Why Europe’s Efforts to Gain AI Autonomy Might Be Too Little Too Late

German Chancellor Olaf Scholz and Microsoft president Brad Smith attend Microsoft’s press conference on the Group’s investments in the AI sector in Germany on Feb. 15, 2024.

This week Microsoft announced that it would invest €3.2 billion ($3.5 billion) in Germany over the next two years. The U.S. tech giant will use the money to double the capacity of its artificial intelligence and data center infrastructure in Germany and expand its training programmes, according to Microsoft vice chair and president Brad Smith.

The move follows a similar announcement from November 2023, when Microsoft said it would invest £2.5 billion ($3.2 billion) in infrastructure in the U.K. over the next three years.

Both countries hailed the investments as significant steps that would permit them to compete on the world stage when it comes to AI. However, the investments are dwarfed by investments made by U.S.-based cloud service providers elsewhere, particularly in the U.S. As AI becomes increasingly economically and militarily important, governments are taking steps to ensure they have control over the technology that they depend on. However, some experts warn that smaller economies may be better off sharing AI power with those that are already leading, than trying to compete.

The rise of AI nationalism

In 2018, Ian Hogarth, who was a tech investor at the time, published an influential essay titled ‘AI Nationalism,’ in which he argued that as AI becomes increasingly powerful and its economic and military importance grows, governments will take steps to promote their domestic AI industries. (Hogarth has since played a key role in the U.K.’s push to lead the world in AI safety regulation).

Hogarth’s essay has proved prescient. Many countries have increased public funding for AI research and commercialization, and the urge to protect domestic companies has shaped regulation. For example, Germany and France led a last minute push—that ultimately proved unsuccessful—to soften crucial aspects of the continent’s landmark AI regulation, due to fears that it could hamper their domestic AI champions.

But while the essay mentions computational power, or “compute,” as one of the AI factors of production that governments could and should try to influence, it didn’t predict how central compute would become to AI progress and policy. Many of the most significant AI nationalism-associated actions taken by governments have aimed to secure their access to compute, or deprive rivals’ access.

For example, in 2022, the U.S. imposed export restrictions to prevent countries such as China from accessing the most advanced semiconductor chips and the equipment required to make them. It updated the restrictions in 2023, and in January 2024 the Department of Commerce proposed a rule that would require cloud compute providers to monitor who was using their services and block access to anyone found to be using them for cyberattacks.

Countries including the U.S. and China have offered subsidies to promote domestic semiconductor chip production. Many countries have expressed their intention to build government-owned compute clusters. Finally, governments have been trying to incentivize the large cloud compute providers by ensuring adequate power supplies are available and removing delays associated with planning permission, all of which are headquartered in the U.S. and China, to build more data centers in their territories.

Size matters

In a statement accompanying Microsoft’s announcement that it would be investing in the U.K., Prime Minister Rishi Sunak said that the announcement was a “turning point for the future of AI infrastructure and development in the U.K.”

But of the roughly $120 billion that research firm SemiAnalysis predicts Microsoft will spend on data centers and other AI infrastructure over the next two years, the investments in Germany and the U.K. represent less than 3%. The majority of Microsoft’s funding will go on investments in the U.S., according to Dylan Patel, chief analyst at SemiAnalysis. U.S.-based companies prefer to build data centers domestically, primarily because of the country’s ample supply of renewable energy backed up by natural gas, and looser permitting restrictions, says Patel.

Efforts to ensure access to sovereign compute by building state-controlled compute clusters are even more futile. While the U.K. plans to invest £900 million in a government compute cluster, this figure pales in comparison to the tens of billions of dollars U.S. cloud providers are spending each year on AI infrastructure.

“A lot of places probably can’t afford to build the scale of compute necessary to do the sorts of projects that they want to do,” says Robert Trager, director of the Oxford Martin AI Governance Initiative. “It may just be the more sensible thing for those countries to train models on cloud systems that are physically based elsewhere.”

States often rely on others for strategically important technologies. For example, U.K. spy agencies decided to store their top-secret information with Amazon Web Services, despite the company being U.S.-based, reportedly because they could not find a U.K. provider with the “scale or capabilities needed.” More broadly, many countries decide not to build their own nuclear weapons, and instead rely upon security guarantees from nuclear armed states.

Given it may be practically infeasible for some countries to keep up, power sharing may be required to ensure some countries don’t fall behind, says Trager. “We need multinational governance frameworks that give many stakeholders a voice,” he says. “Without that we’re going to see competition over these resources that do influence power in the world.”

Write to Will Henshall at [email protected].