Despite Fable 5 warning, European firms resist AI sovereignty
When export controls abruptly locked European enterprises out of the world’s leading AI models, it exposed the fragile reality of a continent dependent on foreign tech. But as the EU pushes for digital independence, businesses are pushing back — warning that isolation breeds inefficiency. Can European industry afford to buy local, or is a pragmatic, multi-cloud compromise the only way to survive?
In disabling Fable 5 and Mythos 5 for all users worldwide, Anthropic has underlined the dangers of not having a sovereign AI strategy. Users in Europe, UK and elsewhere have been shut out of the lab’s latest models, and while negotiations with the US government are ongoing, the withdrawal raises the disturbing possibility that Europe could permanently live under the threat of an AI ‘kill switch.’
The bloc as a whole depends on non-EU providers for over 80% of key digital products
In fact, examples currently abound of businesses doing more or less the opposite of what the EU’s tech sovereignty drive would recommend. France-based Thales signed a strategic partnership with Google Cloud in May aimed at delivering a “sovereign cloud in Germany,” while Telefónica signed a very similar agreement in the same month with the same US hyperscaler, albeit in Spain. More generally, the EU has noted that the bloc as a whole depends on non-EU providers for “over 80% of key digital products, services, infrastructure, and intellectual property.”
Given recent protestations and agreements, it seems hard to imagine this percentage will decline in the near term. Market research from Veeam indicates a disconnect between organisational pressures to roll out AI speedily, on the one hand, and control and visibility over tech resources, on the other. Meanwhile, technology analysts affirm that European enterprises are prioritising tech performance over sovereignty. However, analysts also argue that businesses are becoming open to a more pragmatic approach that mixes non-sovereign and sovereign tech.
Sovereignty remains a ‘contested and negotiated concept’
“The evidence suggests that most European businesses are currently prioritising AI adoption, productivity gains and competitive performance ahead of full AI or data sovereignty, although sovereignty concerns are becoming increasingly important,” says Francesca Musiani, a Senior Researcher at the Paris-based National Centre for Scientific Research (CNRS).
Musiani cites research from (the CNRS-affiliated) DIGISOV and ClaimSov, which have highlighted how tech sovereignty is becoming increasingly prominent as a European policy objective, yet it remains a “contested and negotiated concept” rather than a consistently applied business priority.
“DIGISOV in particular highlights the tension between strategic autonomy, innovation and economic competitiveness, noting that European digital governance is shaped by ongoing trade-offs between regulatory ambitions and market realities,” she explains. “In practice, many European firms continue to rely on AI models, cloud infrastructure and platforms provided by US hyperscalers because these s olutions offer superior scale, performance, maturity and speed to market.”
European firms to rely on AI models, cloud infrastructure and platforms provided by US hyperscalers
A similar overview is offered by other commentators, with Veeam’s GM and SVP for EMEA, Tim Pfaelzer, reporting that many organisations are accelerating AI adoption without having “a full understanding” of where their data resides or how it’s managed. He says, “There is a growing gap between the pace of AI deployment and the maturity of data visibility and control.”
This is reflected in the aforementioned research from Veeam, which found that 83% of CEOs worldwide report pressure to accelerate rollouts of AI and data capabilities. However, only 30% have visibility over the provenance and makeup of their organisations’ data resources. Such a gap creates significant risk, according to Pfaelzer, since AI systems depend strongly on the quality and accessibility of data.
He adds, “If organisations do not have clear insight into their data estate, they risk exposing sensitive information, breaching regulatory requirements and undermining the very outcomes they are trying to achieve with AI.”
Sector-specific sovereignty
Despite the preference for faster rollouts and superior performance, some experts testify that European enterprises are increasingly paying attention to sovereignty concerns. “In my experience, by and large European businesses are more focused on AI rollout, adoption and performance,” says Zach Meyers, Director of Research at the Centre on Regulation in Europe (CERRE). “However, sovereignty considerations are being taken more seriously in certain cases.”
Meyers suggests that certain kinds of organisations are more likely to prioritise digital sovereignty. This includes enterprises that place particular stock on protecting their IP and trade secrets. He says, “This concern seems particularly acute in industries of geopolitical significance — for example, it seems likely that it was a major consideration in ASML’s acquisition of a stake in Mistral.”
Similarly, more regulated sectors, such as finance and energy, tend to produce organisations more interested in sovereignty. For Pfaelzer, regulators in such sectors are concerned about the risks that arise when companies are dependent on only a few providers for critical infrastructure and services, especially if these providers are all based in the same country (i.e. the United States).
Choice of AI provider is driven primarily by performance and value-for-money
“However, for most European businesses the choice of AI provider is driven primarily by performance and value-for-money,” he adds. “One concern is that many companies that are worried about digital sovereignty might choose simply not to adopt new technologies at all — particularly when choosing a European technology provider is not as easy and seamless as using a US company.”
Resistance to a 100% sovereign stack
Given such tendencies, experts suspect that most European enterprises will not be open to full digital sovereignty. “While support for European technological sovereignty is widespread in principle, many businesses are likely to resist fully sovereign AI and data infrastructure if they believe it will increase costs, reduce performance, limit access to innovation or create operational complexity,” says Musiani. She agrees that the concerns expressed by Volvo and Stellantis reflect a tension that prevails across European industry, and that businesses operating in highly competitive global markets will lean towards the best available tech, regardless of where such tech originates.
“If sovereign solutions are perceived as more expensive, less capable or slower to evolve than alternatives offered by major US technology firms, businesses may be reluctant to adopt them exclusively,” she adds. Such reluctance has been reflected by pronouncements from major European firms in recent months, with SAP declaring in a June blog that “fully localized infrastructure may be neither feasible nor desirable,” and that the priority for European companies should be adopting AI as rapidly as realistically possible.
“These firms have a point,” says Meyer, “though proponents of ‘buy European’ would argue that it involves short term pain for long term gain — because driving demand to European tech firms will enable them to gain scale, become more efficient, and then be competitive with their US equivalents.”
A pragmatic, mixed approach
While there would certainly be resistance to imperatives to go fully sovereign, Musiani argues that many enterprises already “favour a pragmatic approach” that balances a sufficient degree of autonomy with commercial performance. “Many European businesses are likely to support measures that reduce dependency on foreign technology while avoiding strict requirements that could undermine competitiveness,” she explains. “Large multinational manufacturers, retailers, logistics firms and technology companies often operate globally and therefore value interoperability, scale and access to world-leading AI ecosystems.”
By spreading usage, such firms are reducing vulnerability to potential kill switches
This account coheres with what many leading European companies are now doing. In recent weeks, the likes of Siemens, Renault, Orange and ChapsVision have affirmed that they use AI models from Europe, the US and China. Siemens, for example, uses China’s DeepSeek and Qwen, as well as several US and European models, while Renault has been working with models from Google, Microsoft, Mistral, DeepSeek and Dataiku. By spreading usage in this way, such firms are reducing their vulnerability to potential ‘kill switches,’ without compromising on capabilities or productivity.
And while there is concern that the EU may be placing too much emphasis on tech sovereignty, Meyers points out that the bloc is not proposing or forcing a ‘buy European’ mandate. As such, the fears recently expressed by certain major corporations may be to some degree overstated.
He says, “The Cloud & AI Development Act only directly impacts purchasing decisions by the public sector, and in almost all cases foreign cloud companies will be able to supply the public sector so long as they prove that they have designed their services and processes so that European interests are protected.”
In disabling Fable 5 and Mythos 5 for all users worldwide, Anthropic has underlined the dangers of not having a sovereign AI strategy. Users in Europe, UK and elsewhere have been shut out of the lab’s latest models, and while negotiations with the US government are ongoing, the withdrawal raises the disturbing possibility that Europe could permanently live under the threat of an AI ‘kill switch.’
