U.S. AI Restrictions Wake Up European Giants: Firms Rush to Diversify or Risk Being Cut Off

.

A Single Order Changed Everything

It was a quiet but significant move. The U.S. government recently ordered Anthropic — the San Francisco company behind the Claude AI system — to suspend access to two of its most advanced models, Fable 5 and Mythos 5, for users outside the United States. The reason given: national security.

The decision sent a shockwave through European boardrooms. Suddenly, the risk that had long been discussed in theory became very real. A company in Paris or Berlin could lose access to a critical AI tool overnight — not because of anything it had done, but because Washington decided so.

Big Names Are Already Hedging Their Bets

Major European corporations were not entirely unprepared. Executives from Siemens, Renault Group, Orange and French AI firm ChapsVision told Reuters at last week's VivaTech conference in Paris that they had already begun mixing providers — combining American, Chinese and European AI models rather than committing fully to any one.

Siemens, for instance, works with Chinese models DeepSeek and Alibaba's Qwen alongside Nvidia's Nemotron and various U.S. and European systems. Renault Group collaborates with Google, Microsoft, Mistral and Dataiku, using both open and proprietary models.

For Cedrik Neike, Chief Executive of Digital Industries at Siemens, the lesson is straightforward: "You need flexibility. Sovereignty often gets confused with autarky — economic self-sufficiency — and autarky is absolutely not the way to do it."

The Core Problem: You Don't Own What You Can't Run

The Anthropic restrictions exposed a fundamental weakness in how most companies use AI today. So-called proprietary models are accessed remotely, over the internet, and remain fully under the developer's control. If that developer — or its government — decides to cut access, there is no workaround. The model cannot be downloaded and run on a company's own servers.

Open-source or open-weight models offer an alternative: companies can download and operate them independently. But European offerings in this space remain limited. France's Mistral is the most prominent example, while others like DeepL have carved out strong positions in narrow tasks like translation.

"Today, in open source, when you look at European models, they're not impressive," said Octave Klaba, CEO of OVHcloud, a French cloud provider. "At one point, the Americans were there, then they moved to closed source, and now there are only Chinese models in open source."

The China Dilemma

Using Chinese AI models raises its own set of concerns — chiefly around data privacy and the reach of the Chinese Communist Party (CCP) over its tech companies. Chinese law can compel domestic firms to hand over data to state authorities, creating legitimate questions about what happens to information processed by models like DeepSeek or Qwen.

Orange, France's largest telecom group, offered one way of thinking about this risk: running a Chinese model on European infrastructure is like buying a painting in China and bringing it home. Once it operates locally, the model itself does not send data back abroad. The key is where and how it is deployed.

Still, the broader dependence on non-European technology — whether American or Chinese — remains a strategic vulnerability that European leaders are increasingly unwilling to accept.

Orange CEO Christel Heydemann used her keynote address at VivaTech to call on Europe to build AI it can "access, govern and challenge on its own terms."

ChapsVision: Always Have a Backup

ChapsVision, a French AI and data analytics company that has secured government contracts in France and Germany — partly replacing U.S. giant Palantir — puts it bluntly: sovereignty means having a credible backup ready at all times.

The firm currently uses a mix of Mistral, Anthropic, OpenAI and Qwen models. But it treats no single provider as irreplaceable. If a key service is suddenly switched off, ChapsVision needs to keep its government clients running without interruption.

This pragmatic view is shared by larger software groups. SAP and Sopra Steria have both said resilience comes through diversification, not isolation. IT services giant Capgemini noted that most AI providers are already adapting their offerings to reduce dependency concerns in Europe — because the European market is too large and valuable to lose.

The Hidden Cost: Token Bills Are Exploding

Beyond access risk, companies are confronting a second, fast-growing problem: cost.

AI services charge by the "token" — roughly speaking, each word or fragment of text processed. As companies move from simple AI chatbots to complex automated systems — where software agents carry out tasks independently — the number of tokens consumed can multiply dramatically.

Orange executives said they expected to be "obsessed with cost per token" by the end of this year. As an example of how quickly costs can spiral, they cited ride-hailing company Uber, which reportedly burned through its entire 2026 AI token budget within just four months.

Renault Group acknowledged the same pressure: "Token costs have risen sharply and are pushing us to adapt," a spokesperson said.

Rudy Kuhn, a senior executive at German software company Celonis — whose clients include BMW and Siemens — warned that deploying AI agents without proper preparation could be financially ruinous. "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."

What This Means for Europe

The broader picture is one of a continent caught between two technological superpowers — the United States and China — both of which present their own risks. American models can be restricted for geopolitical reasons. Chinese models raise data sovereignty and security questions.

The European Union has been working on a technology sovereignty agenda, including measures on semiconductors, AI and digital autonomy. But as industry leaders at VivaTech made clear, that agenda needs to translate into real alternatives — not just policy frameworks.

For now, the most practical answer for European businesses is diversification: using multiple AI providers simultaneously, maintaining open-source fallbacks, and building infrastructure that does not depend on any single foreign supplier. It is not perfect sovereignty — but in a world where a government order can cut off your AI tools overnight, it may be the best available insurance policy.

.

Sources

1. Reuters – "US curbs on AI spur European firms to spread the risk," Leo Marchandon, June 22, 2026: https://www.reuters.com/legal/litigation/us-curbs-ai-spur-european-firms-spread-risk-2026-06-22/
2. European Commission – EU AI Act and Digital Sovereignty Package overview: https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence
3. Human Rights Watch – Report on Chinese tech companies and state data access obligations: https://www.hrw.org/news/2022/03/22/china-new-data-laws-tighten-communist-party-control

.