Why European AI Startup eustella Uses Chinese Open-Source Models

Originally published on Trending Topics on 10 April 2026.

At first glance, it sounds like a contradiction — perhaps even a PR problem. A European AI startup developing a sovereign smartphone agent chooses to build on models from China. Ten days ago, eustella announced its plan to build a European AI agent for smartphones, offering more than 100 million European AI users an alternative to ChatGPT, Claude, and others. The decision to use open-source models from DeepSeek, Alibaba's Qwen, or Moonshot looks like a strategic mistake in the current geopolitical climate. On closer inspection, it is the exact opposite.

The question is not where a model comes from — it is who controls it

The reflex is understandable. For months, geopolitical headlines have dominated the AI debate: export controls, chip embargoes, security reviews, the struggle for digital sovereignty. In this climate, integrating open models of Chinese origin into a European product seems questionable. But the real question is not where a model was built. It is who operates it.

A closed model from California, accessed through an API, sends every single request outside European territory — along with user data, context information, and everything else that makes a smartphone agent useful. An open model from China, whose weights are freely available, can be hosted on European infrastructure, audited, fine-tuned, and if necessary taken offline entirely.

Why China is aggressively pursuing open source

Chinese companies are not opening their models out of altruism. It is a strategic calculation with several motives:

• Circumventing export controls. Without access to the most powerful Nvidia chips, Chinese labs cannot compete with OpenAI, Anthropic, or Google in a pure compute arms race. Open source serves as leverage to build global relevance through reach and community rather than raw computing power.
• Free R&D at scale. Opening a model generates free feedback, fine-tuning, benchmarks, and tooling from tens of thousands of developers worldwide. For latecomers, it is the fastest way to close the gap.
• The model is not the product. Alibaba makes its money from cloud and commerce, Tencent from gaming and WeChat, Baidu from search. The model itself does not need to be monetised — it serves as a lead generator for cloud services or as a strategic asset.
• Commodifying the competition. When Chinese labs distribute strong models for free, they compress the prices and margins of GPT-class APIs worldwide. DeepSeek-V3 and R1 triggered exactly this effect — including sharp stock market reactions against Nvidia and Western AI equities.
• Geopolitical influence. Open models of Chinese origin running in products worldwide are a geopolitical statement. They give China influence over technical standards, tokenisers, and language coverage. In China's brutal AI market — the so-called "war of a hundred models" — a top benchmark result on Hugging Face doubles as a recruiting and reputation tool.

For eustella, these motives are secondary. What matters is the consequence: the interest of Chinese labs in commodifying American closed-source models happens to align with Europe's interest in technological independence. You just have to seize the opportunity.

eustella does not rely exclusively on Chinese AI. Google, OpenAI, Mistral, and potentially soon Meta all offer strong open-source LLMs that can be deployed depending on user needs.

The security concerns are real — but they concern something else

Open model weights are not remotely controllable. They do not phone home. They contain no backdoors that activate with the next update — an open model does not receive updates. It is a file.

What does need to be examined is bias, censorship behaviour, gaps in training data, and how a model responds to sensitive topics. These are solvable problems: through fine-tuning, through evaluation, through combining multiple models, through clear guardrails in the agent layer above. And it is precisely there — in orchestration and product design — that an AI startup creates its value, not in the base model.

The real controversy

The real controversy should not be that a European startup uses Chinese models. The real controversy should be that Europe spent years talking about building its own foundation models and ended up with a handful of labs that, despite billions in announcements, have not delivered what has long been available on Hugging Face. Mistral is the honourable exception, but one swallow does not make a summer.

As long as that is the case, it would be negligent to reflexively reject the best alternatives to American API dependency.

eustella is therefore taking the only pragmatic path: the company takes the best open models the market offers — regardless of whether they come from China, the United States, or Europe — runs them under European control, and builds an agent whose value lies in the product, not in the base model.

The truly European answer to the AI race is not choosing between Silicon Valley and Beijing. It is taking what is open from both camps and building something of your own.

Source: Trending Topics — "Warum das europaische KI-Startup eustella auf chinesische Modelle setzt", 10 April 2026.