Lorenzo Maria Pacini
Almost no one expected it, yet it was predictable: China has entered the global AI competition and intends to win.
China takes measurements
Chinese companies in the AI field aim to achieve success in the same way that Chinese companies in the electric car, renewable energy and biopharmaceutical sectors are conquering the market: by radically changing the economics of these sectors. In other words, they aim to outcompete their rivals by making AI adoption low-cost and large-scale, thus eliminating rivals that operate with high-cost, high-margin business models.
Since 2018, the US government has sought to hinder the development of Chinese artificial intelligence, imposing restrictions on chip exports and banning access to the most advanced AI models created in the US. The launch of DeepSeek broke this blockade, demonstrating China's resilience and capacity for innovation.
Subsequently, the US has started to exclude DeepSeek from government agencies, while OpenAI is lobbying for a large-scale ban in the US. It is also possible that the US government will put pressure on its allies to impose restrictions on DeepSeek, as it has already done with Huawei. Other Chinese AI companies may soon face similar restrictions.
The response of Chinese companies is interesting. Since the launch of DeepSeek, there has been a flood of new high-performance AI models from China, such as Qwen from Alibaba, Doubao from ByteDance, Hunyuan from Tencent and Ernie from Baidu. Unlike their US competitors, these models are open source and free: they are available to anyone in the world to download, modify and integrate. But why are Chinese companies adopting this strategy?
Since the launch of ChatGPT in November 2022, major US tech players such as OpenAI, Microsoft, Google and Meta have followed a similar strategy: they have accumulated Nvidia's most advanced AI chips, invested huge resources in data centers, developed proprietary and closed language models, and applied high subscription or license fees to monetize their products.
These companies treat AI as an exclusive resource, limiting access to their most powerful models through paywalls. OpenAI, Google DeepMind, and Anthropic restrict access to their most advanced models, offering them only through paid subscriptions or corporate contracts. These AI programs are valued in the billions of dollars, with investors expecting huge financial returns.
In practice, Silicon Valley companies' investment in AI is based on a high-cost, high-margin business model, protected by a moat of intellectual property. The model is further supported by the prohibitive costs of accessing computing resources, which are only accessible to the richest tech giants, and which effectively prevent competition.
The Chinese strategy, on the other hand, is exactly the opposite. While advanced computing resources are difficult to obtain, even large Chinese companies are forced to develop innovative solutions to create high-performance models without resorting to the most advanced chips. Instead of focusing on raw processing, Chinese companies concentrate on intelligent engineering and algorithm optimization to develop their AI models. As their models are starting to reach the level of those in the US, the companies have decided to make their products open source to share resources with developers all over the world and accelerate improvements.
This approach offers numerous advantages:
- Low dependence on advanced AI chips
- Lower capital expenditure (capex) requirements
- Decentralization of development to take advantage of global AI talent
- Opportunities for developers who have access to more advanced chips to contribute to the refinement of the model
- Faster iterations: AI advances through continuous improvement, with each new version building on the previous one to refine capabilities and improve efficiency.
Thanks to open source, Chinese companies are creating an ecosystem where developers from all over the world can contribute to improving the models, without having to bear all the development costs.
Such an approach could profoundly transform the AI economy. If Chinese open source models were to achieve the same power as proprietary US models, the business model based on monetizing AI models would be called into question. Why pay for closed models when there are free, open and equally powerful alternatives?
By making fundamental AI models free and accessible, Chinese companies could destroy the pay-to-use business model based on closed and proprietary systems, which rely on huge capital investments. Such an approach would also reduce the importance of control over chips and nullify the economic advantages of US AI companies.
Of course, the free open source model is not a goal in itself, but rather part of a broader strategy. The ultimate goal of Chinese companies is to move AI from foundational models to applications, areas where China has concrete advantages, such as data and the market. Monetization will occur at the application level as AI is integrated into various industries and consumer use cases.
Instead of making money from AI models, Chinese companies will generate profits by selling AI solutions, building integrated artificial intelligence and incorporating AI into consumer goods and services. There are huge profit opportunities in areas such as humanoid robots, autonomous driving, intelligent infrastructure, industrial and healthcare applications, and much more.
The Chinese government is already accelerating the application of AI in its state-owned enterprises, from telecommunications and banking to ports, energy, and public services such as hospitals, schools, and government offices. Private companies in the automotive, electronics, pharmaceutical, and consumer goods sectors are also adopting AI. Once widespread adoption occurs, AI will be ubiquitous and accessible to everyone.
The open-source nature of Chinese AI models will stimulate global competition, creating a fair development environment. China aims to take full advantage of this situation, thanks to its huge market and the data that is essential for developing the best applications.
If China is successful in this endeavor, its AI success would be a victory similar to that achieved in the electric vehicle sector, where it "changed lanes" and beat the competition with a more agile and innovative approach.
DeepSeek as a threat
Since DeepSeek unleashed a global wave in artificial intelligence, the US narrative on the "Chinese threat" has evolved. From news about the US Department of Commerce banning the use of DeepSeek in government devices, to statements by Secretary of Commerce Howard Lutnick calling for stricter restrictions on open source AI models, especially Chinese ones, the US is expanding its containment strategies to the AI sector as well. Thus a new variant of the "Chinese threat" has emerged: the "Chinese AI threat".
Washington has already added 80 companies to its export control list, more than 50 of which are based in China, accusing them of seeking advanced expertise in supercomputing, AI and quantum technology for military applications. Furthermore, the annual report on global threats by US intelligence, published on Tuesday, claims that Beijing is developing linguistic models to spread fake news and aims to overtake the USA as the leading power in AI by 2030.
It's certainly no coincidence: every technological advance made by China in recent years has been met with alarm by the US. The underlying logic is clear: China must not prevail. As soon as Beijing shows signs of progress in a strategic sector, it is immediately labeled as a threat, followed by restrictive measures.
Looking back, the United States has already restricted Chinese companies' access to its market in the battery and electric vehicle sectors, but internal technological difficulties have prevented it from catching up. Now, the same strategy is being applied to AI. From imposing bans on the sale of chips to Chinese companies to pressuring its allies to adhere to the restrictions, every action aims to exclude China from the global technological system. However, history shows that these blockades have not only failed to work, but have often had the opposite effect, stimulating Chinese innovation and destabilizing international supply chains.
In addition to changing their anti-Chinese narrative, US restrictions risk backfiring on the US itself. The blocks will push Chinese companies to intensify independent research, accelerating their technological autonomy. The "threat of Chinese AI" is actually a reflection of American insecurity and fear of losing primacy in this sector.
The progress of AI depends on global cooperation. The United States insists on transforming AI into a geopolitical issue, promoting isolation and division, even going so far as to build a sort of "technological iron curtain".
Artificial intelligence could not only destabilize the Chinese labor market, but also put pressure on its energy and infrastructure systems. Despite the reluctance of large technology companies to recognize the problem - and even less to reveal the energy consumption of their data centers - AI, in particular large language models (LLMs), requires huge amounts of natural resources and energy. According to forecasts by the International Energy Agency (IEA), by 2026 data centers in China will account for almost 6% of the country's total electricity demand. The production of energy and the cooling of these structures require huge volumes of water. China Water Risk, an organization based in Hong Kong, estimates that the total water consumption of Chinese data centers could exceed 3 billion cubic meters by 2030, a value comparable to the annual water consumption of the entire population of Singapore. The so-called "hundred-model AI war" in China could lead to excessive competition for already limited computing resources, putting the country in a bad light on ecological issues.
Reconciling AI ambitions with climate goals represents a colossal challenge for Beijing. China aims to reach peak CO2 emissions by 2030, adopting a reduction strategy based on two pillars: instead of limiting itself to containing energy consumption, the government aims to control both the carbon intensity per unit of GDP and total greenhouse gas emissions. Although China is the world leader in renewable energy production, socio-economic factors and structural obstacles to the electricity grid mean that coal still accounts for two-thirds of the national energy mix. With the accelerated expansion of computing infrastructure to support the growing demand for LLM, there is a risk that the country's energy systems will not be able to keep pace with the AI boom. To mitigate this problem, the government is moving data centers and computational hubs towards cleaner and cheaper energy sources, introducing stricter standards on energy intensity and improving coordination in the use of computational resources.
However, AI could also offer opportunities for the Chinese energy sector. The concept of an "intelligent energy brain" is gaining popularity among state-affiliated policy makers and researchers, who promote the integration of computing power, artificial intelligence and energy economics. A government-led project, the Tianshu-1 system, has reduced energy consumption by more than 15% through the use of AI and big data for the prediction, management and maintenance of electrical networks. Chinese LLM developers are also trying to take advantage of this opportunity, targeting new customers and designing specific models for industrial applications. For example, China Southern Power Grid has collaborated with Baidu to develop artificial intelligence models for the energy sector. However, the success of these initiatives is not guaranteed.
These internal challenges are compounded by external factors, in particular China's dependence on US semiconductor technology for the development of AI. The competition between China and the United States in this field has taken on the characteristics of a real "technological arms race". In October 2022, the Biden administration introduced restrictions on the export of advanced semiconductors to China, including the latest graphics processing units (GPUs), which are essential for processing machine learning models. These restrictions also include the tools, software and expertise needed to produce cutting-edge chips. These measures, motivated by China's strategy of merging the military and civilian spheres and the use of AI in authoritarian surveillance programs, were further tightened in October 2023, with more expected in the future.
The US restrictions, which have extraterritorial application, further complicate Chinese energy management, as local companies are forced to use more older and less efficient chips for artificial intelligence activities. The CEO of DeepSeek, a Chinese LLM developer, admitted that the national models require four times the computational resources compared to the US ones, while still lagging a generation behind in terms of performance. A study conducted by researchers at Yale University estimated that if China could use the restricted chips, the resulting energy savings would be equivalent to the annual energy consumption of 12,000-67,000 American households. Furthermore, protectionism reduces the possibilities for algorithmic improvements, risking energy waste equivalent to the consumption of 1.8 million US homes, both in China and in the USA.
Although the consolidation of the Chinese generative AI sector is still a long way off, numerous companies are competing for limited computational resources. Some academic and corporate laboratories are, however, exploring more efficient alternatives, such as intelligence inspired by the functioning of the brain. Neuromorphic models, based on brain structures and less energy-intensive, represent a promising global perspective, and China is making significant progress in this field. However, in the context of fierce geopolitical competition to develop increasingly advanced models, it is unlikely that the country will radically revise its AI development strategy, which currently appears to be inefficient.
What Chinese AI could mean for Europe
Despite the difficulties, China remains the main competitor of the United States in the AI race. For policy makers, companies and civil society in Europe, ignoring the growing Chinese technological ecosystem is no longer an option, especially considering the progress towards cutting-edge AI systems. In defining its global technological strategy, Europe must address two key priorities.
Firstly, as highlighted by MERICS research, the ties between the European and Chinese AI ecosystems are deeper than is often thought, particularly thanks to research collaboration. However, China's centralized approach, its geopolitical objectives and its ambitions for leadership in the sector make it necessary to have a strategy based on risk assessment when collaborating with Chinese companies, universities and research institutions. US policies, which sometimes have extraterritorial effects, further complicate the balance between national security, ethical technological development and competitiveness. In this scenario, European governments must not only incentivize innovation in AI, but also protect local talent and strategic technologies from the influence of large American and Chinese companies.
Secondly, Europe must define a clear vision on how to relate to China in the global governance of AI. The Chinese government has introduced some of the most ambitious regulations in the world on artificial intelligence and is pursuing active diplomacy on two fronts: on the one hand, it presents itself as a leader of the developing world, and on the other, it collaborates with the West on issues related to the security and risks of AI.
Despite the differences, bilateral talks with the United States on these issues are ongoing. China has also signed the Bletchley Declaration, which emerged from the AI security summit hosted by the United Kingdom in 2023. With few exceptions, the European Union has so far shown little interest in overcoming political and value differences to better understand and, if necessary, collaborate with the Chinese approach to AI regulation.
The real threat is not China's technological advancement, but the US and European attempt to hinder global innovation for political reasons.