Is a digital blockade coming? The end of free AI? What if China now turns off the AI ​​tap for the West?

Double dependency: Why Europe's AI strategy could now fail spectacularly – China's radical paradigm shift is endangering Europe's economy

For months, Western startups and large corporations profited from open and, above all, extremely affordable AI models from China. But this era of free access appears to be coming to an abrupt end. As new insider reports reveal, Beijing is planning far-reaching export and access restrictions for its cutting-edge AI – a direct response to Washington's increasing technological isolation. For the global technological order, this looming "Silicon Curtain" marks a historic turning point: Artificial intelligence is no longer viewed by the superpowers as a global commodity, but rather as a critical national resource and a potential vector for espionage. This double isolation could have particularly disastrous economic consequences for Europe, which finds itself caught between the fronts of the American-Chinese technology war without its own comparable models. The following article examines the background of the new Chinese strategy and shows why the free fall of the digital Iron Curtain heralds the end of globalized AI development.

Related to this:

• Jiu-Jitsu instead of boxing: Learning to win from the best – What Europe and Germany should learn from Apple's AI strategy

When the free pass for Chinese AI ends – and the West suddenly finds itself alone

Just a few years ago, the open, cost-oriented strategy of Chinese AI models was considered a kind of quiet revolution against the expensive, closed US ecosystem. Chinese providers like Alibaba's Qwen or ByteDance's Doubao readily made their models available to the world – either as open-weight models for free download or via inexpensive API access, whose prices were sometimes 10 to 50 times lower than comparable US offerings. Around 80 percent of American AI startups relied on cost-effective models from China. But this status quo could soon be a thing of the past.

As Reuters reported in July 2026, citing three people familiar with the talks, Chinese authorities began consultations last month with leading technology companies to discuss potential restrictions on foreign access to China's most advanced AI models. Representatives from Alibaba, ByteDance, and the AI ​​startup Zhipu AI participated in the talks, which were led by the Ministry of Commerce. What is emerging is a fundamental shift in Beijing's AI strategy: away from global proliferation and toward strategic custody.

Technological parity as the trigger for the paradigm shift

The central reason for this change in strategy is technological: Chinese AI models have dramatically narrowed the gap with their American competitors. According to the Stanford HAI AI Index 2026, the performance advantage between the leading US model, Anthropic's Claude Opus 4.6, and the best Chinese model, Dola-Seed 2.0, shrank to just 2.7 percentage points. While the US still boasts over 50 top-performing models compared to 30 Chinese ones, the qualitative gap is closing at a breathtaking pace.

Particularly symptomatic of this development is the GLM-5.2 model from the AI ​​startup Z.ai (formerly Zhipu AI), which can compete with leading US offerings like Anthropic's Mythos at a fraction of the cost. And in terms of global download figures, Chinese AI models have already overtaken their US competitors: In 2025, they achieved a global download share of 17.1 percent compared to 15.8 percent for US models. According to data from spring 2026, on the OpenRouter platform, Chinese providers account for around 61 percent of total token consumption among the ten most-used models.

It is precisely this parity that is changing the dynamics. As long as Chinese models lagged significantly behind the West, their widespread adoption was strategically advantageous – they brought reach, user data, and global influence without posing geopolitically significant risks. Now, however, as these models operate at the technological frontier, their status is changing: what was previously a marketing tool is becoming a national resource.

A reflection of Washington: The principle of mutual isolation

Beijing's considerations cannot be viewed in isolation. They are a direct response to a strategy that Washington has pursued with increasing determination in recent years. The US has gradually built an export control apparatus that not only covers hardware supply chains but now increasingly includes the model layer of AI.

On June 12, 2026, the US Department of Commerce and the Bureau of Industry and Security (BIS) mandated that foreign users would require a license to access Anthropic's top-of-the-line models—specifically, the Fable and Mythos models. The measure took effect with only 90 minutes' notice and drew sharp criticism from G7 partners. Security vulnerabilities and concerns about potential cyber espionage by China were cited as justification. Export controls for Fable, the model designed for the general public, were later lifted after new security measures were implemented.

In parallel, in May 2026, the US government implemented new guidelines intended to close a loophole in chip export controls: Chinese companies with foreign subsidiaries had previously been able to access state-of-the-art processors like Nvidia's Blackwell and Rubin chips through these structures – a situation the Trump administration aimed to effectively prevent with new licensing requirements. The US and China are thus pursuing structurally similar strategies: Both sides are trying to keep the other away from critical technological resources.

The paradox is that US export controls on chips ultimately spurred innovation in China. Companies like DeepSeek were driven by the enforced resource scarcity to develop unusually efficient training methods – a boomerang effect that fundamentally challenges the underlying assumptions of the US strategy.

Related to this:

• US corporate alert “Digital Kill Switch”: Framework for action for companies – Seven concrete steps towards AI resilience

The security narrative as a driving force for both sides

A central motivation behind Chinese considerations is the perception that advanced AI models could be used as counterintelligence tools. According to two Reuters sources, Chinese authorities are deeply concerned that Anthropic's Mythos model could identify software vulnerabilities in Chinese systems and that the US government could use the model against Chinese interests. This concern led Alibaba to completely ban Anthropic's Claude model from its internal systems—Alibaba employees are no longer permitted to use Claude.

On the other hand, Anthropic is leveling serious accusations against Alibaba. In a letter to US senators, the US company accuses the Chinese corporation of illegally copying functions of its AI software – through so-called model distillation, in which a smaller model imitates the behavior of a larger one and thus adopts its capabilities. Anthropic is calling on the US Congress to take stricter action against Chinese companies that resort to this practice.

The security narratives of both sides reveal the fundamental logic of digital technology warfare: the same tool—a language model—that serves a company as a productivity instrument is classified as a potential attack vector from a national security perspective. That powerful models can identify vulnerabilities in software systems is undisputed. The question is whose vulnerabilities they find—and on whose behalf they do so.

The arsenal of control: measures beyond the AI ​​layer

What appears in the Reuters reports as a possible restriction on access to AI models is merely the latest escalation of a much broader control strategy that Beijing has been gradually expanding this year. This strategy encompasses several levels.

In an effort to protect AI companies and capital flows, China's National Development and Reform Commission (NDRC) blocked Meta's already completed acquisition of the AI ​​startup Manus for two billion US dollars at the end of April 2026 and ordered the deal to be reversed. Particularly noteworthy is that Manus had already relocated its headquarters to Singapore by this time. Nevertheless, Beijing asserted its jurisdiction and enforced the veto – a precedent demonstrating the extraterritorial reach of Chinese regulation. Legal experts have since referred to China's national security clearance as a "regular closing condition for cross-border tech deals.".

To control human resources, Beijing imposed travel restrictions on leading AI specialists at companies like Alibaba and DeepSeek at the end of May 2026. Experts on strategically important AI projects now require government approval before they can travel abroad—a measure with historical parallels to controls on nuclear scientists. The NDRC also barred Manus founders Xiao Hong and Ji Yichao from leaving the country while authorities reviewed the deal.

At the level of regulatory architecture, Beijing issued sweeping new regulations in early June 2026 to control foreign business dealings involving Chinese investors, technologies, and data. In addition, authorities are investigating Chinese AI startups that have relocated abroad for potential violations of export control laws. Following the model of rare earths—for which China introduced an export licensing requirement in October 2025, which was even extended to products manufactured abroad if they contain at least 0.1 percent Chinese rare earths—a similar control regime is now emerging for AI technologies.

The tiered model: Differentiated isolation instead of a blanket ban

A key element of these considerations is a multi-tiered regulatory model presented by Chinese legal experts at a panel discussion held in May, a summary of which was published in an official journal of the Supreme People's Court. The proposed system stipulates that simple open-source AI models should be subject only to a straightforward reporting requirement, advanced models should undergo a security review, and top-tier models should either not be made publicly available at all or restricted to the domestic market.

This differentiation is pragmatic from a regulatory perspective. A blanket export ban on all Chinese AI models would also entail commercially pointless restrictions – and raise significant questions about how control could be technically enforced for open-weighted models, which, by definition, are copyable once published. Reuters explicitly points out that it is still unclear how exactly access restrictions for open-source models could be implemented. Apparently, the focus is therefore on models yet to be published: Two of the sources confirm that potential restrictions might only apply to future models.

The proposed tiered model follows a similar logic to the export control system for arms: mass-produced goods remain accessible, while strategic cutting-edge technology is subject to strict control. The parallel is not coincidental – in Beijing, as Reuters notes, AI is now considered, much like in Washington, a critical national asset requiring state control.

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:

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

Economic consequences: Who bears the costs of isolation?

The economic consequences of a significant restriction on access to Chinese AI models would be substantial – and would not affect all parties equally. For international companies that have built their infrastructure on cost-effective Chinese models, the loss of this option would mean a return to significantly more expensive US alternatives. Reuters explicitly notes that this would likely result in increased costs for many companies.

The global market for generative AI was estimated at around US$103.58 billion in 2025 and is projected to grow to US$467 billion by 2030. In this rapidly expanding market, Chinese models have established themselves as key cost-cutting drivers. European companies, in particular, have benefited from being able to choose between regulated US models and the more affordable Chinese alternatives. This choice could disappear.

For China itself, the strategy poses short-term economic risks. The global spread of Chinese AI models has not only generated geopolitical influence but also commercial revenue, user data, and technological feedback. Companies like Alibaba and ByteDance have viewed international users as a valuable resource for the further development of their models. An abrupt withdrawal could disrupt these feedback loops. However, the logic of the control strategy apparently prioritizes other factors: the long-term preservation of technological superiority—and the prevention of strategic technology outflow—is valued more highly than short-term revenue maximization.

For the Chinese economy as a whole, another area of ​​tension arises: travel restrictions for top AI talent reduce China's attractiveness as a location for international AI collaborations. Critics compare the measures to Soviet practices of controlling scientists and warn of a brain drain due to deterrent effects. Indeed, the Stanford AI Index 2026 shows that the number of AI researchers emigrating to the US has plummeted by 89 percent since 2017 – but the trend has since reversed, with more specialists returning to China or leaving the US.

Related to this:

• Hidden billion-dollar potential: AI development in Romania, Hungary, Greece and Turkey

Europe caught between fronts: Dual dependency as a structural risk

For Europe, current developments reveal a structural failure of its own strategy. The continent had implicitly settled into a comfortable position: on the one hand, access to high-performance US models, and on the other, cost-effective Chinese alternatives as leverage and substitutes. Now both doors are closing simultaneously – or at least this trend is becoming apparent.

The US export control measures imposed on Anthropic's models in June 2026 severely impacted European users and were sharply criticized by G7 partners. In response, Vienna proposed an EU partnership with Anthropic to circumvent US export restrictions. Should Chinese models now be similarly restricted for European users, Europe would find itself in a double bind: without its own frontier models, without reliable access to cutting-edge US technology, and without a cost-effective Chinese alternative.

The costs of such strategic isolation would be substantial. According to a KPMG study, the EU's plans alone to push Chinese technology out of critical sectors could cost €367.8 billion by 2030 – €170.8 billion of which would be attributable to Germany alone. While this figure is disputed, as the study was commissioned by the Chinese Chamber of Commerce to the EU, it does at least indicate the magnitude of the structural dependencies that have developed over the years.

The strategic conclusion is clear: Europe must massively accelerate the development of its own AI capabilities. Digital sovereignty will remain mere lip service as long as the technological foundation for it is lacking.

The paradox of the open-source world: Control over the uncontrollable?

The planned Chinese regulation is particularly complicated in the area of ​​open-weight models. Unlike proprietary closed-source systems, whose use can be controlled via API access, open-weight models are inherently uncontrollable: once published, they can be downloaded, copied, and operated without any connection to their original developer. This applies to Alibaba's entire Qwen family, of which, according to the company, over 400 model versions have been published and which have spawned more than 180,000 derivative versions.

The proposed phased model only solves this problem for the future: Models published so far remain available worldwide. Restrictions can only meaningfully apply to future releases. However, the difference between Chinese open-weight models and classic open-source projects lies in a subtle dimension of control: While purely open-source projects manage without central control, companies like Alibaba retain the ability to distribute weights only to specific user groups and to selectively grant API access. Technically, geo-based access control for API services could be implemented relatively easily – even if it would have no effect on model weights already distributed.

According to Hugging Face's Spring 2026 Report, 41 percent of all new model uploads on the platform come from Chinese developers. Limiting this presence without harming its own ecosystem is a regulatory task that presents significant technical and legal challenges. By early February 2026, major platforms like Doubao and Qwen had collectively produced over 150 billion pieces of AI-generated content—figures that illustrate the massive scale of the Chinese AI ecosystem.

Technology policy in the age of digital security states

What is happening in the discussions between Beijing and its tech companies is more than just a debate about export controls. It reflects a profound paradigm shift in global technology policy: The era of free information flow as the norm – with state restrictions as the exception – is over. It is being replaced by a regime in which states treat AI systems similarly to weapons, nuclear technologies, or state secrets.

Both superpowers pursue structurally similar strategies, even if their narratives differ: The US emphasizes protecting national security and preventing technology theft. China emphasizes sovereignty, protecting strategic resources, and countering American influence. In both cases, the economic calculation is the same: Control over the most powerful AI technology is a geopolitical lever of the first order—too valuable to leave to the market.

The Stanford AI Index 2026 illustrates the urgency of this assessment: In a single metric – the Arena scores for language models – China trails the US leader by 2.7 percentage points. Globally, China leads in AI patents with a share of over 74 percent, compared to 12 percent for the US and 3 percent for the EU. US investments in AI, at $285.9 billion, exceed China's $12.4 billion by more than thirteen times – but money alone no longer defines technological leadership.

In addition, there is China's strategic AI infrastructure: With more than 295,000 installed industrial robot units – almost nine times as many as the US – and the ability to add more electricity capacity annually than Germany consumes in total, China has an industrial base for AI implementation that Western decision-makers still underestimate.

Geopolitical Tectonics: The Changing Power Geography of AI

The potential Silicon Curtain – as Reuters aptly calls the emerging isolation of Chinese AI technology – has structural implications far beyond the immediate market for language models. It signals that the global AI market is fragmenting into two largely separate ecospheres: a US-dominated sphere and a Chinese sphere, with an increasingly squeezed Europe and a Global South struggling to find its place in between.

This bifurcation has industrial policy consequences for every company operating internationally. Those who have relied on Chinese models so far could be forced into more costly migrations. Those who rely on US models are at the mercy of the US export regime, which, as in the case of Anthropic, can prove to be arbitrary and short-term. And those who rely on European alternatives will not yet find a competitive offering at the technological frontier.

The Chinese government has thus unequivocally declared AI a strategic resource – comparable to rare earth elements, semiconductors, or military technology. The difference to these physical goods: AI exists in layers of weight and code that can move across borders without passing through customs. The challenge of regulation is therefore not only political but fundamentally technical. Whether Beijing succeeds in making its Silicon Curtain a genuine barrier or whether it degenerates into a porous symbolic gesture depends on the answer to precisely this question – and on how companies react as they operate in an increasingly shared digital world.

Uncertainty as the only certainty

The Reuters reports make it clear that concrete decisions are still pending: The scope of potential restrictions is still under discussion, it is unclear whether and when they will come into effect, and neither the Chinese government nor the companies affected have commented publicly. What is clear, however, is the direction: China is no longer treating its most advanced AI models as export goods, but as national assets – with all the consequences for a globally networked economy that is still wavering between naivety and a strategic response on this issue.

The development described is not an isolated event. It is part of a systematic reconfiguration of the global technological order, in which digital infrastructure, data flows, and AI capabilities have become core national security issues. Therefore, anyone who wants to understand the economic implications must look not only at the AI ​​market, but also at the geopolitical architecture that is currently being built around it.

☑️ Our business language is English or German

☑️ NEW: Correspondence in your native language!

Konrad Wolfenstein

I and my team are happy to be available to you as your personal advisor.

You can contact me by filling out the contact form here [email protected]:or simply call me at +49 7348 4088 965. My email address is

I'm looking forward to our joint project.

☑️ SME support in strategy, consulting, planning and implementation

☑️ Creation or realignment of the digital strategy and digitization

☑️ Expansion and optimization of international sales processes

☑️ Global & Digital B2B trading platforms

☑️ Pioneer Business Development / Marketing / PR / Trade Fairs

From visibility to trust: Your scalable path with Xpert.Digital - Image: Xpert.Digital

In industrial B2B, sustainable business relationships rarely emerge overnight. They develop step by step – through visibility, professional relevance, recurring touchpoints, and growing trust. Xpert.Digital's 4-stage model addresses precisely this: It offers a structured path that begins with a manageable entry point and can evolve into deeper collaboration in business development if needed.

Instead of relying on loud marketing promises, this model puts the relationship at the forefront. Companies start with clearly defined, easily calculable measures and then decide, based on their own experience, how far they want to expand the collaboration. A key factor for this undisturbed trust-building process: The platform completely avoids annoying advertising ads, so the editorial focus remains solely on the companies' expertise.

More information here:

• From visibility to trust: Your scalable path with Xpert.Digital