Get all your news in one place.
100’s of premium titles.
One app.
Start reading
Radio France Internationale
Radio France Internationale
Jan van der Made

Europe's tech sector sees silver lining in DeepSeek's AI shake up

DeepSeek claims to have developed its AI model with just €6.23 million, far below its Western competitors. AP - Andy Wong

China’s DeepSeek AI chatbot may have rattled US tech giants, but in Europe some industry players see a potential advantage.

As US-based company Nvidia – the world’s leading manufacturer of AI chips – reels from a record-breaking stock drop, European semiconductor firms and AI developers are weighing what the disruption could mean for them.

Philippe Notton, CEO of SiPearl, a European company developing processors for supercomputers, told RFI that DeepSeek’s ability to develop AI with fewer resources could be a turning point.

"That's bad news for Nvidia in terms of future sales, because if you can develop some competitive solution with fewer Graphics Processing Units (GPUs), it means that Nvidia will sell fewer chips," he said.

"All the forecasts predict that the hype on GPUs and Nvidia is collapsing."

Artificial intelligence may lead humanity to extinction, industry leaders warn

Unexpected crash

DeepSeek’s launch last week sent tech stocks plummeting.

On 27 January, Nvidia, called the "posterchild of America's AI frenzy" by Bloomberg, lost $589 billion in market value – the biggest market-cap loss for a single stock ever.

The Nasdaq 100 fell 3 percent, and the S&P 500 dropped 1.5 percent.

DeepSeek claims to have developed its model with just €6.23 million, far below its Western competitors.

For comparison, Stephen Walker, an AI developer and founder of Klu.ai, estimates that training OpenAI's ChatGPT requires about 25,000 Nvidia H100 chips costing between €23,900 and €29,700 each – bringing total development costs to nearly €920 million.

"If what DeepSeek said is true, they can develop such a model for only €5.75 million with some 2,000 to 4,000 GPUs, which is very, very low compared to what the others are using," Notton said.

"It's good news for the planet because it's going to use much less energy to build this. It's good news for Europe because they could do it for a limited budget."

AI development cannot be left to market whim, UN experts warn

Plagiarism concerns

As the dust settled, accusations surfaced that DeepSeek may have built its model using data from US companies. OpenAI and Microsoft are investigating whether DeepSeek trained its chatbot using their proprietary data.

Daniel Castro, vice-president of the Washington-based Information Technology and Innovation Foundation (ITIF), compared the situation to past innovations.

"Apple didn't invent the smartphone," he said. "They just invented the best one, and that's why they were so successful with the iPhone.

"When something like this comes out, all the other companies are asking themselves: what are we doing to make sure to lower the costs. Ultimately that competition will be very good for the AI industry."

Visitors look at artist Refik Anadol's "Unsupervised" exhibit at the Museum of Modern Art, Wednesday, Jan. 11, 2023, in New York. The new AI-generated installation is meant to be a thought-provoking interpretation of the New York City museum's prestigious collection. AP - John Minchillo

French industry

Meanwhile, European tech firms scrutinise the DeepSeek phenomenon with interest.

According to Semiconductor Review, the industry generates €25 billion in yearly revenue and employs more than 50,000 people, constituting over 10 percent of the country's total exports.

However, Acsiel, which monitors trends in the French electronics sector, reports the market fell 19 percent in the third quarter of 2024, to €486 million.

A massive reduction of research and developing costs in the increasingly competitive AI market may prove very welcome.

The French AI start-up Mistral on Thursday hailed the latest DeepSeek model as "great," and announced another new release of its own.

"R1 is a great and complementary piece of open-source technology," Mistral said in a statement, while announcing its own new release "Mistral Small 3", which it claims is competitive with larger models including Meta's Llama and Alibaba's Qwen.

This picture taken on March 25, 2024, shows the Mistral Ai logo on a smartphone in Mulhouse, eastern France. AFP - SEBASTIEN BOZON

Over-regulation?

However, US analysts argue that European AI regulations could hinder innovation.

Stephen Ezell, vice-president at ITIF, said the EU's AI Act, introduced in July 2024, was one of the most restrictive regulatory regimes we've seen for AI globally.

"If I was a European policymaker looking at how a company like DeepSeek is challenging probably the top European AI company, Mistral ... I would be very concerned about the approach of now putting them at a further disadvantage through this regulatory regime that really restricts access," Ezell said.

The developments gain added significance following US President Donald Trump's announcement of the €500 billion Stargate project, a joint venture between OpenAI, Oracle and SoftBank that he called "the largest AI infrastructure project in US history".

This is something that Europe cannot compete with, Notton said.

"But if finally, we can do it for let's say 100 times cheaper, it becomes much more reasonable," which makes the DeepSeek phenomenon "a kind of revolution because if it can be produced for a lower price, much more countries will be able to do it."

France's Mistral AI signs partnership with Microsoft

Fears for free speech 

Advocates of free speech and data protection advocates are worried.

If users ask DeepSeek questions that are sensitive to China's Communist Party, it suddenly stops functioning properly.

Probes about the June 4, 1989 Tiananmen Square massacre, independence for Tibet or Taiwan, or about deposed politicians or Chinese dissidents are flatly answered with "sorry, that's beyond my current scope. Let’s talk about something else", or "I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses."

But Castro is not worried. "American and European researchers are not going to use this AI chatpod to research Chinese history or politics," he told RFI. "If that was the primary use, that would be of concern."

Screenshot of DeepSeek giving evasive answers on questions that may be politically sensitive in China © Screenshot DeepSeek

Yet other worries remain.

Dieter Kugelmann, president of Germany's Authority for Data Protection and Freedom of Information, warned that DeepSeek "seems to be lacking in pretty much everything in terms of data protection law".

The app collects extensive user data – including IP addresses, chat histories and keystroke patterns – which could be stored on servers in Hangzhou, thus potentially accessible to China's Ministry of State Security.

If these concerns prove valid, the EU may need to act. The European General Data Protection Regulation only allows data transfers with countries offering comparable protections to the EU.

No such agreement exists between China and the EU.

Sign up to read this article
Read news from 100’s of titles, curated specifically for you.
Already a member? Sign in here
Related Stories
Top stories on inkl right now
One subscription that gives you access to news from hundreds of sites
Already a member? Sign in here
Our Picks
Fourteen days free
Download the app
One app. One membership.
100+ trusted global sources.