Sovereign AI: Why European companies want to keep their data out of Big Tech’s hands

From the European Commission comes a steady push for technological sovereignty, something Europe still largely lacks. In recent years, that effort has intensified as artificial intelligence has burst onto the scene like a bull in a china shop. The European Union has stepped up its attempts to manufacture advanced semiconductors while also promoting the construction of data centers and the development of homegrown AI.The urgency is growing, though the available options are not. In reality, an AI system built entirely on European technology is impossible. The dependence on chips from U.S. company Nvidia alone is enough to dash any such ambition. But technological sovereignty, like most things, exists on a spectrum. And one of its most critical dimensions is the infrastructure where data is stored and processed.As companies make increasing use of artificial intelligence, their internal data are becoming ever more valuable. And some businesses are reluctant to let their information travel to the servers of OpenAI, Anthropic, Google or Microsoft in order to use their AI tools.Using those companies’ services directly often means sending sensitive corporate information to the cloud. “People have started reacting negatively to these AI models,” explains Charles Sansbury, chief executive of U.S. multinational Cloudera. “They raise concerns because companies lose control of their data, which leaves their perimeter, whether that perimeter is technological or geographic.”Sansbury’s company operates behind the scenes, in the areas where businesses store their data and seek to extract value from them. Today, part of its business revolves around helping companies apply AI to these vast repositories of internal information. During EVOLVE 2026, an event organized by Cloudera in New York on October 27 and attended by EL PAÍS as a guest, the firm highlighted how companies are increasingly protective of their data. Many now prefer to run AI models in a private cloud or on their own local servers.Some companies go so far as to run AI systems on information stored on machines that are completely disconnected from the internet, further protecting what has become one of their most valuable assets: data, the so-called gold of the 21st century. “The term sovereign AI has a more geopolitical nuance,” Sansbury says. “But really it’s about controlling where your data go, who processes them and with what systems. In the end it’s about giving customers the tools to run any application within a control framework they have defined beforehand.”Growing distrust across the Atlantic and Europe’s determination to loosen the grip of U.S. tech giants on the digital economy have heightened geopolitical tensions. Those dynamics are helping to fuel ambitions for sovereign AI. One of the companies best positioned to benefit from that demand is the French company Mistral.“It is absolutely essential for companies to inject their data, protected by their intellectual property, into AI models,” says Marjorie Janiewicz, chief revenue officer at Mistral. “There is a direct correlation between doing that and getting value from AI systems. There are clear signs that if you don’t enrich models with your intellectual property, you lose a competitive edge to other companies.” The French startup, which recently partnered with Cloudera, whose data-management platform now offers Mistral’s models, works with some of Europe’s industrial heavyweights. Companies including ASML, TotalEnergies and Airbus have integrated their data into Mistral’s models. All operate in highly regulated sectors where safeguarding information is even more critical.Keeping the data at homeThe European Commission has drawn up a specific roadmap for the digitalization of the energy sector and the deployment of AI. The industry, already subject to extensive regulation, has become increasingly strategic in the wake of the war in Ukraine and rising electricity prices. The EU does not want the rollout of artificial intelligence in this sector to depend on foreign providers.A similar dynamic is playing out in other industries that are also regarded as critical to the functioning of the state. “For companies in regulated sectors in Europe, whether in Spain, France, Germany or the Nordic countries, national rules are emerging that prevent data from leaving the country’s geographic perimeter,” says Sansbury, whose company works with much of the banking and telecoms sector in Spain.In this environment, companies seeking to ensure that their data never leave their own systems are increasingly turning to so-called open-weight models. These models can be downloaded, unlike those developed by OpenAI or Anthropic, and allow users to modify certain parameters. The result is that companies can tailor AI systems to their own context and requirements.But European companies do not have many options. “If you need a high-quality [open] AI model and you need it to run on your servers, you can turn to Nvidia’s models, but those are American. And half of the [open] models available on the market are Chinese, which brings its own set of geopolitical concerns,” Sansbury notes.There is a certain irony in the fact that Cloudera, a U.S. company with a global footprint, is helping drive sovereign AI among European businesses. Its systems, built on an open-source architecture, allow companies to feed in data privately and process it using AI models from different providers. Within the European Union, however, the preferred option is increasingly Mistral.Janiewicz from the French company offers an example of how its systems can be useful to businesses: “Applying artificial intelligence today is very different from just a few years ago. Today we can have real use cases ready in three months. Working on crash-testing trials with BMW, we built a test model very quickly that speeds up BMW’s vehicle evaluation.”In this case, the data BMW uses to build its crash-test model are far too valuable to upload to the servers of OpenAI or Anthropic. No matter what guarantees those companies provide on paper, a degree of concern remains. Could confidential information eventually leak? Might it be used to train new AI models? What happens if the terms of service change, and with them the way data are handled?Companies seek to avoid such uncertainties by running AI models on their own systems, without sending information to the cloud. A Cloudera report, based on surveys of more than 1,500 technology leaders worldwide, found that nearly half of businesses have already moved part of their AI operations to private systems.“A bank, which holds transaction, deposit and investment data of its customers, does not want to upload that information to a cloud-based system,” Sansbury explains. “It does not want to place all those proprietary data somewhere outside its own domain. What we can do through our tools and the AI models we provide access to is enable companies to build AI in an environment they control, without information leaving their security perimeter.”In practice, having sovereign AI is little more than having private AI. In Europe, the former term has gained traction, but at its core it reflects a simple concern shared by many companies: reducing the risk that sensitive business information might end up in someone else’s hands.Sign up for our weekly newsletter to get more English-language news coverage from EL PAÍS USA Edition