In 2024, China's breakthroughs in the field of large models are quite exciting:

Baidu's Wenxin Yiyan, Zhishu GLM-4, and other domestic models have defeated GPT-4 in global authoritative tests; three domestic large model unicorns such as Moon's Dark Side have broken through the valuation of 20 billion yuan; "domestic Sora" dark horses like Aishi Technology and Shengshu Technology have refreshed the AI video generation time to 6 seconds, 14 seconds, and 30 seconds; Wan Ka cluster has broken out of domestic GPU dark horses such as Moore Threads, supporting the training of trillion-parameter AI large models...

All of these are related to our country's "AI First City" - Beijing.

In July of this year, the person in charge mentioned at the 2024 Global Digital Economy Conference Artificial Intelligence Special Forum that as of then, there were 166 large model products nationwide that had been filed, of which 71 were in Beijing, accounting for nearly half of the country. It can be known from public information that the number of large model unicorns in Beijing accounts for 80% of the country, and at the same time, these large models have accelerated the landing in various fields such as the Internet, education, medical care, finance, and government.

Advertisement

At the end of July, Zhidongdong had counted the financing situation of domestic large model industry-related enterprises over ten million yuan in the first half of 2024 and July, and the results showed that Beijing enterprises obtained 43 financings, while Shanghai and Shenzhen were 11 and 9 respectively, showing a clear gap. ("Large Model Hot Money Surge: Hundreds of Billions of Financing in Half a Year, the Highest Over 50 Billion Yuan")

All signs indicate that Beijing is becoming China's "First City of Large Models."

From starting to leading, it only took Beijing more than a year to break through the large model industry. In May 2023, Beijing issued "Several Measures to Promote the Innovative Development of General Artificial Intelligence in Beijing," announcing 21 measures such as "carrying out research on large model innovative algorithms and key technologies," and took the lead in the country's cities to fire the first shot in large models.

More than a year later, Beijing has formed a comprehensive coverage of large models, AI computing power, data, applications, and solutions, as well as the entire industry chain and industry-university-research integration of large model "roadmap." In Beijing, not only can you see China, but also the world.Every "first," "strongest," and "blockbuster" development in Beijing is a testament to the Chinese large model industry's transition from catching up to leading. Why is it Beijing? And how has Beijing taken the central position in the "hundred model wars"? Let's explore together.

I. Nearly half of the large models are filed in Beijing, with top large models "made in Beijing"

Entering 2024, Beijing's strength in "producing" large models has been continuously confirmed.

At the second Beijing Artificial Intelligence Industry Innovation and Development Conference in March, the person in charge of Beijing mentioned that the number of AI-related enterprises in Beijing is about 2,200, accounting for about 40% of the country;

At the 2024 Global Digital Economy Conference in July, Beijing announced that the number of large models that have been filed and launched has reached 71, close to half of the national total of 166.

Previously, in October 2023, the "Beijing Artificial Intelligence Industry Large Model Innovation Application White Paper (2023)" showed that about half of the large model manufacturers in China with a scale of more than one billion parameters are in Beijing, with a number as high as 122.

Star large model products and companies "made in Beijing" have been continuously emerging in the Chinese technology circle.

The technology giants have set off the "hundred model wars," and Beijing has become the most cutting-edge innovation base. Technology giants such as Alibaba, Tencent, Baidu, ByteDance, Huawei, 360, JD.com, Xiaomi, and NetEase have all announced the latest large model research and development and application progress, with large model e-commerce, large model on the car, and AI mobile phone launches becoming hot trends.

For example, Baidu launched the Wenxin large model 4.0 version in October 2023, which is comparable to GPT-4. More than ten AI native applications and large model hardware were launched with it. By the end of the year, the user scale had broken through 100 million, and this year, it has upgraded multiple applications such as AI programming assistants, digital people platforms, intelligent customer service platforms, and knowledge management platforms.

The "Five Tigers of Entrepreneurship" in the domestic large model have emerged, becoming unicorns with a valuation close to 20 billion yuan. Among them, Zhi Pu AI, Bai Chuan Intelligence, Moon Dark Side, and Zero One All Things are all registered in Beijing, and MiniMax is located in Shanghai.Founded in 2019, ZhiPu AI launched its new generation base model GLM-4 in January 2024, with overall evaluation results approaching GPT-4; in July 2024, it launched the "Chinese version of Sora" Qingying, which supports the generation of 6-second videos. The company has raised more than 400 million USD in financing this year, and its latest valuation has exceeded 20 billion yuan.

Baichuan Intelligence, founded in 2023 by former Sogou CEO Wang Xiaochuan, has now launched 12 large models including Baichuan-7B/13B, Baichuan2-7B/13B, Baichuan2-192K, Baichuan 4, and "Pengcheng-Baichuan·Brain Sea 33B". The company just completed a 5 billion yuan Series A financing round in July this year, and its valuation has also broken through the "big threshold" of 20 billion yuan.

Founded in 2023, Moon's Dark Side has created the AI hit product Kimi with a monthly visit of more than 20 million people through its unique "To C" strategy. The company obtained more than 1 billion USD in financing in February 2024, setting a new record for the largest single financing in the domestic AI field. The latest valuation has reached 3 billion USD, also exceeding 20 billion yuan.

AI 2.0 company Zero One All Things, incubated by Dr. Li Kaifu in person, open-sourced the Yi-9B model in March 2024 and released the Yi large model API open platform. It has also recently been reported to have completed a new round of financing of several hundred million USD.

Some large model "new seedlings" have not yet joined the unicorn ranks, but they have become "kings" in the segmented track after only about one year of establishment.

In the popular field of AI video generation, Ai Shi Technology, led by former ByteDance visual technology director Wang Changhu, launched the Sora-like video generation product PixVerse overseas version as early as January 2024. Shengshu Technology, led by Tsinghua University Professor Zhu Jun and others, launched China's first original full self-developed long-duration, high consistency, and high dynamic video large model Vidu in April 2024, both leading OpenAI's "futures" product Sora.

In the highly watched field of AI infrastructure (AI Infra), Silicon Flow, founded by former OneFlow founder and CEO Yuan Jinhui, just launched a one-stop large model API cloud service platform SiliconCloud in June. It is reported that its self-developed SiliconLLM large model inference engine is more than 10 times faster than similar open-source products.

Industry-university-research joint innovation has become a trend. For example, in the field of general large models, the Beijing Academy of Artificial Intelligence and China Telecom jointly released the world's first low-carbon, high-performance open-source multi-language large model Tele-FLM in July 2024, and announced that they will release large language models with 100 billion, 60 billion, and 1 trillion parameters next.

The landing of large model industry applications has just begun to show results. In June 2023, Beijing released the first batch of 10 typical application cases of industry large models, including Baidu, ZhiPu Huazhang, the Institute of Automation of the Chinese Academy of Sciences, iFLYTEK, Yunzhisheng, the Institute of Scientific Intelligence, the Fourth Paradigm, Haomowei Zhixing, Xian Yuan, and Mianbi Intelligence, covering energy, medical, financial, consumer, intelligent driving, government, real estate and other industry fields.

 In July 2024, the Beijing Industry Grand Model Innovation Application Competition held its award ceremony, with a total of 6 companies winning the first prize. They are RuiLai Technology (Government Affairs), ZhongKeWenGe (Finance), RuiBo (Industrial Upgrade), ShuKun Technology (Medical), ZhongKe Brain (Cultural Education), and ZhongKe RuiJian (Smart City).

In just one year, the number of Beijing's main large models has grown from 40 in July 2023 to 71 in July 2024. Technology giants have laid out the entire industrial chain of large model computing power, algorithms, and applications. Dozens of unicorns and leading startups in niche tracks have caught up with international peers. The integration of industry, academia, and research has promoted the explosive application of landing results and has taken the lead in forming an AI grand model cluster area nationwide.

Beijing has run out the Chinese speed of large model industry development and has also competed for the Chinese strength.

II. The launch of the four major E-level intelligent computing centers activates the data elements with 36,000P intelligent computing power.

To open Beijing's "model" power, it is inseparable from the "key" of computing power.

As of December 2023, Beijing has formed a computing power supply scale of 12,000P (PFlops, 100 trillion floating-point operations per second) - among which general computing power is 8,672P, intelligent computing power is 3,402P, and supercomputing power is 340P. The intelligent computing power of about 4,082P deployed by the capital's cloud service providers in the Tianjin, Hebei, Shanxi, and Inner Mongolia regions has provided a solid foundation for the development of the large model industry.

What is this concept? 1P refers to a computing speed of 1,000 trillion times per second, and 100P is equivalent to the computing power of 50,000 high-performance computers. For example, to locate a certain characteristic star in a starry sky map of 200,000 celestial bodies, it only takes 100 seconds. What kind of "model" power 12,000P will bring to Beijing can be imagined.

Entering 2024, Beijing's intelligent computing infrastructure construction is still strong, continuing to provide energy for the development of the large model industry.

In January 2024, the Beijing Artificial Intelligence Public Computing Platform (Shangzhuang) located in Haidian District was released, and the first phase of 500P computing power has been officially launched, providing inclusive computing power services for the development of the large model industry.

In March 2024, as the largest public computing platform in Beijing, the Beijing Yizhuang Artificial Intelligence Public Computing Platform was officially put into use, providing computing power of 3,000P, equivalent to the computing power of more than one million high-performance computers.The area will issue "100 million computing power vouchers + 100 million model vouchers" annually to serve enterprises; looking to the future, the economic development zone where the platform is located has planned a computing power scale of over 10,000P.

In May 2024, the infrastructure construction of the Beijing Digital Economy Computing Power Center project located in Chaoyang District has started, and it is expected to be completed by the end of 2024. After production, it will cumulatively achieve an intelligent computing power supply of 2000PFlops.

△Beijing Digital Economy Computing Power Center effect drawing

According to the latest released "Beijing City Computing Power Infrastructure Construction Implementation Plan (2024-2027)", Beijing plans to build 4 E-level (EFlops, floating-point operations per second) intelligent computing power highlands, and the new intelligent computing power will reach 36,000P by 2027.

Beijing not only provides momentum for the development of the large model industry from the computing power level but also provides sufficient "fuel" from the data element aspect.

Data from the Beijing Municipal Economic and Information Bureau shows that after the Beijing International Big Data Exchange achieved the landing of the license, as of November 2023, the exchange issued 27 data asset registration vouchers, and the data transaction scale exceeded 2 billion yuan.

Behind the scenes, Beijing took the lead in creating the first national data basic system pilot zone, issued a plan for the creation of the data basic system pilot zone and a policy list; introduced innovative incentive measures such as the first registration, first transaction, first entry, and first opening of data assets, and continuously released policy dividends.

In March 2024, Beijing launched the first domestic AI data training base, while ensuring large model training with the "regulatory sandbox" system. The AI data training base, in conjunction with Beijing's computing power infrastructure, greatly helped enterprises in the large model field to develop the "strongest brain".

It can be seen that by grasping the three major elements of models, computing power, and data, Beijing has grasped the "bull's nose" of the development of large model technology industry.III. 30 Years of AI Evolution: Beijing Becomes China's "Capital of Large Models"

How did Beijing gradually become the capital of AI? This story dates back to more than 30 years ago.

The rise of Beijing's AI industry has a long historical origin. The artificial intelligence research of the Chinese Academy of Sciences began in 1987, the Microsoft Research Asia was established in 1998, and the famous Tsinghua Yao Class was founded in 2005... They have created batches of talents for the development of Beijing's AI, accompanying the first wave of AI tide led by SenseTime, Megvii, and the Beijing Academy of Artificial Intelligence in the early 2010s, and also becoming the mainstay of the current large model industry development in the 2020s.

Accumulating strength and then unleashing it, with the AI technology industry represented by computer vision landing and maturing, 2020 was the year of the outbreak of China's AI landing, and also the first major outbreak of Beijing's AI.

According to the "2020 Beijing Artificial Intelligence Development Report" released by the Beijing Academy of Artificial Intelligence, the output value of Beijing's artificial intelligence industry in 2019 has reached a scale of 170 billion yuan; by the end of June 2020, the total financing amount of AI startups exceeded 300 billion yuan, and has given birth to 32 AI unicorn enterprises such as SenseTime and Megvii; the number of AI-related papers published and patent applications are both the first in the country, and a number of major original achievements with international influence have been produced, all ranking first in the country.

This is inseparable from Beijing's ability to grasp the key points in AI policy.

In September 2017, the Management Committee of the Zhongguancun Science and Technology Park took the lead in releasing Beijing's first AI policy "Zhongguancun National Innovation Demonstration Zone Artificial Intelligence Industry Cultivation Action Plan (2017-2020)", and in December, Beijing issued the "Guiding Opinions on Accelerating Scientific and Technological Innovation and Cultivating the Artificial Intelligence Industry in Beijing", providing a policy foundation and support for the development of Beijing's AI industry. In 2019, Beijing was approved to establish the first national new generation of artificial intelligence innovation and development pilot zone, which also provided conditions for its AI outbreak.

Taking advantage of the situation, with the global explosion of large model applications represented by ChatGPT, 2023 is the year of evolution for Beijing's AI industry.

1. The first shot in the national AI large model policy, 100 billion yuan to support the investment and financing environment

In terms of AI large model policy, Beijing started early, without hesitation, and without waiting.On May 30, 2023, Beijing took the lead in launching the first local policy for large-scale models. The Beijing municipal government issued the "Implementation Plan for Accelerating the Construction of an AI Innovation Hub with Global Influence" and the "Several Measures to Promote the Innovation and Development of General Artificial Intelligence in Beijing," proposing 21 measures including strengthening the development of data collection and governance tools for large model training, setting the tone for the development of large model industries in Beijing.

In September 2023, the Haidian District of Beijing took the first step to set an example. The district issued the "Implementation Plan for the Innovation and Leading Development of General Artificial Intelligence in Zhongguancun Science City" (referred to as the "Plan"), which proposed to create a core industrial scale of 230 billion yuan, providing talent subsidies from multiple aspects such as talent settlement, financial subsidies, and public rental housing.

In April 2024, Beijing released another major policy. The Development and Reform Commission and other departments further jointly issued the "Several Measures of Beijing on Accelerating the Leading Development of the General Artificial Intelligence Industry" (referred to as the "Measures"), focusing on improving the supply of intelligent computing power and accelerating the innovative application of large models, proposing 10 measures, and providing a "big gift package" to large model enterprises from a full range of computing power, models, data, applications, and environment.

These "big gift packages" include but are not limited to: the highest reward of 100 million yuan for AI technology research and development projects included in the national major strategic tasks, the highest support of 50 million yuan for projects to create benchmark applications of artificial intelligence, an investment of more than 100 billion yuan in the next five years to optimize the investment and financing environment, and reducing the operating costs of large model enterprises through financial subsidies, issuing computing power coupons, model coupons, and other methods.

2. Create a 45EFLOPS intelligent computing supply, and support the training of large models with trillion-level parameters with domestic computing power

In terms of the construction of large model infrastructure, Beijing maintains its strong investment strength.

In April 2024, the Beijing Municipal Economic and Information Bureau and the Beijing Municipal Communications Administration issued the "Beijing Municipal Implementation Plan for the Construction of Computing Power Infrastructure (2024-2027)" (referred to as the "Plan"). The plan proposes that by 2025, the scale of intelligent computing supply in Beijing will reach 45EFLOPS; by 2027, the hardware and software products of intelligent computing infrastructure will be fully controllable, the overall performance will reach the leading level in the country, and have the ability to build 100% controllable intelligent computing centers.

Beijing also announced the creation of more than 200 industry benchmark cases driven by intelligent computing power, focusing on building Haidian, Chaoyang, Yizhuang, Jingxi (Shijingshan, Mentougou) and other E-level intelligent computing highlands, and accelerating the construction progress of two 10EFLOPS large-scale intelligent computing clusters.

However, Beijing's computing power map is not limited to this. It is planned to build a Beijing-Tianjin-Hebei-Inner Mongolia computing power supply corridor, plan and build a super-large-scale intelligent computing cluster that supports the training needs of large models with trillion-level parameters, and accelerate the commercial landing of artificial intelligence large models in a more grand layout.

It is worth mentioning that accelerating the independent control of core hardware and software products will be a key task.In the proposal, Beijing has put forward that for enterprises that purchase independently controllable GPU chips to carry out intelligent computing power services, the implementation plan's safeguard measures are to provide support according to a certain proportion of the investment, accelerating the realization of independently controllable intelligent computing resource supply.

In the "Beijing Municipal Implementation Plan for Accelerating the Construction of an Artificial Intelligence Innovation Origin with Global Influence (2023-2025)" in May, Beijing also clearly proposed to promote breakthroughs in domestic artificial intelligence chips. It actively guides large model development enterprises to apply domestic artificial intelligence chips, accelerating the rate of domestication of artificial intelligence computing power supply.

Beijing's substantial investment in domestic computing power is also a "big move".

For example, in October 2023, the Beijing Municipal Economic and Information Bureau released the "Beijing High-end and Precision Industry Development Fund Implementation Guide (Second Batch)" last night, which proposed to focus on supporting the direction of integrated circuit design products for the first round of tape-out rewards, with a single enterprise's annual reward amount not exceeding 30 million yuan.

3. Building a nest to attract phoenixes, mobilizing the national scientific and technological strength to "form a game" and make the government-industry-academia-research capital ecosystem "active"

The integration of the government-industry-academia-research capital ecosystem is the source of vitality for the development of Beijing's large model industry.

Building a nest to attract phoenixes, Beijing has allocated several plots of land in the capital, where land is scarce and expensive, to develop the AI industry.

In September 2023, the "Zhongguancun Artificial Intelligence Large Model Industry Cluster Area" was unveiled, creating the Wudaokou Artificial Intelligence Industrial Park, Peking University West Gate Artificial Intelligence Industrial Park, Zhongguancun West Area Artificial Intelligence Industrial Park, and Tsinghua Science and Technology Park, covering an area of 670,000 square meters. This provides a broad physical space for the grand blueprint of Beijing's large model industry.

In July 2024, the concept of the artificial intelligence innovation block was released. Beijing plans to create an artificial intelligence innovation block in the 53 square kilometers of the southern part of Haidian, connecting 52 national key laboratories, 10 new research institutions, 106 national research institutions, 37 top universities, 89 AI2000 top global scholars, 12,300 artificial intelligence scholars, 1,300 artificial intelligence enterprises, more than 90 units related to large models, and 37 units of large models for the record.

Through a series of partnership plans and alliances, Beijing has made AI entrepreneurship spaces "active".In May 2023, Beijing launched the General Artificial Intelligence Industry Innovation Partnership Program, announcing the first batch of 39 members that month, and in July, the second batch of 63 enterprises was announced, including computing partners such as Alibaba Cloud and Beijing Super Computing Center, data partners like Beijing Municipal Big Data Center, Beijing Academy of Science and Technology, and Beijing Academy of Artificial Intelligence, model partners including Baidu, ZhiPu AI, Alibaba's DAMO Academy, and Kunlun Wanwei, application partners like Beijing Municipal Government Services Administration, Bank of Beijing, and Kingsoft Office, and investment partners such as IDG Capital, CDH Investments, and Miracle Plus.

In July 2024, the Zhongguancun Artificial Intelligence Industry Alliance was initiated and established, launching the General Artificial Intelligence Industry Innovation Ecological Plan. The alliance members include the Beijing Academy of Artificial Intelligence, China Information and Communication Technology Research Institute, Peking University, Tsinghua University, Renmin University of China, Institute of Automation of the Chinese Academy of Sciences, Zhiyuan Research Institute, Haitian Ruisheng, ZhiPu AI, Tencent, CRRC, Baidu, Baichuan Intelligence, and Zhongguancun Science City Company, among hundreds of universities and institutions.

It can be seen that to reflect the market-oriented open cooperation and docking mechanism, Beijing has mobilized the investment, industry, and academic forces of more than half of China to jointly promote the development of the large model industry.

The results of these actions are not just a "show of strength," but also a model for the whole country, using the results to nourish the national large model industry and the development of the digital economy.

Conclusion: To create a benchmark for the digital economy, Beijing occupies the "hundred models battle" C position.

From digitalization to intelligence, Beijing is once again at the forefront of the era's leap.

In 1994, China first accessed the Internet, and Beijing took the lead in establishing the Municipal Information Office in 1997 and set a long-term "Digital Beijing" three-step goal in 1999: 2000-2010, to form the basic framework of "Digital Beijing". 2011-2020, to improve the capital's digital level. 2021-2050, to build Beijing into a modern international digital city.

Leading the trend of broadband Internet, high-definition digital TV, and seizing the opportunities of e-commerce, digital logistics, and digital government, Beijing has created historic events such as the launch of the "Capital Window" website, the popularization of the municipal traffic one-card pass, the 2008 digital Olympics, and the 3G network service... According to official data, Beijing's digital economy added value reached 871.9 billion yuan in 2015 and has grown to 1.7 trillion yuan by 2022, becoming one of the main engines of economic growth.

AI, as an important sub-topic of digitalization and a representative of new quality productivity, has not only taken root in the fertile soil of Beijing's digitalization but also sprouted under the active AI special policy irrigation. With policies, industries, academia, capital, and applications working together, Beijing has become a leading center for artificial intelligence industry talent in China and even globally. Behind the first city of China's large models, Beijing's moat is increasingly deepening.