German companies rely almost exclusively on AI models and services from the USA

# US AI dominates companies: China models hardly used, EU AI plays no role Source: [https://www.heise.de/en/news/US-AI-dominates-companies-China-models-hardly-used-EU-AI-plays-no-role-11446969.html](https://www.heise.de/en/news/US-AI-dominates-companies-China-models-hardly-used-EU-AI-plays-no-role-11446969.html) Open\-weight models from China currently play a minor role in the German economy, according to a survey by the digital association Bitkom\. Only two percent of companies using AI use DeepSeek, and one percent use Alibaba’s Qwen models\. This puts Chinese providers far behind the leading models from the USA\. ChatGPT from OpenAI remains the dominant player by a clear margin: 76 percent of AI\-using companies use it, up from 70 percent in the previous year \(multiple answers possible\)\. This is followed by Microsoft’s AI assistant system Copilot with 35 percent \(2025: 28 percent\), which uses both Microsoft’s own AI models and models from OpenAI and Anthropic\. Google’s Gemini models follow with 28 percent \(2025: 22 percent\)\. Clearly trailing, Meta’s Llama models reach 8 percent, SAP’s Joule AI functions 6 percent, and Anthropic’s Claude 3 percent of companies\. According to the survey, Grok from xAI plays virtually no role, not even being used in one in a hundred German companies\. Furthermore, no surveyed company indicated using models from Perplexity or the leading European provider Mistral, based in France\.[According to Bitkom, 603 companies with 20 or more employees were surveyed](https://www.bitkom.org/Presse/Presseinformation/KI-aus-China-fuer-deutsche-Wirtschaft-kein-Thema), and the results are representative\. ### Trend towards open models? “The German economy currently relies almost exclusively on applications from the USA for artificial intelligence\. Those who want digital sovereignty must give European providers a chance,” commented Bitkom President Dr\. Ralf Wintergerst on the results\. “At the same time, not every task in a company requires the largest and newest model\. For many applications, specialized, smaller systems are sufficient, which can then also be used more cost\-effectively – this is precisely where there is also an opportunity for European providers\.” Open\-weight models from China[like Kimi K3](https://www.heise.de/news/Chinas-neuer-KI-Riese-Moonshot-AI-fordert-US-Dominanz-heraus-11369310.html?from-en=1)or[Qwen 3\.8](https://www.heise.de/news/Qwen3-8-Max-Alibaba-legt-mit-neuem-KI-Modell-nach-11395102.html?from-en=1)have caused a sensation in recent months\. They compete with top US models in benchmarks and promise significantly lower token costs than proprietary US top LLMs from OpenAI and Anthropic\. Thanks to published weights, companies can also host the models locally\. Companies like Coinbase and Airbnb have already publicly stated that they use Chinese open\-weight models on Western infrastructure for certain tasks\.[Nvidia is strongly advocating for a US ecosystem of more open models](https://www.heise.de/news/Offener-Brief-Microsoft-Nvidia-Co-fordern-offenes-KI-Oekosystem-11378076.html?from-en=1)and[recently acquired the Hugging Face platform](https://www.heise.de/news/Nvidia-uebernimmt-Hugging-Face-fuer-12-93-Milliarden-US-Dollar-11440209.html?from-en=1)\. It remains to be seen whether this will become a sustainable trend that puts pressure on providers of closed models and also finds its way into the German economy\. \([axk](mailto:[email protected])\) Don't miss any news – follow us on[Facebook](https://www.facebook.com/heiseonlineEnglish),[LinkedIn](https://www.linkedin.com/company/104691972)or[Mastodon](https://social.heise.de/@heiseonlineenglish)\. *This article was originally published in[German](https://www.heise.de/news/US-KI-dominiert-in-Firmen-China-Modelle-kaum-genutzt-EU-KI-spielt-keine-Rolle-11446802.html)\. It was translated with technical assistance and editorially reviewed before publication\.*