The Vise Closes: AI Control Becomes an Arms Race

The Vise Closes: China and the USA Monopolize Frontier AI

According to Reuters, China held talks in June 2026 with Alibaba, ByteDance, and Z.ai about export controls for its advanced AI models. The Chinese Ministry of Commerce is discussing restricting the release of top-tier models to domestic use — both open and closed systems.

This is not an isolated move. This is the closure of a strategic vise: The U.S. already controls frontier models (Fable, Mythos), now China is doing the same. For Europe, this opens a dilemma that will define the next decade.

China’s Approach: From Alibaba to Z.ai

The Ministry of Commerce is examining a three-tier control system:

1. Open-source tools: With reporting requirements, keep it simple
2. Advanced technologies: With security review
3. Frontier models: Domestic-only use or no release at all

The affected companies — Alibaba’s Qwen, ByteDance’s Doubao, and Z.ai’s GLM-5.2 — are among the world’s most widely used models. GLM-5.2 especially: It competes with U.S. top models but costs a fraction of the price.

The justification is not new. Chinese authorities fear the U.S. will use the Mythos model — developed explicitly for cybersecurity — against Chinese systems. Zhou Hongyi, founder of security firm 360, has called for a “Chinese Mythos”. This logic leads directly to export controls.

The U.S. Precedent: Fable, Mythos, and the Reversal of Freedom

China is not doing something new — it is copying the U.S. In June, the Trump administration shut down Anthropic’s Fable and Mythos globally. Because Anthropic had no real-time nationality verification, the company had to shut down for all users.

Fable was re-released after new security measures. Mythos remains available only to “trusted” U.S. organizations.

This is the strategic signal: Frontier AI is no longer a commodity. It is a weapon. Whoever controls it controls it for strategic purposes, not commercial ones.

The European Dilemma: Dependent on the Grace of Two Powers

Europe now faces a classic dependency scenario:

Before: “We buy from the U.S. or use cheap Chinese open-source models.”

Now: Both sources are being controlled.

According to analysis, this dependency has already led to foreign providers dominating over 80% of all digital products, services, and infrastructure in the EU. DeepMind (Google), ARM (SoftBank) — European innovations, bought and relocated.

Now a new, faster form of exploitation is emerging.

The Second Channel: Knowledge Drain Through Training Data

The classic leak was slow: Founders build startups, venture capitalists buy from outside, companies relocate.

The new channel is digital and faster.

Platforms like Mercor directly sell the knowledge of European experts to AI labs as training data. A junior software developer in Berlin can share their knowledge for compensation into an American or Chinese model — and in doing so, pass on exactly the knowledge that makes that model competitive.

This is not theft. This is a market. But the effect is the emigration of intellectual labor toward models that Europe does not own.

The InvestAI Initiative: Too Slow, Too Little, Too Late

The EU announced InvestAI: 200 billion euros over several years, with 20 billion for “AI gigafactories”.

That sounds massive. In context, it is underfunded.

The four major U.S. companies (Amazon, Alphabet, Microsoft, Meta) are spending around 700 billion dollars in 2026 together — three times Europe’s entire, multi-year initiative.

And the gigafactories themselves are delayed. The calls for proposals have been postponed multiple times. First infrastructure is scheduled to come online in 2027 — a year when the U.S. and China will already be a generation ahead again.

Moreover: With the “Digital Omnibus” package in November 2025, the EU loosened its AI regulations — the high-risk requirements of the AI Act are shifting to 2027-2028. Officially to give European companies room. Practically: Yielding to U.S. and corporate pressure.

The Structural Core: Compute and Data Are Real

Some in Europe hope for a “leapfrog” — a clever jump that skips missing compute and data. This hope is naive.

Models like Mythos emerged from “raw compute, massive data, and practical experience”. These are not substitutable by cleverness. They are accumulated, over years, with hundreds of billions in investment.

There is no trick for missing compute. There is no shortcut for missing data.

Europe’s Weight: Market and Regulation — But for How Long?

Why doesn’t China shut Europe out immediately? Two reasons:

1. Market: With 450 million consumers, the EU remains a market access
2. Regulation: The EU AI Act sets standards with global effect (“Brussels Effect”)

But this weight is eroding. The more European intellectual labor flows into foreign models, the less differentiation Europe can demand. The smaller European AI capacity remains, the less negotiating power.

In 5-10 years, the situation could be reversed: Not Europe dictating standards, but the U.S. and China — and Europe must accept.

What Happens Now Determines the Next Decade

China’s export controls are not an error. They are a declaration of war in the AI arms race.

Europe’s response has so far relied on investments and regulation loosening. This is not wrong, but it is too defensive and too slow.

What Europe needs:

1. Massive investment in compute (not 20 billion, but 100+)
2. Talent protection: Regulate knowledge drain through training data markets
3. Speed: 2027 is too late if the U.S. and China are already deploying further generations in 2026
4. A clear goal: Not regulation, but capacity

Without these, Europe in 2030 will be the land of data protection — while the U.S. and China build the AI that no longer respects data protection, because they have the capacity to ignore it.

The vise is closing. And Europe sits in it.