China's AI models are flooding the global market – and Europe must decide: play along or fall behind

Xpert Pre-Release

Language selection 📢

Published on: May 30, 2026 / Updated on: May 30, 2026 – Author: Konrad Wolfenstein

China's AI models are flooding the global market – and Europe must decide: play along or fall behind – Image: Xpert.Digital

DeepSeek, Qwen & Co.: How European companies use China's AI without risk

The dangerous AI temptation: Why German companies are now relying on Chinese technology

World power through open source: China's ingenious master plan for AI dominance

The tectonic plates of the global technology landscape are shifting at breakneck speed. For a long time, the USA, led by giants like OpenAI, Google, and Anthropic, was considered the undisputed pioneer of artificial intelligence. But this consensus is crumbling. With an unprecedented strategic offensive, China is currently flooding the global market with high-performance, freely available open-source models. Names like DeepSeek, Qwen, and MiniMax are no longer niche products, but serious competitors that massively undercut Western premium models in performance and, above all, in price. For European companies, from ambitious startups to established medium-sized businesses, this development exerts enormous economic attraction. But opting for cost-effective Chinese AI has its pitfalls: Anyone launching cross-border AI projects maneuvers their company into a highly complex field of tension between European data protection (GDPR), Chinese state control, and tangible geopolitical risks. The following article examines China's master plan to become an AI superpower and shows how European companies can operationally resolve the strategic dilemma between economic pragmatism and data policy sovereignty.

Technological advancement with aspirations for global power

China's rise to global AI superpower status is no longer a prediction, but a measurable fact. By 2025, Chinese companies had released 1,509 major language models – roughly 40 percent of all newly released AI models worldwide. Nine of the world's fourteen leading open-source models originated in China, while not a single US open-source model made the top 14. The underlying philosophy is remarkable: China strategically prioritizes openness. While Western providers like OpenAI rely on proprietary, paid models, Chinese labs like DeepSeek, Qwen, Kimi, and MiniMax are flooding the international developer community with freely available code.

The cost difference is not gradual, but structural. DeepSeek R1 was trained on 2,000 NVIDIA H800 GPUs for approximately $5.6 million – comparable Western models devour budgets of $80 to $100 million on significantly larger cluster infrastructures. API pricing follows the same logic: Qwen 2.5-Max costs only $0.38 per million tokens processed, while premium US models charge between $4.50 and $15. This cost advantage has real consequences: Western companies are already adopting Chinese models. Airbnb uses Alibaba's Qwen for its customer service bots, the code development tool Cursor employs Chinese models, and even Meta is reportedly using Qwen models to train its own AI, "Avocado.".

Infrastructure offensive behind the Great Wall

China's computing ambitions extend far beyond individual model releases. On December 3, 2025, China activated the world's largest distributed AI computing network: the Future Network Test Facility (FNTF), which spans over 2,000 kilometers, connects 40 cities via 55,000 kilometers of fiber optic cable, and, according to its operators, achieves 98 percent of the efficiency of a single data center. In Zhengzhou, China has launched a 30,000-chip computing center specifically for the next generation of physical AI—robots and autonomous systems. The national supercomputing network comprises more than 150,000 accelerator chips and over two million CPU cores, already accessed by more than one million users—researchers and businesses.

In parallel, Chinese industry is circumventing US export restrictions on state-of-the-art NVIDIA chips with pragmatic solutions: Alibaba, ByteDance, and other tech giants are leasing computing time in data centers in Singapore and Malaysia, operated by non-Chinese companies. This practice is perfectly legal following President Trump's repeal of the Biden-era "Diffusion Rule." Goldman Sachs predicts that Chinese internet companies alone will invest over $70 billion in data centers by 2026. These figures illustrate that China is not building an academic playground, but rather an industrial infrastructure designed for global scaling.

The State Council document as a blueprint for world power ambitions

On August 21, 2025, the Chinese State Council published the strategy document "Guofa No. 11"—the so-called "AI+" Action—a 14-point plan for the deep integration of AI into all areas of the economy and society. The goals are precise: By 2027, AI is to be deeply embedded in six core areas, with a penetration of AI agents and smart devices exceeding 70 percent. By 2030, the so-called "intelligent economy" is to become the main driver of growth, with a penetration rate exceeding 90 percent. The long-term goal for 2035 envisions a complete transition to an AI-permeated economy and society.

In parallel, on July 26, 2025, China presented a foreign policy counterpart, the “Action Plan on Global Governance of Artificial Intelligence,” which aims for inclusive, multilateral AI governance—with an explicit focus on supporting developing countries in building their own AI capacities. While Europe debates regulation and the US pursues a “Build, Baby, Build!” approach to deregulation, China is pursuing a two-pronged strategy: domestically, a massive claim to state control, and internationally, self-portrayal as a fair, inclusive partner of the Global South. This combination of strategic infrastructure investment, academic openness through open-source models, and diplomatic strategy makes China’s AI offensive a phenomenon whose complexity is unique in modern technological history.

Europe's strategic dilemma: cheap cooperation or expensive regulation?

For European companies, the rise of Chinese AI resources presents a compelling economic opportunity. The performance gap between top Chinese and US models has shrunk dramatically: while it was over 100 points in relevant benchmarks at the beginning of 2024, it had shrunk to around 20 points by the beginning of 2025. In specialized domains such as mathematics and programming, Chinese models now even outperform their US competitors. Added to this are the significant cost advantages: according to available data, Chinese providers achieve 90 percent of the performance of US models at training costs that are 82 percent lower.

This economic pull is virtually impossible for European SMEs and startups to ignore. Anyone developing an AI-powered product today faces a decision that wasn't even a question two years ago: Do I pay premium US prices for OpenAI or Anthropic, or do I use Chinese open-source models that I run on my own infrastructure? The answer to this question depends not only on technical criteria, but above all on one's own risk tolerance in the areas of data protection, geopolitical dependency, and regulatory compliance. Because this is precisely where the real complexity of cross-border AI projects involving China begins.

The dual legal system: When GDPR and PIPL collide

Cross-border AI projects between Europe and China operate within a legal gray area, defined by two sides. On the European side, the General Data Protection Regulation (GDPR) stipulates that personal data may only be transferred to third countries if an adequate level of data protection is guaranteed there – something that has not yet been confirmed for China by a formal adequacy decision from the EU Commission. On the Chinese side, the Personal Information Protection Law (PIPL) has been in effect since November 2021. While similar in its basic structure to the GDPR, it differs from it in key aspects.

The PIPL applies extraterritorially: European companies that process data of Chinese citizens also fall within its scope. Furthermore, it obliges data controllers to handle personal data according to the principles of purpose limitation, data minimization, and transparency. However, what structurally distinguishes the PIPL from the GDPR is its relationship to state actors: While the GDPR also applies to state bodies, Chinese authorities are largely exempt from the PIPL. This blind spot is not accidental, but rather inherent to the system: The Chinese intelligence law obliges all organizations and individuals to cooperate with the security authorities, which China observers largely interpret as a de facto right of access to all data stored in the People's Republic.

The DeepSeek case exemplifies these tensions. The German Federal Office for Information Security (BSI) considers DeepSeek's storage of keystroke patterns problematic, at least in security-critical areas, as this data can be used to create user profiles with the help of AI. Under Chinese law, DeepSeek is obligated to store all user data within the People's Republic. Several European countries, including Italy, Denmark, and the Czech Republic, have prohibited their authorities from using DeepSeek models on official devices. The German Federal Commissioner for Data Protection, Louisa Specht-Riemenschneider, demanded that DeepSeek be removed from app stores for violating European law, while several German data protection authorities have launched investigations.

Operational architecture of cross-border AI projects

Despite these regulatory and security policy tensions, the practice is more nuanced than a simple prohibition or approval dictate. European companies seeking to utilize Chinese AI resources for cross-border projects have several operating models to choose from, representing varying compromises between performance, cost savings, and risk exposure.

The safest model for European companies is so-called on-premise deployment: Chinese open-source models such as DeepSeek-V3, Qwen, or MiniMax are operated on the company's own servers within the EU. In this case, no user data leaves the European infrastructure, thus ensuring both GDPR compliance and circumvention of the Chinese intelligence law. This approach has already proven practical for technically proficient companies: Over 180,000 derivative models have been created based on Alibaba's Qwen alone, a significant portion of which run on European infrastructure. The second model—using Chinese cloud APIs directly from Europe—is legally risky as long as there is no standard contractual clauses framework or comparable safeguard, since transferring personal data to a country without an adequacy decision constitutes a GDPR violation.

This results in a clear operational logic for international AI project management: European project managers assume responsibility for data classification, compliance architecture, and the operation of the more production-oriented systems on European infrastructure. Chinese data engineering teams can be responsible for model optimization, fine-tuning, and benchmarking—as long as no sensitive real-world data flows into China, but only anonymized training data or synthetic datasets. This form of division of labor is not only more legally robust but also economically rational: Chinese AI engineers, especially specialized data engineering teams, offer a very attractive price-performance ratio compared to their international counterparts.

A new dimension of digital transformation with 'Managed AI' (Artificial Intelligence) - Platform & B2B solution | Xpert Consulting

A new dimension of digital transformation with 'Managed AI' (Artificial Intelligence) – Platform & B2B solution | Xpert Consulting - Image: Xpert.Digital

Here you will learn how your company can implement customized AI solutions quickly, securely and without high entry barriers.

A managed AI platform is your all-inclusive, worry-free solution for artificial intelligence. Instead of dealing with complex technology, expensive infrastructure, and lengthy development processes, you receive a ready-made solution tailored to your needs from a specialized partner – often within just a few days.

The key advantages at a glance:

⚡ Rapid implementation: From idea to ready-to-use application in days, not months. We deliver practical solutions that create immediate added value.

🔒 Maximum data security: Your sensitive data stays with you. We guarantee secure and compliant processing without sharing data with third parties.

💸 No financial risk: You only pay for results. High upfront investments in hardware, software, or personnel are completely eliminated.

🎯 Focus on your core business: Concentrate on what you do best. We take care of the entire technical implementation, operation, and maintenance of your AI solution.

📈 Future-proof & scalable: Your AI grows with you. We ensure continuous optimization and scalability, and flexibly adapt the models to new requirements.

More information here:

The Managed AI Solution - Industrial AI Services: The Key to Competitiveness in the Services, Industry and Mechanical Engineering Sectors

Contractual pitfalls and patent protection: Practical tips for AI cooperation with China

IP protection as a critical bottleneck in any cooperation

Besides data sovereignty, the protection of intellectual property is the second strategic weakness of cross-border AI cooperation with China. In no other area of technology cooperation is the discrepancy between the formal legal framework and operational reality greater. For years, China has had a sophisticated patent and copyright system that, on paper, meets international standards. In practice, however, access to legal remedies for foreign companies in cases of IP infringement remains complex, time-consuming, and fraught with considerable risk.

With 1,576,000 AI patents, China holds a 38.6 percent share of the global market – a figure that reflects both the high level of innovation and the strategic importance of IP protection in the Chinese AI landscape. For European companies conducting AI projects with Chinese teams, this leads to a clear recommendation from experts: all proprietary algorithms, trained model weights, and architectures must be fully documented before the project begins, secured through international patent applications, and protected by contractual clauses regarding confidentiality and transfer of ownership. Particular attention should be paid to the handling of training infrastructures: anyone training or fine-tuning proprietary data or models on Chinese servers risks, without contractual protection, effectively disclosing training insights to third parties.

Experienced consultants for the Chinese market further recommend structuring AI development contracts according to internationally recognized standards, with explicit clauses regarding ownership rights to trained models, the allocation of improvement rights, and the handling of derivative works. The so-called "work-for-hire" principle, which applies under US law and automatically makes the client the owner of the commissioned work (similarly regulated in German copyright law regarding usage rights), is not mandatory in this form under Chinese law. Without explicit regulations, gray areas can arise in which Chinese contractors could assert claims to developed model components.

The EU AI Act as a global regulatory paradigm

While China and the US are focusing their AI strategies on growth and market penetration, the European Union has enacted the world's first comprehensive AI regulatory framework with the AI Act. The regulation entered into force on August 2, 2024, and is being phased in: since February 2, 2025, bans have been in place for AI systems posing unacceptable risks. Governance rules and additional obligations for providers of general-purpose AI systems came into effect on August 2, 2025. Mandatory compliance for high-risk AI systems will follow on August 2, 2026, with full implementation planned for 2027.

The AI Act applies extraterritorially to all AI systems placed on the market in the EU or whose use affects EU citizens – regardless of where the provider is based. This means that Chinese AI providers seeking to serve European customers must meet the same transparency, documentation, and compliance obligations as US or European providers. New models will be reviewed by the EU AI Office starting in 2026, and existing models starting in 2027. Providers that violate the rules risk fines of up to €35 million or 7 percent of their global annual turnover.

For the cooperation model between European companies and Chinese AI teams, this has a direct consequence: The European project management, as the "operator" within the meaning of the AI Act, bears responsibility for the regulatory compliance of the AI systems used – regardless of whether the underlying models originate from China, the USA, or Europe. This allocation of responsibility makes a careful risk classification of each AI module used an indispensable step in project design. Particularly in applications in the high-risk areas defined by the AI Act – such as human resources, lending, or medical diagnostics – the entire AI value creation process must be fully documented and safeguarded by human oversight mechanisms.

Geopolitical asymmetries and strategic dependencies

The economic appeal of Chinese AI resources cannot be separated from their geopolitical context. China pursues its AI strategy as an integral part of its state industrial policy and national security strategy. The State Council not only controls and subsidizes model development but, through the 2017 National Intelligence Law, has also created the legal framework within which private companies are obligated to cooperate with intelligence services. This situation is not directly comparable to that of Western cloud providers: While the US Cloud Act also grants government access to data stored abroad by US companies, it is subject to judicial review and diplomatic agreements that channel data access.

Twelve of the fifteen leading open-source AI models now originate from China. This finding has two opposing implications. On the one hand, China's open-source strategy democratizes global access to powerful AI models and reduces dependence on US providers who secure their monopoly through high prices and restrictive terms of use. On the other hand, a structural dependence on Chinese base models—even when run on-premises—carries the risk that embedded preferences, training data biases, or politically motivated content restrictions will unwittingly permeate European applications. The question of whether Chinese models have deliberate blind spots for certain topics—Taiwan, Tibet, Tiananmen Square—is well-documented empirically and poses a real quality risk for companies in certain use cases.

Furthermore, there is the risk of technological path dependency: Anyone building their development infrastructure on a Chinese base model invests in customizations, fine-tuning, and integration interfaces that are completely lost when migrating to a different vendor. While this lock-in risk is lower with open-source models than with proprietary APIs, it is not entirely eliminated—especially when proprietary extensions or specific model architectures are used that do not guarantee full portability.

Operational success factors for international AI project teams

Cross-border AI projects involving China rarely fail due to technical shortcomings, but rather due to structural coordination problems resulting from differing working methods, communication norms, and institutional frameworks. Experience from German-Chinese technology projects repeatedly demonstrates that intercultural competence and a clearly defined escalation protocol are often more important than the purely technical excellence of the participating teams.

Several principles have proven effective in practice for the collaboration between European project managers and Chinese data engineering teams. First, the data strategy should be fully defined before the project begins: What data leaves the EU and under what conditions? What classification schemes apply? What anonymization and pseudonymization standards are used? Second, the compliance architecture requires continuous, shared responsibility: The European side is responsible for GDPR and AI Act compliance, while the Chinese side is responsible for PIPL compliance when processing data of Chinese citizens or companies. Third, IP ownership structures must be clearly defined in a contract before even a single line of code is written jointly.

Furthermore, the technical infrastructure should be designed in such a way that it safeguards the principle of data sovereignty through architectural decisions, not just through contractual promises. Hybrid deployment models – in which sensitive processing stages are mandatory on European servers, while computationally intensive, non-personal training tasks may be performed on international or Chinese infrastructures – offer a practical middle ground between economic efficiency and legal compliance.

Europe's AI sovereignty strategy as a counterweight

The European Union has recognized the challenge and is responding with its own investment initiative. The “AI Continent Action Plan” focuses on five strategic pillars: expanding computing infrastructure, including planned AI gigafactories with investments of up to €20 billion; improving data access; targeted AI skills development; developing trustworthy algorithms; and simplifying regulatory processes. The flagship initiative GenAI4EU provides almost €700 million for the development and deployment of generative AI in strategic European sectors.

In parallel, German industrial companies are investing in their own local AI infrastructures. Bosch, Trumpf, and Siemens are working on proprietary AI solutions that aim for independence from both US cloud giants and Chinese models. This trend toward sovereign AI infrastructure, however, does not contradict the use of Chinese open-source models as core components—rather, it defines the conditions under which such use is responsible: local hosting, full model control, GDPR-compliant data processing, and transparent documentation for regulatory authorities.

The real question for Europe is not whether to use Chinese AI models – from an economic perspective, this is almost unavoidable if one wants to remain competitive. The crucial question is how this use can be structured in such a way that Europe does not relinquish either technological sovereignty or control over data policy. Cross-border AI projects under European leadership, which treat Chinese development capacities as a resource and not as a strategic dependency, are not a contradiction – they are the most complex, but also the most realistic form of a European AI strategy in the age of global technological competition.

Six areas of strategic decision-making

European companies entering into cross-border AI projects with China must actively shape six strategic decision areas that cannot be separated: data sovereignty through architecture rather than solely through contracts; compliance duality in the tension between GDPR and PIPL; IP protection before project start through international patenting and precise ownership clauses; AI Act compliance as operator even with externally developed models; geopolitical risk management through continuous monitoring of regulatory and political developments; and intercultural project management that productively integrates different work and communication cultures instead of ignoring them.

China's AI offensive is real; it is well-funded, technologically competitive, and strategically driven. European companies that ignore these resources are forfeiting economic potential. Those, however, that use them uncritically and without a structured governance architecture risk data sovereignty, trade secrets, and regulatory compliance. The truth—as is so often the case with the most pressing economic policy issues—lies not in a binary decision, but in the quality of managing the inevitable complexity.

🎯🎯🎯 Data-driven B2B industry hub as a quasi-in-house solution

The quasi-in-house solution: How Xpert.Digital closes operational gaps in B2B marketing and sales – Smart Content-Driven Business - Image: Xpert.Digital

Xpert.Digital is a data-driven B2B industry hub led by Konrad Wolfenstein . The company acts as an external, quasi-in-house solution for industrial partners, closing operational gaps in marketing, content, and sales – without requiring additional resources on the client side.

More information here: