Beyond the Race: China’s AI and the Choices Facing Europe
Green European JournalAI development is often framed as a two-player contest between the US and China over who can develop “superintelligence” first. Yet China’s moves suggest Beijing is focusing primarily on diffusion and global influence. Can Europe navigate this rivalry and actively engage with China while reducing its technological dependence on foreign superpowers?
AI development is often framed as a two-player contest between the US and China over who can develop “superintelligence” first. Yet China’s moves suggest Beijing is focusing primarily on diffusion and global influence. Can Europe navigate this rivalry and actively engage with China while reducing its technological dependence on foreign superpowers?
This article will feature in the Green European Journal’s winter 2026 print edition Far West: Europe in the “Chinese Century”. Subscribe now and be among the first to receive a copy in December.
In recent years, China has firmly established itself as the leading global AI superpower alongside the US. Chinese open-weight models developed by companies including Alibaba, DeepSeek, Moonshot AI, and others are among the most downloaded in the world. Chinese AI labs are also quickly catching up with the leading US labs: according to some estimates, Chinese frontier capabilities are only three to six months behind the US, with the gap closing. In 2024, China also accounted for the largest share of AI patents granted worldwide,1 at 74.2 per cent, compared with 12.1 per cent for the US and 3 per cent for Europe.
These developments have sparked debate in the West, particularly in the US, about the trajectory and goals of China’s AI policy. One of the dominant narratives is that China is locked in an “AI race” with the US to develop so-called artificial general intelligence (AGI) or superintelligence first, and that whoever gets there first will command unprecedented power in the international order. For the US, losing this race could mean a decline in global influence and even the inability to protect its political and economic system.

Playing up the threat of China’s technological catch-up in frontier AI has enabled Silicon Valley to push back against regulation domestically. It has also prevented an informed and grounded debate on China’s AI strategy. But what if China is not pursuing the race towards AGI in quite the same way as the US? What if China is indeed in a race, but running on a different track, with the finish line further south?
The answer to these questions has implications for Europe. If we misunderstand China’s path and ambitions, we risk overlooking important developments in the global AI landscape and directing our policies towards the wrong priorities.
Building an ecosystem
A closer look at the Chinese Communist Party (CCP)’s actions and policies suggests that China is pursuing a strategy focused on building an AI ecosystem. It aims to create the economic, social, and political conditions for the widespread diffusion of Chinese AI, both domestically and internationally. To reach this goal, China doesn’t need to lead every technological breakthrough or be the first to develop the most advanced AI systems; rather, it needs to maximise technological diffusion and integration.
As political scientist Jeffrey Ding argues in his book Technology and the Rise of Great Powers,2 the country that benefits the most from a general-purpose technology is not the one that invents it, but the one that spreads and integrates it most widely across its economy. Ding links the rise of the British Empire to the diffusion of the steam engine, and the United States’ later overtaking of Britain as the dominant global power to its widespread adoption of electricity, which transformed production processes and strengthened its economy. China’s policies appear to regard AI as a general-purpose technology, and to follow much the same underlying logic.
Four pillars are central to China’s strategy: promoting the domestic diffusion of AI across traditional sectors; strengthening technological self-reliance along the AI value chain; betting on open-weight models for competitiveness and international diffusion; and expanding international cooperation on AI capacity-building and governance, particularly with countries in the Global South.
The government aims to integrate AI into as many sectors of China’s economy as possible
Diffusion over superintelligence
China’s focus on AI adoption does not mean that Chinese frontier labs don’t take AGI seriously or have no ambitions in that direction. DeepSeek founder Liang Wenfeng, for example, has explicitly stated that achieving AGI remains a key goal for the company. Moonshot AI founder Yang Zhilin has expressed similar ambitions. However, China’s political approach has been directed towards other priorities. AGI and superintelligence do not feature particularly prominently in the CCP’s vocabulary. In the 15th Five-Year Plan (2026-2030), for example, AGI is explicitly mentioned only once, as something to be explored alongside other fields of AI.
Instead, the government aims to integrate AI into as many sectors of China’s economy as possible, from manufacturing to agriculture, logistics, healthcare, education, public administration, and urban management. More recently, CCP policies have also placed greater emphasis on embodied intelligence and robotics, AI agents, and integrating AI into everyday products, from smartphones and computers to cars and household appliances. These objectives are being pursued under the AI Plus initiative, first introduced in 2024 and further developed in 2025, and again taken up in the latest Five-Year Plan.
There are signs that this strategy is paying off, at least in part. At this year’s World Artificial Intelligence Conference (WAIC) in Shanghai, more than 1100 companies showcased a vast range of ways AI technologies are being integrated across sectors. The conference is China’s flagship AI event and offers good insight into general developments in its AI economy.
Trivium China, a research and strategic advisory firm, also recently pointed out that while large parts of the Chinese economy are struggling, technology-related industries, including AI and semiconductors, have been performing exceptionally well.3 Of course, not all of this success can be attributed to government policies. Much of it is driven by private companies and entrepreneurs. Still, it suggests that China has created a policy environment in which AI diffusion can thrive.
Self-sufficiency across the stack
Chinese companies have faced increasing difficulties accessing foreign technologies essential to developing frontier AI. In October 2022, the Biden administration introduced export controls on advanced AI chips, including Nvidia’s A100 and H100, which are widely used to train frontier AI models. Restrictions were later extended to the equipment and software needed to manufacture advanced semiconductors. Although some restrictions have since been eased following Chinese retaliation and negotiations between Washington and Beijing, dependence on foreign technologies remains a strategic vulnerability for China.
Pursuing self-sufficiency across the AI stack, including rare earths, energy, chips, compute, data, models, and applications has therefore been a major strategic priority. This is not simply a response to US restrictions: it also reflects China’s broader ambition to strengthen its domestic economy and develop its own technological capabilities. At a Politburo study session in April 2025, Xi Jinping stressed the importance of technological self-reliance in AI, emphasising breakthroughs in advanced chips and foundational software, as well as the development of an independently controllable AI hardware and software system.4
Despite considerable progress, manufacturing the most advanced semiconductors at scale remains a major challenge for China.
A 2025 report by the Mercator Institute for China Studies shows that this ambition for self-reliance translates into different strategies and levels of government intervention depending on which layer of the AI stack it targets.5 Interestingly, some of the highest levels of innovation and development are happening in areas where the government intervenes the least. This is particularly the case in AI models and applications, where direct government intervention is comparatively limited. Chinese companies like DeepSeek, Moonshot AI, Zhipu AI, Tencent, and Alibaba have developed models that are now competing with those of leading US frontier labs. Here, the government’s role has largely been to support the wider ecosystem and help companies scale up through procurement or access to computing power.
In computing infrastructure, meanwhile, the government is investing heavily in the construction of data centres. Under its East Data, West Computing initiative, Chinese local governments, often in collaboration with private companies, are building large data centre clusters in the country’s western regions, where land and electricity are cheaper, to provide computing power for the AI industry (much of which is concentrated in eastern China).
In semiconductors, the picture looks very different. Here, China still faces some of its biggest technological bottlenecks, and government intervention has been extensive. The state-backed National Integrated Circuit Industry Investment Fund, known as the Big Fund, is a major source of financing for China’s domestic semiconductor industry. Its third phase was established in 2024 with a registered capital of 344 billion yuan (around 47.5 billion US dollars).
Companies such as Huawei and SMIC play a central role in developing domestic alternatives to foreign chips. But despite considerable progress, manufacturing the most advanced semiconductors at scale remains a major challenge for China.
Open weights to speed up diffusion
Unlike many leading US labs, most major Chinese frontier AI companies pursue an open-weight approach, which consists in making the underlying parameters of their AI models publicly available, allowing other developers to download, use, and adapt them to their own needs. However, the training data and the full development process are often not made publicly available, as is the case with open-source models. Chinese companies like DeepSeek and Alibaba also provide their models on internationally popular platforms such as Hugging Face, making them available to developers around the world.
This approach has two main benefits. First, open-weight (and occasionally open-source) models allow Chinese companies to tap into the international developer community, speeding up development and significantly reducing the costs of building and adapting AI models. According to a 2026 analysis by Hugging Face, Alibaba’s Qwen models have become one of the largest foundations in the open model ecosystem, recording more than two billion downloads in the first seven months of 2026. Other popular Chinese models also boast millions of downloads.6
Developers across the world now experiment with these models, adapt them for different applications, improve their efficiency, and identify or fix safety problems. In the case of Qwen alone, more than 151,000 derivative models had been uploaded to Hugging Face by August 2026. Many developers make their adaptations and improvements freely available, allowing others, including the original Chinese companies, to benefit from their work without hiring them directly. This gives Chinese companies a significant competitive advantage over American companies that keep their models closed.
The second key benefit is the international diffusion of Chinese AI and the development of applications built on its model ecosystem. Especially in the Global South, the costs of accessing and adapting AI models can be a major barrier. Chinese open-weight models can be downloaded, adapted to local languages and needs, and operated on infrastructure outside China. This flexibility makes them a more attractive, cheaper alternative to closed US models.
Even though much of this open-weight and open-source push has been driven by the private sector, the Chinese government actively supports it and has made it a priority in several policy documents. In the AI Plus initiative, for instance, the State Council explicitly calls for supporting open-source AI communities, making models, tools, and datasets more widely available, and developing globally accessible open ecosystems. The same document also calls for helping countries in the Global South develop their own AI capabilities. The 15th Five-Year Plan similarly encourages strengthening open-source systems.
Cooperation with the Global South
China’s ecosystem-building is not limited to the domestic level or the international diffusion of its open-weight models. The Chinese government is also actively building ties with governments worldwide, and particularly in the Global South, to help countries access, use, and build on Chinese AI technologies, and to collaborate on international AI governance.
The United Nations has been an important channel for this. In 2024, China initiated a UN resolution on international AI capacity-building, which was later adopted by consensus. At the UN’s first Global Dialogue on AI Governance in July 2026, China sent a ministerial-level delegation and actively promoted international cooperation on open-source AI, capacity-building, and governance.
The World Artificial Intelligence Cooperation Organization (WAICO) has been another important forum in this regard. Initiated by China, it was founded in July 2026 with 29 signatory countries, mostly from the Global South. The organisation has two main goals: first, capacity-building to improve access to AI technologies and services through policy exchange, technical cooperation, training, and joint research. Second, connecting members’ respective strengths and improving technology supply, furthering the joint construction of open-source ecosystems, and helping advance AI Plus initiatives in different countries.
Chinese diplomats are also using these forums to position China as an active defender of multilateralism, particularly as the current US administration is sceptical of, and actively undermines, attempts at international AI governance. This could become an important source of soft power, making it easier for Chinese AI technologies to spread internationally while strengthening the country’s position and influence in shaping how AI is governed around the world.
What does this mean for Europe?
China already is, and increasingly will be, a key AI player that Europe cannot ignore. Many European companies and developers are building on Chinese AI models, and they are likely to use them even more as these models become more competitive. For Europe, which is trying to become less dependent on American technologies and diversify its options, this trend could also open up new strategic opportunities.
One argument that often comes up in European debates is that Chinese technologies should not be used because they come from an authoritarian country, while US technologies are considered more democratic and therefore safer. The problem with this framing is that it doesn’t ask the most important questions that determine whether a technology is safe to use: is the model open or closed? Can we download it, run it on European infrastructure, and adapt it to our needs? Is it secure? Do we retain control over our data? Can we independently test the model and understand how it behaves?
When Chinese AI meets these conditions, Europe should not be so quick to dismiss it. Open-weight models, no matter where they are from, can offer European companies greater control over their systems than some of their closed alternatives.
Of course, these models still need to be used with caution. China has very different ideas about freedom of information from Europe, so models need to be tested, fine-tuned, and adapted for local use. They also need to be checked for security vulnerabilities. In October 2026, the Canadian AI company Cohere reported that an Alibaba Qwen model had unexpectedly attempted to send a probe file to an Alibaba Cloud server in Beijing during a security test.7 The attempt was blocked, and there is no established evidence that it was deliberate. Still, the incident shows why testing is essential. This is where European AI safety institutes could play an important role. Governments should help set them up and ensure they are equipped with the necessary resources and technical expertise to independently evaluate AI systems, identify potential risks, and help companies adapt them for safe use. This capacity would also be an important step towards European tech sovereignty.
Moreover, Europe should not build an AI strategy that assumes a continued supply of Chinese open-weight models. China is currently discussing a proposal by the Ministry of Commerce and the Ministry of Science and Technology to potentially introduce export controls on certain advanced AI technologies. While China is unlikely to completely change course on open-weight models, the proposal shows the risks of overreliance on Chinese AI.
Europe should therefore build on the options that currently exist, but it also needs to create the economic conditions for European AI companies to emerge and compete, so that it can rely on its own models in the future.
Open-weight models, no matter where they are from, can offer European companies greater control than some of their closed alternatives
Keeping options open also means that Europe must resist pressure from the US to become locked into its technological ecosystem. We cannot know who will lead in AI in 10 years’ time, and betting too heavily on continued US technological leadership would be risky. While Europe should seek to maintain good relations and preserve its partnerships with the US, it must also have credible alternatives that allow it to walk away from the negotiating table when necessary. Preserving that freedom of choice would give us more room to manoeuvre and greater control over our technological future.
China’s approach to AI diffusion also has important implications. Europe has already begun pursuing a similar approach through its Apply AI Strategy, but many traditional industries are struggling to integrate AI. They should pay closer attention to how their Chinese counterparts are using these technologies beyond frontier AI labs, and what can be learned from their experiences. If Chinese companies succeed in using AI to significantly improve productivity or develop whole ranges of new products, this could create another wave of competitive pressure on European industries. Europe needs to anticipate and prepare for this possibility.
Lastly, Europe should question the AI race rhetoric pushed by Silicon Valley and the Trump administration, and design policies that respond to what China is actually doing. Because worse than falling even further behind is spending billions trying to win a race we have fundamentally misunderstood.