Europe Cannot Avoid an AI Reckoning

These deficiencies can no longer be ignored, because the goal of US policy is shifting from “managing China” to “outpacing everyone” globally. Alongside an export-control regime, the latest 25% tariff on selected advanced AI chips under the Trump administration aims to pull more investment in high-end semiconductor production back onshore, making US domestic fabrication more competitive and accelerating the development of its own AI infrastructure.

The tariff is one expression of a broader AI strategy that has been taking shape for years and that discards three longstanding assumptions: that the United States would privilege market efficiency over industrial policy; that China would import compute capacity rather than build it; and that Europe could regulate the industry without building sovereign capabilities of its own.

The US has moved well beyond the belief that markets will optimize supply chains. It is layering export controls on top of subsidies, tax incentives, and procurement policies to reshape where chips are designed, fabricated, and deployed. At the same time, China has been rolling out domestic “AI accelerators” (new chips), expanding fabrication capacity, and tying AI infrastructure to its overseas lending and economic diplomacy. Europe, by contrast, has treated AI mainly as an afterthought, refining legal definitions while continuing to rely on foreign cloud capacity, chips, and models.

Having entered the AI era over-regulated and under-industrialized, Europe imports the vast majority of its advanced semiconductors, pays materially higher industrial electricity prices than the US, and still relies on American cloud providers for the bulk of its compute. If this dangerous dependence was not obvious before, it has become harder to ignore amid explicit US threats to assert control over sovereign territory belonging to a longstanding European ally. 

Europe finds itself squeezed between an aggressive revisionist power (Russia) that is already probing its defenses and a US administration that is prepared to weaponize its industrial, infrastructural, and trade ties with the continent. If the US were to leverage access to AI and advanced compute coercively, the effects could be immediate: European defense networks, intelligence systems, hospitals, financial markets, and industrial firms could face sudden restrictions on critical cloud services, with few domestic alternatives. In this scenario, the Kremlin would have an opportunity to escalate its hybrid war against Europe, knowing that the continent is digitally exposed and politically constrained.

Given such risks, Europe must move beyond its focus on regulatory excellence, risk classifications, and compliance systems. Unless it makes serious strides toward building the physical and financial infrastructure that a domestic European AI industry will need, these preoccupations will be more of a liability than an asset. Specifically, European policymakers must support the creation of vast compute clusters, ensure cheap, reliable electricity, and commit to sustained capital expenditures in strategic sectors.

The bad news is that Europe can’t change course overnight. A single cutting-edge data center can easily cost more than €1 billion ($1.2 billion) and consume more power than a midsize European city, and a state-of-the-art fab (chip production facility) now requires more than €20 billion in upfront capital expenditures. Yet Europe’s energy prices are already elevated, its venture capital markets are shallow, its cloud infrastructure is dominated by foreign providers, and its semiconductor goals remain largely aspirational. One recent analysis estimates that roughly €3 trillion in investment will be needed over the next five years to upgrade Europe’s AI industry.

The good news is that Europe is not starting from scratch. It controls several crucial choke-point technologies. For example, the Dutch company ASML holds a monopoly over extreme ultraviolet lithography, with its machines underpinning TSMC and Samsung’s most advanced production lines. Similarly, German and Dutch suppliers such as Zeiss (optics) and Trumpf (high-power lasers) occupy important strategic niches in the AI production chain. Together, these domestic nodes give the European Union the means to anchor parts of the global AI hardware stack in Europe.

After compute, capital is Europe’s scarcest input in the AI race. But unlike compute, it can migrate quickly in response to policy signals and incentives. Although Europe lags in overall technology and AI funding – US AI companies attracted roughly $47 billion in 2024, compared to about $11 billion for European firms – it produced more high-tech startups than the US between 2019 and 2024 (albeit with a deal value of only $62 billion in 2024, compared to $209 billion in the US).

Moreover, AI-specific funding has already grown from a low base, with European firms raising nearly €3 billion across 137 deals in 2024, about 35% more than the previous year. Venture-capital investments in European defense and security technologies have climbed to record levels, reflecting a reassessment of the continent’s strategic industries. This momentum partly reflects a gradual shift in private-capital allocation toward Europe, as investors respond to US policy uncertainty and seek longer-term exposure to Europe’s strategic infrastructure and industrial assets. 

The EU must start playing hard ball, too. That means leveraging Europe’s market power by conditioning market access, procurement, and regulatory approvals on concrete local commitments – local packaging, data-center build-out, assembly, or research and development – much as the US has done under the CHIPS and Science Act with TSMC in Arizona. Europe must also mobilize long-term capital through public guarantees and blended finance, so that pension funds, insurers, and sovereign vehicles can underwrite fabs and compute clusters (investments that venture capital will hardly touch).

Finally, Europe must treat energy, compute, and data-center deployment as a single planning challenge, rather than as three separate issues. Prices, permitting, and infrastructure all must be aligned so that fabs and compute clusters have predictable power, off-take agreements, and siting. Fortunately, ensuring such consistency is well within Europe’s institutional reach.

The lesson from recent US policy changes is not that Europe should deregulate, but that regulation without hardware, compute, and capital leaves it dangerously exposed in an increasingly dog-eat-dog world. Europe can still catch the train, but only if it starts building the capacity that makes regulation meaningful.