[Visible Commentary] DeepSeek's Impact and China's AI Innovation Ecosystem

YouTube link: https://www.youtube.com/watch?v=LM5DWwRjx84

Video Script

I will omit the parts that Center Director Ha Jeong-woo has already mentioned and focus my presentation on China and its innovation ecosystem. When we look at China as a whole, AI gained attention in stages. We paid significant attention during the era of Vision AI. Before COVID-19, China was highly recognized for its advancements through companies like SenseTime. At that time, prominent Chinese think tanks and government-designated large corporations included those we know well today: Alibaba, Baidu, Tencent, and Didi. They also continuously selected the top 100 promising AI startups in 2019, 2018, and 2017.

At that time, companies like SenseTime and Megvii were receiving attention. While these companies, excelling in Vision AI, have become the parent companies of the deep learning firms that impress us today, they did not consistently receive such spotlight. After experiencing COVID-19, China's innovation ecosystem faced a shock. Subsequently, as the LLM trend shifted from Vision AI, the general perception was that China lacked capabilities, leading to assessments of it being a 'paper tiger.' Following that period, lists selected by China in 2024 began to feature LLM-related companies, and there were indeed many excellent companies. Companies like MiniMax, which developed Keyi, and Day, mentioned by Dr. Ha earlier, gained attention then. Thus, when these companies emerged, and China was recognized in 2023 and 2024 for still possessing AI capabilities and fostering good companies after COVID-19, they were not widely known to the public.

Structure of China's AI Ecosystem and the Emergence of DeepSeek

or perhaps only to those deeply involved in tech since last year, but generally, they did not receive sustained attention. What does this imply? It is often said that once Chinese companies emerge, the government plays a significant role in nurturing them. While China's policy and AI ecosystem support are generally more robust than in other countries, it is more appropriate to consider their emergence as unplanned. In 2023, China's Ministry of Science and Technology mapped out the entire LLM ecosystem in China, and it did not differ significantly from the Vision AI era. Beijing remained first, Shanghai second, followed by Guangdong and then Hangzhou. While we must look at Hangzhou, where DeepSeek originated, to understand DeepSeek, Hangzhou possesses a strong innovation ecosystem, but it ranks only about fourth nationally. It can still be considered Tier 1.

This implies that the companies most heavily promoted and supported by the Chinese government were the earlier ones, and even the support from key cities like Beijing, Shanghai, and Guangzhou was more concentrated on them. These companies, however, emerged as later entrants. This distinction seems meaningful. Regarding the shock, which has been discussed extensively, we are aware that price or performance is not fundamentally changing the paradigm.

However, the significance lies in the fact that regardless of how meticulously the US implements export controls, it is fundamentally difficult to keep pace with the speed and agility of private sector R&D and pivoting. Therefore, while export controls have had and will continue to have an impact, it is difficult to track all private companies or prominent Chinese companies, which would also be of concern to the US, that pivot from entirely different sectors. Secondly, fundamentally, users want what is cheap and good.

Drivers and Ecosystem of China's AI Innovation

Services, but this cannot be stopped. When companies are determined to offer their development for free and compete on open-source advantages, it is difficult to block this through security concerns. I believe there is a structural issue here. From China's perspective, while US data is widely introduced, in-depth content emerges from Chinese internal perspectives and interviews with Chinese officials. While there may be some intent, the consistent message is that they are not focused on immediate profits; their goal, as stated in interviews, is AGI. They aim to conduct research that fundamentally changes paradigms and will focus solely on that. As mentioned earlier, the parent companies had substantial capital and sufficient GPUs, allowing them to support talent in pursuing challenging, engineering-focused tasks without GPU constraints, often with intense work demands. Most of this can be seen as bottom-up. Based on interviews, this is the case. Although DeepSeek may seem to have appeared out of nowhere, technically, much of it was disclosed last year, suggesting its existence. Ultimately, it was able to emerge because the ecosystem's water level was already high. The ecosystem was largely set up with China's national AI strategy in 2017. Firstly, there was data, infrastructure, and universities were adequately established by then, and the proliferation of companies was highly beneficial.

Consequently, as these individuals graduated and entered the workforce, they were able to acquire practical skills by solving real-world problems and developing businesses. Structurally, due to the US-China hegemony and export controls since Trump's first term starting in 2018, there has been a strong imperative to reduce reliance on China. Prior to that, China relied externally for core technologies. This is often criticized within China as a chronic issue in indigenous technological innovation. Recognizing that these dependencies became targets for export controls and sanctions, and hindered growth, significant efforts were made. For instance, open-source projects like Baidu's reached 10,000 contributions during COVID-19. Preferences for RISC-V, a decline in students studying in the US, and difficulties in joint research with the US led to

environmental requirements for collaboration with domestic Chinese institutions and Alibaba. These factors played a role. I will skip the core competencies as they have been discussed extensively. Please refer to Dr. Ha's earlier remarks. The significance lies in the Chinese perspective, namely the pursuit of indigenous core technologies. China recognizes that paradigm-shifting advancements in cutting-edge technology originate in the US, and they need more companies capable of achieving this.

These emerged as indigenous companies. Another significant aspect is that the research teams within them consist of highly skilled individuals, many without prior overseas study experience. Technically, one must acknowledge their originality and innovation. From a Chinese perspective, however, it represents a step backward, originating from the product application stage, an area where China typically excels. They are utilizing AI, but the foundational models are largely developed elsewhere, in the US or other countries. The innovation lies in the endeavor to develop their own foundational models. This signifies a deeper penetration into the field. While companies like Alibaba and Baidu have traditionally excelled, the emergence of a highly capable, albeit small, company from a favorable region, which was not previously considered a leading contender but is now highly promising, is noteworthy.

The Interplay Between China's Industrial Policy and AI Innovation

Secondly, China's economic growth over the past 20-30 years has led to its evaluation as an innovative nation, but this has evolved. For example, with the rise of smartphones, was the innovation truly indigenous, or was Apple's role immense, given its presence in China? The situation with electric vehicles is similar; China allowed Tesla full ownership without a joint venture stake. Thus, a method China has effectively employed, especially recently, to foster industrial growth is by attracting global leaders to establish a presence, thereby creating a ripple effect. In artificial intelligence, while there was a desire for assistance from foreign companies, it can be seen that the 'bait' role, which was previously fulfilled by imports, has now naturally transitioned to indigenous 'bait.' This is significant.

AI Innovation in Authoritarian States and the Impact of DeepSeek

While we are not political scientists, from this perspective, it seems counterintuitive that China, a controlled, authoritarian state, would be capable of scientific and technological innovation. The democratization of AI, paradoxically playing a role in China, is a complex issue. What is the current effect within China? Following DeepSeek's emergence, recent reports indicate that major provincial governments and corporations in China have announced the adoption of DeepSeek. Yesterday, Tencent finalized its decision to integrate it, and all 15 state-owned enterprises are reportedly adopting it. As you can see in the bottom right, all three major telecommunication companies—China Mobile, China Telecom—are adopting it. It has also been implemented in government administration, and while the pace has likely accelerated, as of the compilation of these materials yesterday, Beijing, Haidian, and other administrative bodies have fully integrated DeepSeek.

Secondly, DeepSeek's emergence has sent shockwaves globally and within China's big tech sector. Consequently, Baidu has released its open-source model, and TikTok has lowered prices, among other actions, seemingly invigorating the Chinese AI spirit and promoting overall activation. What will be the effect? In terms of AI adoption and diffusion, competition is intensifying, leading to lower prices, faster speeds, and the emergence of more alternatives and solutions. This is how diffusion will likely occur. You may proceed to the next slide. Yes, the Chinese Ministry of Industry and Information Technology annually selects and announces the best AI application cases at the end of the year. As you can see on the left, in Chinese characters, it pertains to creating basic guidelines and industry codes. The 2023 selections primarily utilized Baidu and Alibaba's offerings.

The Emergence of DeepSeek and the Revitalization of China's AI Industry

Second, when DeepSeek exploded, it not only caused a global shock but also impacted Chinese big tech. Consequently, Baidu immediately released its open source, and TikTok lowered its prices, which seems to have invigorated the Chinese AI spirit and led to overall activation. What effect might this have on the diffusion of AI adoption? Firstly, overall competition has intensified, leading to lower prices, faster speeds, and the emergence of alternatives, which in turn can foster more solutions. Diffusion is expected to occur along these lines. You may press again. Yes, the Chinese Ministry of Industry and Information Technology selects and publicizes the best AI application cases annually at the end of the year. As you can see on the left, it's in Chinese characters. For the existing 2023 guidelines, such as establishing basic guidelines and industry codes, almost all of them

However, with the advent of DeepSeek, as mentioned earlier, there is a potential for significant change. Moving on, this is similar. This was selected at the end of this year. Ultimately, China's comparative advantage lies in the vast number of areas where AI can be applied, given its developed manufacturing sector and extensive value chains. This diffusion is expected to accelerate significantly. Baidu's LLM provision for Chinese power plants yielded results, but they were not substantially improved. With the emergence of DeepSeek models, efficiency could improve, or conversely, Baidu might be spurred to develop superior solutions due to competitive pressure. Compared to last year, the application areas have expanded considerably, now encompassing almost all sectors. The background to DeepSeek's emergence suggests that within China, there is a significant potential to accelerate the diffusion and application of AI at better prices and with higher quality. This is the impact it has within China. Globally, as previously discussed, while it may not be a complete paradigm shift, it represents a meaningful engineering breakthrough. Regarding export controls, there are sentiments that they should have been stricter and will need to be in the future. A recent publication by the Director of the CSIS AI Center Policy Center acknowledged the technical breakthroughs and progress but concluded with an impactful statement: 'What Chinese companies can do, we can do, and the gap still exists.' The US companies will continue to advance because Chinese companies cannot replicate their work. There is a fundamental difference in competitiveness; their work is replication, while much of ours remains unreproducible. The impact on countries other than the US has been discussed extensively. New opportunities have opened up, and from a corporate perspective, concerns about how to collaborate with China technologically have intensified since DeepSeek's emergence. Initially, the thought was to cease and desist, but with DeepSeek's appearance, large corporations are now considering whether collaboration with China is necessary, thinking that such excellent advancements might exist in other areas.

US-China AI Competition and the Shifting Landscape of Global Technology Cooperation

This has led to a sudden surge in interest in China. As mentioned, the AI landscape is entering its second round. In terms of application areas, humanoid robots are likely to continue gaining attention. AI's integration with physical robots and its expansion into the market structurally favors China. China excels in robot production and cost-effective manufacturing. The companies on the right, known as the 'Six Dragons of Hangzhou,' include Unitree, a renowned humanoid robot company. With Unitree's backing, the overall value chain shows that prior to DeepSeek's release, Unitree had already launched products equipped with Baidu's Wanibot, integrating it into components. Following OpenAI's product release, they immediately replicated it.

The Convergence of AI and Robotics: China's Structural Advantages

and launched it, which is now incorporated into DeepSeek. Chinese humanoid robots have significantly improved by diligently imitating US models. With the support of the regional and robotics ecosystems, the integration of AI robots will complicate competition. It is a fact that China holds a structural advantage as production and manufacturing become closer. This is something to watch moving forward. Beyond Hangzhou, the ecosystem in the Shanghai region is also excellent. While DeepSeek is currently the focus, more informed individuals mention MiniMax and Qwen. In fact, Shanghai hosts one of the nation's premier AI research institutions, the Shanghai AI Laboratory, alongside numerous AI research centers within prestigious universities. Examining the research conducted here is likely to reveal significant potential for progress.

While acknowledging this, the concluding remark was striking: 'What Chinese companies can do, we can do, and the gap still exists.' We must not overlook the fact that the gap still exists. The reason US companies continue to advance is that Chinese companies cannot replicate their work. There is a fundamental difference in competitiveness; their work is replication, while much of ours remains unreproducible. Therefore, regarding the impact on countries other than the US, much has been discussed previously. New opportunities have opened up, and from a corporate perspective, concerns about how to collaborate with China technologically have intensified since DeepSeek's emergence. Initially, the thought was to cease and desist, but with DeepSeek's appearance, large corporations are now considering whether collaboration with China is necessary, thinking that such excellent advancements might exist in other areas.

The US-China Technology Gap and China's Capacity for Imitation

This has led to a sudden surge in interest in China. As mentioned, the AI landscape is entering its second round. In terms of application areas, humanoid robots are likely to continue gaining attention. AI's integration with physical robots and its expansion into the market structurally favors China. China excels in robot production and cost-effective manufacturing. The companies on the right, known as the 'Six Dragons of Hangzhou,' include Unitree, a renowned humanoid robot company. Unitree's backing means that in the overall value chain, prior to DeepSeek's release, Unitree had already launched products equipped with Baidu's Wanibot, integrating it into components. Following OpenAI's product release, they immediately replicated it.

and launched it, which is now incorporated into DeepSeek. Chinese humanoid robots have significantly improved by diligently imitating US models. With the support of the regional and robotics ecosystems, the integration of AI robots will complicate competition. It is a fact that China holds a structural advantage as production and manufacturing become closer. This is something to watch moving forward. Beyond Hangzhou, the ecosystem in the Shanghai region is also excellent. Currently, the focus is on DeepSeek, but more informed individuals mention MiniMax and Qwen. In fact, Shanghai hosts one of the nation's premier AI research institutions, the Shanghai AI Laboratory, alongside numerous AI research centers within prestigious universities. Examining the research conducted here is likely to reveal significant potential for progress.

and launched it, which is now incorporated into DeepSeek. Chinese humanoid robots have significantly improved by diligently imitating US models. With the support of the regional and robotics ecosystems, the integration of AI robots will complicate competition. It is a fact that China holds a structural advantage as production and manufacturing become closer. This is something to watch moving forward. Beyond Hangzhou, the ecosystem in the Shanghai region is also excellent. Currently, the focus is on DeepSeek, but more informed individuals mention MiniMax and Qwen. In fact, Shanghai hosts one of the nation's premier AI research institutions, the Shanghai AI Laboratory, alongside numerous AI research centers within prestigious universities. Examining the research conducted here is likely to reveal significant potential for progress.

China's AI Technological Self-Reliance and the Importance of its Ecosystem

These shocks are likely to become much more frequent. There is overestimation and underestimation, and Dr. Ha has done a lot of accurate evaluations. Overall, it seems to signify a further step towards meaningful technological self-reliance. Judging from the interviews with Yang Yuan, the assessments of China and DeepSeek are very accurate. Compared to the United States, although China has significantly improved its ecosystem level, the gap with the United States remains large. Cooperation is not smooth, and the technological gap is about 1-2 years.

Simply following technology does not make it all meaningful. Ultimately, it is merely a difference between developing fundamental technology and applied technology. The gap is about 0.1 years. Some argue strongly that even if the gap narrows to 0.01 years, it is meaningless because those who develop fundamental technology create everything. The rebound and subsequent rise in GPU stock prices suggest that China, despite the difficulties, views this field as its final challenge and continues to attempt it. However, this is a very difficult task.

In fact, China is well aware of this internally. When I visited Beijing at the end of last year, Alibaba's presentation indicated that the domestic data center usage rate was only about 10%. They believe there is still a long way to go. Evaluating this, it is analogous to technology or food; the surrounding ecosystem that produces it is more important than the recipe itself. A common narrative is that China's lack of such an ecosystem is the core reason for the difficulty in achieving breakthroughs in GPUs. Therefore, it is judged that at this juncture, tacit knowledge within the ecosystem that can be comprehensively implemented and emerge is far more important than explicit knowledge of specific technological units.

China's AI Governance Index and Future Prospects

What will China's future hold? Highly outstanding companies may emerge from entirely unexpected places. Then, these achievements will begin to spread to the provinces. They will be told to create examples like DeepSeek. Everyone will focus on developing fundamental technology, and substantial support for DeepSeek is expected. If measures such as its adoption in the public sector within just one week are taken, the scale will expand after one cycle. When the AI governance index of the world was announced last year by the AI research institute affiliated with the Chinese Academy of Sciences, South Korea was not included.

South Korea was not included among the 15 major countries, while countries like Italy and India were. At that time, we suspected there might be an intention behind it. Why was it not included when, even with a conservative assessment, it would rank within the top 5? In February of this year, the same research, the AI Governance Index (AGI Index), was developed and released, and South Korea suddenly rose to 4th place. While this could also be seen as intentional, it is unlikely that deliberate efforts were made to elevate South Korea to a top position. However, this indicator suggests that our standing has risen due to the efforts of many, including numerous doctors, and meaningful safety-related efforts within the country. Considering this, even the UK, which led the initiative at the Trump administration's Action Summit, did not sign on and withdrew. How will China penetrate and exploit that vacuum? In multilateralism, we typically...

The Era of AI Security Competition and South Korea's Response Strategy

China will never deviate from negotiation tables favorable to them, or will seek to make them favorable. Regarding AI, China quickly emerged with a stance officially advocating multilateralism after the swift conclusion of the Action Summit. Therefore, their existing position will be maintained for the time being, and preparations are needed to address how to maintain and leverage this situation given the withdrawal of US research institutions. The Promising report, released in January by our Say AI Forum, also suggests that with the emergence of DeepSeek, the non-participation in this summit, and the complete shift to security competition, the previously emphasized aspects are becoming almost meaningless. As intergovernmental dialogue, technology standard setting, and industry consortia enter a full-fledged competition phase alongside the significant legislation you mentioned, the minimum red lines or safety aspects concerning AI...

...even if an agreement could not be reached on the need for US-China cooperation, there were still movements to do something. For instance, discussions arose about not using AI for weaponization, and it was suggested that this could have been possible through PAX this year, but it appears highly unlikely now. That is the current situation. What should we do in the interim? Although our competitiveness and ecosystem are not on par with the US or China, we are in the top three. If the race truly centers on security competition, we too must adopt a two-track approach to enter the top 3-4 positions. Strategies such as scientific diplomacy are being discussed. While goals like sustainable development and human society are important, we must also actively participate in this competition and strive to become one of the top 3-4 nations. This requires parallel efforts in R&D investment and diplomatic activities.

We must pursue a two-track approach, not just separately but by making large-scale R&D investments that act as leverage. While various indicators are being created by different countries and diverse judgments will exist, we need emergency measures and an equally aggressive strategy to ensure we are always on the table as a key negotiating party for that database. I will conclude my presentation here.