US AI dependence puts Europe’s high-tech industry at risk

Europe’s reliance on American artificial intelligence models threatens the development of high-tech products and could slow innovation if access is restricted, BR reported. European experts and a defence startup say the continent has the expertise to build its own AI, but lacks sufficient computing capacity and access to key chips.

The prototype of the RDX-1 races through the workshops of Reactive Dynamics, a young defence startup in Reichertshofen near Manching. The roughly three-metre-long vehicle resembles a small tank, but is an unmanned, electrically powered tracked vehicle designed to move autonomously across a battlefield, including for rescuing wounded soldiers.

Co-founder Joris Bauer-Ludwigs wants the vehicle to be produced entirely in Europe. While its construction is technically feasible, he said using only European components is currently almost impossible, particularly when it comes to software.

“Artificial intelligence is essential to our development,” Bauer-Ludwigs said. “We use the large models from the well-known American companies. At the moment, they are absolutely unrivalled.” He said there was currently no European alternative.

The risks of relying on US-based AI became clear when US President Donald Trump temporarily prohibited non-Americans from using Anthropic’s best AI model, according to the report. “That was a real eye-opener,” Bauer-Ludwigs said. If such a restriction happened again on a larger scale, he said, development would lose considerable speed.

AI is increasingly used to programme software, including systems for controlling vehicles. The newer and more advanced the model, the better the software it can produce. Jörg Bienert, co-founder of the German AI Association, said AI was now the basis for almost every high-tech product.

A European cut-off from US AI would affect products ranging from autonomous vehicles and accounting software to drones for European defence, Bienert said. He therefore launched the initiative to develop the AI model Soofi.

Bienert rejects the view that Europe has missed its chance to develop large language models. “We have the scientific expertise, and we have the data to train the models,” he said. A consortium he initiated trained Soofi within a year, initially as a small version. A larger version is now being developed.

“We have proved that we can train our own language models, powerful language models,” Bienert said.

Work on Soofi also exposed Europe’s shortage of computing capacity, he said. Investment is urgently needed, but funding is not the only obstacle. Most AI data centres around the world use chips made by US manufacturer Nvidia, with practically no alternative available. New data centres can therefore be built only if Nvidia supplies the chips.

Bienert sees European cooperation as the only way to overcome this dependence. The European Commission launched the CHIPS Act in June to promote semiconductor production in Europe and strengthen the continent’s chip industry.

Bauer-Ludwigs has not abandoned hope of eventually building a fully European robotic vehicle. He said Europe had the necessary talent, but lacked the resources and environment available to companies such as OpenAI and Anthropic.

“We do have these people, but they do not have the resources and the environment to develop models on a similar scale,” he said. “We simply have to change that.”