Restrictions imposed by U.S. authorities on access to particular artificial intelligence services have prompted several large European companies to spread their AI usage across multiple suppliers and to strengthen demand for regional alternatives that they can control.
The immediate catalyst cited by executives is the U.S. government order instructing San Francisco-based Anthropic - the developer of the Claude chatbot - to suspend foreign nationals' access to its Fable 5 and Mythos 5 models, citing national security concerns. The move underscored a key vulnerability for businesses that rely on proprietary AI offerings delivered remotely: such services remain under the developer's control and can be restricted or withdrawn, and they cannot always be operated independently on a company's own servers.
At last week’s VivaTech conference in Paris, senior executives from Siemens, Renault Group, Orange and ChapsVision told attendees they already deploy a mix of U.S., Chinese and European AI models to avoid dependence on any single provider. Siemens, for example, reported using Chinese models DeepSeek and Alibaba’s Qwen alongside Nvidia’s Nemotron and other U.S. and European models.
European policy makers have for some time sought to reduce reliance on U.S. technology, framing that dependence as a potential threat to the region’s economic prospects. The European Union has assembled a so-called sovereignty package of measures aimed at strengthening capacity in semiconductors, AI and digital autonomy. Yet corporate leaders at VivaTech framed sovereignty more as the preservation of choice and diversity than as an aspiration for complete self-sufficiency.
Cedrik Neike, chief executive of Digital Industries at Siemens, captured that distinction, saying: "You need flexibility." He added: "Sovereignty often gets confused with autarky (economic self-sufficiency), and autarky is absolutely not the way to do it."
Europe’s set of general-purpose AI providers remains limited, led by France’s Mistral, while other European firms such as translation specialist DeepL maintain strong positions in narrower niches. The AI market is broadly split into two technical models: open-source or open-weight systems that companies can run on their own infrastructure, and proprietary models accessed remotely that stay under the developer’s control.
Octave Klaba, chief executive of OVHcloud, offered a frank assessment of the open-source landscape in Europe: "Today, in open source, when you look at European models, they’re not impressive. At one point, the Americans were there, then they moved to closed source, and now there are only Chinese models in open source," he said.
Orange, a communications and infrastructure provider, said its systems can host all open-source models, including those originating in China, and used an analogy to explain the operational difference: running a Chinese model on European infrastructure is like buying a painting in China and bringing it home, because the model is self-operated and does not send data abroad when run on premises. The company said the Anthropic restrictions made "patently clear, if it wasn’t before, how important it is for Europe to have access to an AI service that it can control, that will never be switched off on a whim."
Christel Heydemann, Orange’s chief executive, used a keynote address to urge Europe to develop artificial intelligence it can "access, govern and challenge on its own terms."
French AI and data analytics company ChapsVision, which has secured government contracts in France and Germany to replace a U.S. rival, reported using a range of models spanning Mistral and Anthropic to OpenAI and Qwen. For ChapsVision, sovereignty means having a credible substitute ready if a primary service is cut off.
Other software and IT services companies echoed the resilience-through-diversification message. SAP and Sopra Steria said they see resilience coming from using multiple providers rather than isolating systems. Capgemini said most AI suppliers were evolving their offerings away from pure remote-only access to address dependency concerns in Europe, though it cautioned this remained a work in progress.
Cost pressures are emerging alongside sovereignty concerns. Token costs - the charges assessed per unit of information processed by AI systems - are rising as firms move to setups in which software agents perform tasks automatically. Orange warned that executives will be "obsessed with cost per token" by the end of this year, and cited Uber as an example of a company that had exhausted its 2026 token budget in just four months.
Renault Group said it works with Google, Microsoft, Mistral, DeepSeek and Dataiku, employing both open-weight and proprietary models, though it noted DeepSeek was not yet used in a meaningful way. "Renault Group already has an in-depth reflection on the cost of AI tokens, which have risen sharply and are pushing us to adapt," a Renault spokesperson said.
Beyond vendor choice and cost, companies face infrastructure and integration challenges. Rudy Kahn, a senior executive at German software firm Celonis, which counts BMW and Siemens among its clients, said organizations must first build the context models that give AI agents an understanding of how their business operates before deploying agents broadly. "If you do not provide a context model, AI needs to extract every single fact from the data itself," he said. "This will just blow your token bill completely."
As European firms respond to restrictive access by major U.S. AI providers, the predominant corporate strategy emerging from discussions at VivaTech is to combine multiple models and hosting approaches - using open-weight systems where feasible, keeping proprietary services in the mix where they add value, and pushing for regional services that can be controlled and governed locally. At the same time, spiraling token costs and the need to develop context-rich infrastructure are key practical constraints shaping how quickly firms can pivot away from single-provider reliance.