AI companies in Europe risk losing their world-leading status in the field of machine translation, industry figures have said, after the decision by one of the continent’s leading startups to partner with Amazon’s cloud computing division provoked alarm.

While businesses in the EU have generally lagged behind the US and China in AI adoption, a small group of European companies have cornered the global market for high-quality machine translations for professional use.

The biggest success story is Cologne-headquartered DeepL, an online translator that regularly outperforms Google Translate in accuracy assessments. Used by governments, courts and half of the Fortune 500 list of highest-earning US companies, last year it was reported to have recorded revenues of $185.2m. Last month DeepL launched a live voice-to-voice translation service, reminiscent of the babel fish device envisaged in Douglas Adams’ 1981 novel The Hitchhiker’s Guide to the Galaxy.

Shortly before the announcement, however, DeepL informed its paying subscribers that it would “no longer process data exclusively on our own servers” and was entering a partnership with Amazon Web Services (AWS), which provides vital infrastructure for much of the internet.

The move prompted concern among users and observers of the sector in Europe, who say it will boost Silicon Valley’s monopoly over digital infrastructure just as the actions of the Trump administration prompt alarm over tech companies’ independence.

“I was not pleased,” said Jörg Weishaupt, the chief executive and founder of Malogica Group, a software business headquartered in Madeira, Portugal. He had been a longtime business customer impressed with DeepL’s performance, but he has decided to cancel his subscription.

His main concern, he said, was that he no longer felt comfortable uploading contracts or company strategy papers to DeepL’s site. “These are confidential documents, and I want to know where they end up.”

DeepL said AWS would not have access to its paying customers’ data, either for viewing its content or training Amazon’s algorithms, and said the partnership was vital to scale up its offering internationally.

A spokesperson said: “DeepL remains the data processor. We have added AWS as a sub-processor to our services providing the necessary infrastructure for global scale. AWS will not control or access customer data in any usable form. Customer data is encrypted in transit and at rest, and we do not use customer data from paid services to train our AI models.”

Weishaupt, however, pointed to the US 2001 Patriot Act and the 2018 Cloud Act, which allow the US government to request information from cloud providers. Last July, a Microsoft director of legal affairs said under oath at a hearing in France that the company “cannot guarantee” data sovereignty to customers in the EU should the Trump administration demand access to customer information held on its servers.

Since the Cloud Act, US companies have rushed to create technical solutions to assure non-US clients that their data is safe. DeepL says concerned customers can choose a data residency option that guarantees their data will not leave Europe. But some question whether such reassurances can be relied upon.

Weishaupt said: “There’s a big movement towards sovereignty in Europe at the moment. It may have been caused by the current geopolitical situation, but it won’t go away. We are all trying to get out of the lock-in with the Americans.”

The Trump administration has repeatedly clashed with the EU over European attempts to regulate big tech companies, and in her 2025 state of the union address the European Commission’s president, Ursula von der Leyen, said that “to take control over the technologies […] that will fuel our economies” could amount to “Europe’s independence moment”.

In such a climate, any collaboration between European AI translators and US cloud providers is likely to draw some criticism, including from within the sector.

Marco Trombetti, the co-founder and chief executive of Translated, a Rome-based company and DeepL competitor, said: “Europe needs to be absolutely independent in terms of infrastructure. Digital infrastructure is the road network of today. We cannot pay a toll when we want to do business. If you pay that toll for every transaction, that will simply increase the toll and not allow you to go on that highway. And that’s literally a big, big error.”

He also said it runs counter to businesses’ own long-term interests. While as much as 80% of his company’s revenue today comes from Silicon Valley, with clients including Airbnb, Uber and Starlink, Trombetti said he had not relocated to the US.

“It would be a disaster. AI translation companies have thrived in Europe because we operate in a multi-lingual market that has made us acutely aware of the problem we are trying to overcome. From Europe, you can easily cover 200 languages within a two-hour timezone window.”

Relying on American infrastructure, he said, would risk European companies giving up their competitive advantage. In January, the US Department of Commerce introduced rules that mean American companies will get priority access to US-made chips, particularly advanced graphics processing units (GPUs), when demand exceeds production.

“The playing field will become increasingly uneven,” said Trombetti. “US companies having priority access to chips creates a strong incentive to relocate to the United States. It works in the short term, but the more they restrict the rest of the world, the more they motivate China and the EU to build alternatives.”

Building a European digital road network, however, poses a significant challenge. Leevi Saari, a Finnish researcher at the University of Amsterdam and AI Now Institute, said: “As DeepL wants to scale up its services, it needs more capacity, and what Amazon offers would have looked very tempting.”

In a global AI boom, building datacentres has become increasingly expensive and the rate at which hardware chips lose their value has increased, meaning that in spite of AWS outages in 2025, few European companies have switched to local clouds. As machine translation companies switch to live text and voice-to-voice translation, the main technical issue they are trying to address is datacentre latency, Saari said.

Datacentre latency refers to the time it takes for data to travel from point A to point B, which is usually measured in milliseconds. AWS achieves low latency by having established a network of datacentres around the world and by laying its own network of subsea fibre-optic cables.

“Currently, the gravitational forces of the AI industry are such that startups will end up being pulled towards the US,” said Saari. “How can Europe create its own AI gravity well? That’s the trillion euro question.”