 12 of 20 areas, including agriculture (John Deere), security (IBM), and personal devices, computing, and human-computer interaction (Microsoft). In addition, between 2012 and 2016, IBM led the world in AI patent applications (3,677), with Google's parent company Alphabet (2,185) and Microsoft (1,952) also ranking among the top five.

 

While the US is leading in AI development, how long it can maintain that lead remains unknown.

 

including Various analyses of funding data for AI startups, including AI, have found that for at least one year, Chinese AI startups received more funding than US startups. For example, in 2017, Chinese AI startups received approximately $8.1 billion in investment, while US startups received approximately $6.2 billion. In addition, research by Chinese tech company Tencent found that the average investment time for US AI startups was 14.8 months, while in China it was 9.7 months.

 

In terms of the number of investments in AI startups, China is also beginning to close the large gap with the US, reducing the gap from 476 in 2016 to 371 in 2018. The reason for the narrowing gap is the significant increase in the number of investments in Chinese startups, while the number of transactions involving US AI startups has remained relatively stagnant. US AI startups did receive record investment in 2018, but only received $10.7 billion, while Chinese AI startups also received approximately $5.4 billion.

 

在 Between 2016 and 2018, private equity and venture capital funding for EU AI startups nearly tripled, yet the EU still lags behind the US and China. For example, the US received more funding in any single year between 2016 and 2018 than Europe received in total over the three years. Similarly, in 2017 and 2018, Chinese AI startups received several billion dollars more in private equity and venture capital funding than the EU. Unless EU startups begin to receive significantly more funding, the EU is likely to fall further behind the US and China.

 

Should use

 

Technological innovation is key to raising living standards, and artificial intelligence is likely to be a major driver of technological innovation in the emerging wave of innovation. By 2030, AI is estimated to generate $13 trillion in gross domestic product (GDP) growth. Companies are increasingly needing to apply AI to remain competitive. In addition to economic benefits, AI can also bring significant societal benefits, such as reducing car accidents and injuries and enabling better disease treatment.

 

The Chinese, EU, and US governments have publicly acknowledged the importance of applying AI.

 

For example, in 2017, China’s Ministry of Industry and Information Technology released the Three-Year Action Plan (2018-2020) to Promote the Development of a New Generation of Artificial Intelligence Industry, calling for the integration of AI into manufacturing. Furthermore, the EU’s coordinated plan on AI calls for the creation of a ‘European common data space’ in sectors such as manufacturing and energy to support the development and adoption of AI. In 2019, US President Trump issued an executive order calling for the development of technical standards to support the adoption of AI.

 

To assess the AI adoption rate in China, the EU, and the US, the report analyzed surveys on AI adoption rates. China leads with a score of 7.7, ahead of the EU (1.3) and the US (1). In terms of the percentage of companies that have applied or piloted AI, China is in the lead (4.7 points), followed by the US (2.9 points) and the EU (2.5 points).

 

Application Percentage of companies applying

 

Measuring A first approach to measuring AI adoption rates is to track the percentage of companies that have successfully applied AI to their business processes. In 2018, China (32% of total companies) led this indicator, followed by the US (22%) and the EU (estimated at 18%).

 

▲Percentage of companies adopting AI

 

Trial Percentage of companies piloting

 

Measuring A second approach to AI is to track the percentage of companies that are piloting AI. This indicator tracks companies that were piloting AI initiatives as of September and October 2018. On this indicator, China also leads (53% of total companies), followed by the US (29%) and the EU (26%).

 

▲Percentage of companies piloting AI

 

Although different surveys find different adoption rates, they also show similar trends: China’s adoption of AI is faster than that of the US and the EU. China’s lead in AI adoption may be partly due to higher awareness of the value of AI among its people and businesses.

 

Unlike the US and the EU, China’s adoption rate is relatively uniform across industries. For example, the difference in the percentage of US companies active in AI (meaning they are adopting or piloting AI) across different industries is as high as 32 percentage points. However, the difference in the share of active AI companies between the highest and lowest adopting industries in China is only 6 percentage points.

 

▲Percentage of companies adopting or piloting AI in China and the US by industry in 2018

 

There are several possible explanations for this phenomenon. The first is that the importance of AI has permeated Chinese culture. Following the release of the “New Generation Artificial Intelligence Development Plan” by the State Council in 2017, the government began aggressively investing in and applying AI to AI startups. By applying AI, the Chinese government not only provides funding to AI companies but also creates role models to demonstrate the benefits of AI to encourage private companies to adopt AI. Furthermore, a higher percentage of Chinese people (76%) believe that AI has a positive impact on the overall economy, exceeding the US (58%), France (52%), Germany (57%), Spain (55%), and the UK (51%). A second possible explanation is that, compared to some Western countries, China’s technological utilitarian culture is willing to adopt AI as long as it can provide broader social welfare, even if some people believe there are ethical concerns about AI.

 

Compared to US companies, Chinese companies do a better job of communicating the importance of AI to their employees.

 

For example, 43% of Americans say their employers consider the development of AI and the digital transformation of the organization to be strategically important, while the figure in China is 85%. The same survey found that 54% of Americans say their workplace has no plans to deploy AI tools, which is not surprising, while the figure in China is only 22%.

 

While US companies may not be properly communicating the importance of AI to their employees, many EU individuals are outright skeptical of AI. Therefore, while the EU lags only slightly behind the US in adoption rates, it lags significantly behind China. Similarly, EU individuals generally have more negative sentiments toward AI in the workplace than US workers, and significantly more negative sentiments than Chinese workers. For example, when considering the impact of AI on their work, a higher percentage of individuals in the UK (55%), Germany (61%), France (65%), and Spain (53%) cite at least one negative feeling about their work, compared to 51% and 24% in the US and China, respectively. EU individuals may lack enthusiasm for AI because they have less positive experience with AI; 77% and 91% of individuals in the US and China, respectively, report that AI tools have a positive impact on their effectiveness. Lower percentages of French (62%), German (65%), Spanish (72%), and British (74%) individuals feel similarly.

 

Number According to

 

Artificial intelligence systems typically rely on large amounts of data for training. Large datasets help AI systems develop highly accurate models. Furthermore, machine learning techniques enable AI systems to identify subtle patterns in large datasets that are difficult or impossible for humans to perceive. This is one reason why many AI systems perform certain tasks better than human experts, such as identifying signs of lung cancer in computed tomography scans.

 

Decision-makers in China, the EU, and the US have recognized the importance of data.

 

In 2015, to support the use of big data, China listed open data as one of ten national projects. The EU's coordinated plan on AI states: “AI requires the development of large amounts of data…the larger the dataset, the better AI can learn and discover even subtle relationships in the data.” In the US, President Trump’s American AI Initiative instructed the government to “enhance access to high-quality and fully traceable federal data,” and directed the US Office of Management and Budget to identify and address data quality limitations.

 

There is no direct metric to measure the relative quantity and value of AI-usable data in a specific location. However, as individuals engage in various online and offline activities (e.g., using search engines, posting on social media, and making purchases), they generate large amounts of data. The data generated from these activities can be of immense value for machine learning models. Therefore, one way to estimate the potential value of data in a country or region is to consider the percentage of the population participating in digital activities.

 

Mobile Payments ( 2018)

 

Consumers also generate data that technology companies can analyze each time they use a mobile device to purchase products. The report defines “mobile payments” as using a mobile device to scan and make transactions at the point of sale, excluding online purchases. It is estimated that over 525 million Chinese people made mobile payments in 2018, compared to 55 million in the US and 44 million in the EU. In 2018, an estimated 45% of the Chinese population used mobile payments, compared to 20% in the US, 13% in the UK, and 8% in Germany.

 

▲Number of people using mobile payments in 2018

 

Internet of Things (IoT) Data ( 2018)

 

IoT devices can generate large amounts of data that organizations can use to train machine learning systems. This metric tracks the estimated amount of IoT data generated in each region in 2018, in terabytes (TB). China (152 million TB) leads the US (69 million TB) and the EU (53 million TB). Per 100 workers, the US (42 TB) leads the EU (21 TB) and China (19 TB).

 

▲Newly generated IoT data volume

 

Productivity Data ( 2018)

 

Organizations constantly generate data that can serve as input to train their AI systems. For example, an airline can analyze its customer, agent, aircraft, and route map data to better control its flight costs. This metric tracks the estimated volume of productivity data, which is a combination of big data and metadata. The US (966 million TB) leads China (684 million TB) and the EU (583 million TB). Per 100 workers, the US (586 TB) leads the EU (234 TB) and China (87 TB).

 

▲Productivity Data

 

Electronic Health Records: Researchers have used electronic health records to develop AI systems that can perform a variety of functions, from predicting whether a patient is likely to be hospitalized to helping track the spread of diseases. Comprehensive data on the adoption of electronic health records in China, all EU member states, and the US are currently unavailable. However, a combination of quantitative and qualitative information suggests that the US has the highest number of electronic health records, followed by the EU and China. Therefore, the US also leads in per capita access, followed by the EU and China.

 

Adoption rates of electronic health record systems are relatively high across all regions surveyed, but the availability of access to electronic health records across borders and among providers is not high. For example, a 2015 survey found that 84% of primary care physicians in the US used electronic health record systems, compared to 99% of Swedish physicians, 98% of Dutch physicians, 98% of British physicians, 84% of German physicians, and 84% of French physicians. In China, a 2012 survey found that 48% of hospitals had basic electronic health record systems. The number of Chinese hospitals using electronic health records has likely grown to over 90% since 2012. In 2017, over 96% of US hospitals used certified electronic health record systems.

 

However, in 2015, only 30% of US hospitals could find, send, and receive electronic health records from other healthcare providers. Evidence suggests that interoperability is even lower in China and the EU. In China, hospitals often use electronic health record systems that are not interoperable, forcing patients to carry printed health records when seeing doctors at different hospitals. In the EU, the ability to access and share medical data across borders varies significantly, limiting the ability to train AI systems on cross-border data.

 

High-Resolution Map Data: High-resolution map data is important for the development of numerous AI systems, including self-driving cars. The US leads this metric, followed by the EU and China. As of April 2019, 45% of US states currently have data at 1-meter resolution or higher. In contrast, only six EU member states (representing about 15% of the EU’s geographical area) provide complete high-resolution 3D elevation data to the public. The rest either provide partial or low-resolution coverage or do not make the data publicly available. In China, the People’s Republic of China Mapping Law requires all entities conducting mapping to have a license, and as of January 2018, only 14 Chinese companies had obtained a license, with Chinese companies viewing the license as a “golden key.”

 

China leads in both the amount of data collected and the amount of data accessible to large internet companies (which are also likely the companies best positioned to leverage AI). This fact, coupled with several data deficiencies that could be mitigated by changes in Chinese policy, means that China may have a greater advantage in the future.

 

Chinese internet companies may have access to more and broader data

 

Chinese large internet companies may have a data advantage over their Western counterparts for at least two reasons.

 

First, services in the West are relatively fragmented across companies.

 

For example, Amazon users can buy groceries but cannot book hotels. On the other hand, Chinese tech companies have created all-in-one super apps. For example, WeChat, an app owned by the Chinese tech company Tencent, allows users to “hail a ride, order food, book hotels, manage phone bills, and buy flights to the US,” whereas in the US, these services (and the data) are divided among companies such as Uber, Postmates, Expedia, Verizon, and Venmo.

 

Second, Chinese tech companies have integrated themselves into traditional offline activities.

 

For example, Didi ( the Chinese version of Uber) has bought gas stations and car repair shops. In addition, Meituan Dianping, which originated similarly to Yelp, not only provides a platform for users to compare businesses but also handles food delivery. Therefore, Chinese internet companies have more opportunities than US companies to collect data of greater variety and depth. However, it should be noted that the broader global reach of some US tech giants provides them with their own data advantages. For example, Facebook has over 2 billion users, while WeChat has only 1.1 billion. If Chinese companies achieve more international success, such as with the social media video app TikTok, the US advantage would weaken.

 

China has also not fully leveraged the data it generates. For example, for decades, US companies have been collecting structured data in sectors such as insurance and finance, such as loan repayment rates. However, Chinese companies have been slower to adopt enterprise data warehousing, making it difficult to extract insights and the value of this data. China also lags behind its Western counterparts in establishing standards to help organizations share data across platforms. Government agencies have neglected basic standards for data collection, resulting in large quantities of data that are unreadable by computers, which reduces the quality and usability of the data for analysis. Although public data was designated as a national project in 2015, China still lags behind its peers in making government data available to the public. Finally, while other countries are benefiting from increased global cross-border data sharing, China’s internet ecosystem remains closed, limiting the amount of data it shares and receives from abroad. This ‘closedness’ reduces the diversity of data collected by Chinese companies.

 

Hard ware

 

AI systems rely on semiconductor devices that can perform a massive number of operations per second

 

Specifically, Graphics Processing Units (GPUs), which are circuits that perform mathematical operations in parallel, catalyzed the recent development of AI. In addition, technologies such as supercomputers combine processing units such as GPUs and Central Processing Units that can scale the capabilities of AI systems through massive computational power. For instance, researchers are combining supercomputers and machine learning techniques to simulate climate change as well as the merger of black holes. The aforementioned hardware is crucial to enhancing a nation’s AI competitiveness for several reasons. First, countries with weak semiconductor industries may be vulnerable to the actions of other nations. For example, in 2018, the US banned US companies from providing parts and software to ZTE, a large Chinese telecommunications equipment manufacturer. Because ZTE relied on semiconductor devices from US companies, the company nearly went bankrupt. Although the US eventually lifted the ban, the situation highlighted China’s dependence on Western technology. More recently, the US has blocked US companies from selling chips to five specific supercomputing entities, and the US Department of Commerce has blacklisted Huawei, preventing companies from selling US technology without a license. Second, many experts believe that AI chips specifically designed for AI applications, such as self-driving cars or facial recognition, will outperform established technologies such as GPUs. As a result, non-semiconductor companies such as Apple, Alphabet, and Amazon are designing their own AI chips to meet their specific needs, which can improve the performance of their AI systems, thereby giving them a competitive advantage. Third, high-performance computing has driven breakthrough discoveries in multiple fields, and the use of top-performing supercomputers provides nations with an advantage to develop cutting-edge weapon systems and applications faster than other countries.

 

Semiconductor R&D spending: Not only are semiconductor sales important, but the R&D spending of semiconductor companies is also important, which is often the main factor influencing who develops the best chips. This metric examines the number of top 10 semiconductor companies in R&D spending in 2017. The United States (5 companies) leads the EU (0) and China (0), with the 5 US companies having a combined spending of $24 billion in R&D.

 

Number of companies designing AI chips (2019) Because some companies have found that developing custom AI chips can improve the performance of their AI systems, it is also important to track the number of companies designing AI chips. We analyzed multiple data sources, including CrunchBase, to track the number of companies developing chips for AI use cases. The United States (55 companies) leads China (26 companies) and the EU (12 companies). Per 10 million workers, the United States (3) also leads the EU (0.5) and China (0.3).

 

Number of supercomputers (2019)

 

This metric counts the number of supercomputers ranked in the top 500 in performance. In the Top 500 Supercomputers, China has more supercomputers (219) than the US (116) and the EU (92) combined. Per 10 million workers, the US (7 supercomputers) leads the EU (4) and China (3). Another way to assess a country/region is to measure the combined system performance of the top 500 supercomputers. In the global Top 500 Supercomputers, the US has the highest share of combined system performance (38%), leading China (30%) and the EU (17%). Per 10,000 workers, the US (36 TFLOPs/s) also leads the EU (10) and China (6).

 

Data analysis shows that the US still leads in hardware but China is challenging the US in supercomputers, China is rising in AI semiconductors, while the EU lags behind other countries. The US’s position in developing the world’s fastest supercomputers demonstrates its strength, but China’s capabilities are constantly improving. For example, 6 out of the 10 fastest supercomputers are located in the US. Furthermore, the two fastest supercomputers in the world, Summit and Sierra, are located at US Department of Energy (DOE) sites. In addition, the US company Intel developed 96% of the processors in the Top 500 supercomputers. Of the 133 supercomputers that use accelerators or co-processors to enhance computer performance, 98% are from US companies Nvidia or Intel.

 

However, in some respects, China has overtaken the US as the global leader in supercomputers.

 

In June 2010, 282 of the world’s 500 top-performing supercomputers were in the US. However, in 2018, the US only had 109 in the Top 500, the lowest number ever. In addition, both the US and China are developing exascale computers, which can perform 500 quintillion calculations per second. China has begun to show that it has the potential to reduce the gap with the US in the semiconductor sector, at least in AI chips. In the past two years, several Chinese

 

AI chip startups have received at least $100 million in funding. Some experts believe that China has a greater competitive advantage in the AI chip market compared to the overall semiconductor market. For example, Horizon Robotics, which develops AI chips for robots, received a $600 million investment in its Series B funding round in 2018, led by SK Hynix, a leading South Korean semiconductor company. Similarly, Bitmain, which initially developed chips for bitcoin mining, has developed an AI chip and received nearly $765 million in funding between 2017 and 2018. Finally, Cambricon Technologies, which developed the world’s first commercial deep learning processor for mobile phones in 2016, received a $1 million grant from the state-backed National Development and Investment Company. 500位的超级计算机的数量。在超级计算机500强中，中国拥有的超级计算机数量（219）超过美国（116）和欧盟（92）的总和。每1000万工人中，美国（7个超级计算机）领先于欧盟（4个）和中国（3个）。

 

▲超级计算机数量（2019年）

 

超级计算机（综合系统性能， China has begun to show that it has the potential to reduce the gap with the US in the semiconductor sector, at least in AI chips. In the past two years, several Chinese

 

评估国家/地区的另一种方法是衡量排名前500位的超级计算机的综合系统性能。在全球前500强超级计算机中，美国的综合系统性能所占比例最高（38 ％），领先于中国（30％）和欧盟（17％）。每10,000名工人中，美国（36 TFLOPs / s）也领先于欧盟（10）和中国（6）。

 

▲2009-2019年排名前500名的超级计算机的综合性能

 

数据分析显示，美国在硬件方面仍然领先，但中国在超级计算机方面挑战美国，中国在AI半导体方面正在崛起，而欧盟则落后于其他国家。

 

美国在发展世界上最快的超级计算机方面的地位既显示了其实力，但中国的能力在不断提高。例如，最快的10台超级计算机中有6台位于美国。此外，世界上最快的两个超级计算机，Summit和Sierra，位于美国能源部（DOE）的站点。此外，美国公司英特尔开发了500强超级计算机中96％的处理器。在使用加速器或协处理器来增强计算机性能的133台超级计算机中，有98％来自美国公司Nvidia或英特尔。

 

但是，从某些方面来说，中国已经赶超美国成为超级计算机的全球领导者。2010年6月，全球500台性能最佳的超级计算机中有282台在美国。然而，在2018年，美国只有有109台世界500强，为历史最低水平。此外，美国和中国都在开发百亿亿次计算机，它们每秒可以执行五百亿次计算。

 

中国已经开始显示出它有可能减少与美国在半导体领域的差距，至少在人工智能芯片方面。在过去两年中，数家中国AI芯片初创企业已获得至少1亿美元的资金。一些专家认为，与在整个半导体市场相比，中国在人工智能芯片市场上的竞争优势更大。例如，为机器人开发人工智能芯片的Horizon Robotics在2018年的B轮融资中获得了6亿美元的投资，该轮融资由世界领先的韩国半导体公司SK Hynix牵头。同样，最初开发用于比特币采矿的芯片的比特大陆，已经开发了一种人工智能芯片，并在2017年至2018年期间获得了近7.65亿美元的资金。最后，Cambricon Technologies，在2016年开发了世界上第一台用于手机的商业深度学习处理器，获得了由中国政府支持的国家开发投资公司的拨款100万美元。

 

While China is rising, the European Union is declining. European industry still holds market share in areas such as sensors, but it has given up production of advanced digital semiconductors. Furthermore, there are indications that the EU will continue to lag in the development of advanced AI chips, which are costly and have long development cycles. Firstly, no EU semiconductor company ranks among the top ten in R&D spending. Secondly, some of the most innovative chip designs come from US and Chinese companies, such as Alphabet, Facebook, and Baidu. However, due to the fragmented market and competition regulations on the continent, EU digital startups have struggled to scale. And in Europe, companies comparable to Alphabet and Baidu lack the money and motivation to design AI chips. Thirdly, non-EU companies are acquiring promising European semiconductor design companies. The Japanese conglomerate SoftBank acquired the British semiconductor company ARM for $32 billion in 2016. Similarly, the Chinese government-backed private equity firm CanyonBridge acquired Imagination Technologies, another British semiconductor designer.

 

 

Policy Recommendations

 

The report also provides a series of policy recommendations to help the three countries improve AI capabilities; only the main points are excerpted below.

 

China

 

Talent Should focus on improving the ability to cultivate talent and retain local talent.

Research Improve the quality of AI papers and patents.

Data Establish standardized data formats and reduce restrictions on cross-border data flow.

Hardware Offer AI degree programs at universities.

 

European Union

 

Talent Take measures to incentivize talent to remain in the EU.

Research Increase R&D spending and provide more funding for startups.

Data Make full use of public data and establish and implement data policies among member states.

Application Reduce barriers to public sector application of AI and promote the application of AI technology by private enterprises.

Regulations Provide equal protection, rather than equal regulation, for emerging business models and existing businesses.

 

United States

 

Talent Increase the domestic talent base and encourage foreign talent to immigrate to the United States.

Research Increase funding for computer science research and increase R&D investment in private enterprises.

Data Maximize the value of the data it generates and create an environment for data sharing.

Application Promote the value of artificial intelligence to the public and businesses, promoting the use of AI.

Hardware Regain its leading position in high-performance computing and develop AI supercomputers.

Regulations Policy should avoid restricting innovation and should restrict Chinese acquisitions of US companies.