Trapped in the Chip War: Europe’s AI ambitions at risk

The AI revolution may appear ethereal and software-driven, epitomized by Large Language Models (LLMs). Yet its foundations are profoundly material, resting on powerful data centers, cutting-edge AI chips, and critical minerals — all embedded in fragile supply chains that are anything but virtual.

A defining feature of these supply chains is their geostrategic rupture between the world’s two greatest powers. The United States dominates advanced AI GPU design and production, while China controls the supply and processing of rare earth elements (REEs). Both increasingly weaponize their leverage in the escalating „chip war” to outcompete the other in the AI race.

The collateral — but also deliberate — damage falls on Europe, which finds itself trapped in a deepening two-level dependency. This comes at a time when the continent is mobilizing resources to close the gap in AI development by heavily investing in advanced, large-scale hardware infrastructure. At its core lie AI factories: massive data and compute facilities that provide the processing power and storage needed to train, test, and scale Large Language Models (LLMs) and other AI systems.

Today, the US and China dominate the global AI infrastructure landscape, jointly operating more than 90% of all AI factories. The implication is clear: Europe lacks the scale and capacity needed to develop and deploy its own AI models at a competitive pace.

For this reason, the EU and its member states have stepped up investments in AI factories, with at least 15 facilities expected to be operational by the end of next year. At the center of this push is the EU’s flagship plan to build up to five AI Gigafactories — each equipped with around 100,000 AI chips and costing between €3 and €5 billion — several times more than the capacity and expense of existing facilities.

Yet these ambitious efforts risk being severely undermined by the growing weaponization of Europe’s strategic dependencies from both East and West. Mutually reinforcing Chinese and American protectionist measures targeting access to the most advanced AI chips and REEs could delay and complicate the rollout of EU AI infrastructure — leaving Europe even further behind in the global AI race.

First level dependency: US AI chips

Europe’s first, and most visible, dependency lies in its near-total reliance on American AI chips. AI factories are, in essence, clusters of thousands of these processors. The most widely used among them are AI GPUs (Graphics Processing Units), specialized chips designed to handle the massively parallel computations required to train and run advanced AI models.

It is the US-based company NVIDIA that dominates the market, controlling an estimated 80–90% of global AI GPU sales. Unsurprisingly, the company produces the world’s most advanced processors, like NVIDIA Blackwell Ultra GPUs, which are considered indispensable for training the highest-performing LLMs. As a result, the GPUs powering Europe’s future AI Gigafactories will almost certainly come from NVIDIA.

Therefore, the fast completion and competitiveness of EU factories are conditioned on an uninterrupted and abundant supply of the most advanced NVIDIA GPUs from the US. Yet there are plenty of reasons why Europe should expect shortages and delays in their provision.

History has already shown how vulnerable this supply can be. In its final days, the Biden administration established the Framework for Artificial Intelligence Diffusion, imposing export controls on advanced AI chips through a three-tier country system. Although the policy was aimed primarily at US adversaries such as China, Russia, and their closest allies, many EU member states in Central and Eastern Europe would also have faced controlled access.

Eventually, the new Trump administration revoked the framework while simultaneously negotiating the EU’s intention to purchase at least $40 billion worth of US AI chips for its computing centers. The move underscores Washington’s interest in maintaining Europe’s dependency: it reinforces NVIDIA’s market dominance while preserving another strategic lever over the EU.

Nevertheless, the US willingness to freely export these chips to the EU may end the moment domestic AI developers face supply shortages. Soaring global demand for AI GPUs is already projected to outpace production capacity within the next year. In such a scenario, Washington could be tempted to compel NVIDIA to prioritize domestic consumers in its deliveries.

The AI GAIN Act, currently under consideration in Congress, shows that this could happen much sooner than expected. The law would require chip manufacturers to fulfill orders from American customers before exporting advanced AI chips abroad. Such a measure could force Europe to queue even longer for its data center orders, giving the US an even greater advantage than it enjoys today.

Designed in the US, made in Taiwan, built with Dutch machinery

What adds a critical layer of vulnerability to AI chip production is its globalized nature. Although designed and owned by NVIDIA, the physical manufacturing of the most advanced GPUs is outsourced to Taiwan’s TSMC (Taiwan Semiconductor Manufacturing Company), which holds a near-monopoly on their fabrication.

Here lies a surprising source of European leverage: TSMC relies on Dutch-made photolithography machines produced by ASML. The Eindhoven-based firm is the world’s sole supplier of extreme ultraviolet lithography equipment, indispensable for manufacturing the most advanced chips.

In effect, the global division of labor for cutting-edge AI chips is dominated by the „big three”: American NVIDIA in design, Taiwan’s TSMC in manufacturing, and Dutch ASML in equipment. Yet this is only the pinnacle of a much broader supply chain — one that ultimately traces back to the raw materials on which these firms depend.

Chief among them are rare earth elements (REEs), a group of 17 metals with unique properties that make them essential for a wide range of modern technologies, from green energy to defense. They are also critical technological enablers in the AI sector, underpinning some of the most advanced and transformative properties of AI chips.

Second level dependence: Chinese AI rare earths

This is where the second, less visible — one could say literally underground — dependency comes into view. Over the past decades, China has steadily consolidated its dominance over global REE supply chains. Today, the PRC controls around 70 percent of rare earth mining as well as 90 percent of separation and processing.

In recent years, Beijing has increasingly weaponized its near-total global monopoly over these materials. In 2020, it enacted the Export Control Law, establishing a legal framework to restrict the export of REEs and other materials under the pretext of safeguarding „national security” in dual-use technologies. This move was widely interpreted as a warning to the US, which at the time had begun sanctioning the Chinese chip industry.

In late 2023, China announced a ban on the export of rare earth extraction and separation technologies. Yet 2025 witnessed the largest escalation to date, marked by two consecutive rounds of restrictions. The first came in April, when Beijing imposed export controls on seven of the 17 REEs (samarium, gadolinium, terbium, dysprosium, lutetium, scandium, and yttrium) introducing monthly export quotas and a licensing system. This resulted in a sharp 74% year-over-year decline in exports in May, along with significant delays. The second blow came in this October, when five additional REEs (holmium, erbium, thulium, europium, and ytterbium) were added to the export control list.

For now, however, the current restrictions pose the most immediate threat to the Western defense sector rather than the AI chip industry, as they could significantly hinder rearmament efforts. The AI chip sector is likely to absorb the shock, at least in the short term, thanks to its existing REE stockpiles, diversified supply chains, and the fact that China’s measures exclude the remaining REEs.

In response to concerns over China’s new restrictions, Taiwan’s Ministry of Economy stated that no significant impact is expected on the country’s semiconductor industry. It emphasized that the restricted REEs differ from those used in chip manufacturing, and that the necessary products or rare-earth derivatives are mainly sourced from Europe, the US, and Japan.

However comforting this may sound, in practice it overlooks the fact that many of TSMC’s suppliers are, in one way or another, ultimately dependent on Chinese REE feedstock — as no alternative source of comparable scale exists globally. The current Taiwanese optimism may soon be tested by licensing backlogs and prolonged review times faced by its suppliers. The conclusion is clear: even if not directly, TSMC is already indirectly vulnerable to the risks posed by Chinese restrictions.

In this context, one of the most important actors to watch is, of course, ASML, which is far more directly exposed. While the Dutch company says it has secured enough raw materials to meet short-term demand from its clients, it already anticipates several weeks of shipment delays from China. Should these delays persist and deplete its reserves over the coming months, ASML’s production capacity will almost certainly take a hit.

This exemplifies that as long as China dominates the mining and processing of REEs, AI chip supply chains — no matter how diversified or seemingly insulated — will ultimately trace back to Beijing, giving it powerful leverage over the sector.

What will happen next?

The consequences of China’s REE restrictions have yet to fully materialize in the coming months. Much will depend on Beijing’s approach to licensing and review procedures, which can either shorten or prolong delays. It is important to remember that export controls are a political and bargaining instrument in the hands of the Chinese government.

For the same reason, the magnitude of their impact will also depend on Washington’s response, which remains uncertain. Shortly after China’s announcement, Trump threatened on his Truth Social platform to impose an additional 100% tariff on all Chinese imports and to introduce export controls on „any and all critical software” by November 1 — the very same date when China’s new restrictions are set to take effect.

But most importantly, just before this deadline, Trump and Xi Jinping are expected to meet in South Korea. This will offer the best, and likely the last, opportunity to negotiate a settlement that could stabilize AI chip supply chains and give producers temporary respite, similar to the May trade truce that reduced tariffs. It is therefore possible that both the Chinese restrictions and US threats are part of their bargaining positions, a way of showing their cards ahead of last-minute negotiations. 

But a diplomatic détente in the chip war does not preclude an AI chip undersupply in the face of skyrocketing global demand. In the coming months or years, the US may still be pressured to restrict exports, which could in turn trigger retaliatory measures from Beijing. Therefore, the problem is ultimately structural rather than merely diplomatic.

Moreover, on the Chinese side, there is a clear and accelerating escalatory trend in REE restrictions — one that could be tightened further if Beijing wishes. First, China can still expand its controls to the remaining five REEs, including light rare earths such as praseodymium and neodymium, which are indispensable for magnet-based precision manufacturing of AI equipment. Second, existing measures can always be strengthened: quotas lowered, licensing tightened, or absolute bans introduced. As seen, Beijing’s repertoire still contains several powerful, unused tools. 

There is a clear self-reinforcing dynamic in this „chip war”: further US restrictions on AI chips provoke Chinese retaliation through tighter controls on AI-critical REEs, and vice versa. Crucially, the biggest impact will not fall on their competing domestic AI developers, but on the rest of the world, which will face sharply reduced access to both the world’s most advanced AI technologies and the materials needed to produce them independently.

EU with (almost) tied hands

What should have already raised alarm bells in the EU is the fact that European countries stand by the escalating chip war with their hands tied, lacking real agency — much like the rest of the world. Even more troubling is that there is little that can be done in the short or even medium term. The EU’s current two-level dependency in the AI chip sector was not built in months or years, but over decades. By the same token, its reconstruction will take at best several years and, at worst, several decades — if it happens at all. By that time, the AI wave will likely have already passed, along with its winners.

Paradoxically, albeit not implausibly, what could offer Europe the opportunity to build lagging computing capacities could be the burst of the „AI bubble” in the US, which many experts increasingly suspect given the mind-boggling pace of demand from American AI developers. Ironically, such a downturn could work to Europe’s advantage: being less exposed to the bubble, Europe could benefit from a sudden loosening of global chip supply and falling prices, speeding up its deployment of AI factories while catching up to a slowed-down US.

Another factor that could tilt the balance in Europe’s favour is if AI development itself were to hit a ceiling — a possibility now gaining traction among analysts. In such a scenario, even with greater computing power and additional chips, the pace of AI progress could stagnate rather than accelerate as expected. These diminishing marginal returns in AI development encountered by Americans could offer Europe the breathing room it needs to narrow the gap.

Yet relying on such external shocks as Europe’s miracle remedy is, to say the least, an irresponsible gamble. Betting on an American market correction or a technological slowdown may offer temporary comfort, but it is neither a realistic nor a safe strategy.

It is indeed a grim picture when we realize that the EU has both the financial and human capital to compete with China and the United States. The problem is that the material means to do so must first be bought from them. Logically, the demands of EU AI factories — and their gigafactory counterparts — will, at best, be treated as a low priority in Washington and Beijing, and at worst, become the first target if strategic pressure is needed.

And that is the first thing Brussels and other European capitals must keep in mind: Europe will inevitably suffer the consequences of its two-level dependency, shaped by both deliberate and unintended Sino-American actions.

For the time being, Europe must learn to play the cards it still holds — no matter how scarce they may seem — while laying the foundations for greater medium- and long-term resilience of its sovereign AI chip industry.

One, but arguably very powerful, strategic asset the EU possesses is ASML. Accordingly, the EU must make it an absolute priority to preserve ASML’s status as the world’s sole supplier of extreme ultraviolet lithography machines, further cementing its symbiosis with TSMC and American partners.

In this way, Europe would retain control over one of the few critical bottlenecks in the global supply chain for advanced AI chips. This leverage would enable the EU, when confronted with restricted access or broader economic coercion, to respond by imposing export or maintenance controls on ASML’s most sophisticated machinery — slowing chip fabrication and raising the cost of data center expansion.

In the medium-term, the EU aims by 2030 to secure 10% of its annual needs for extraction, 40% for processing, and 25% for recycling of critical raw materials, as outlined in the European Critical Raw Materials Act. These still modest targets — and the five long years ahead — have already been widely criticized as too little, too late. Yet they also illustrate how difficult it will be to disentangle Europe from its current dependency on Chinese REEs and other critical materials.

In the meantime, the EU plans to create strategic rare earth stockpiles to prepare for supply chain disruptions and larger shocks that are likely to come. In doing so, it will be critical to ensure that these reserves are sufficient to meet the needs of critical industries during a crisis — ideally for several months or more.

In the long term, one promising avenue for reducing these dependencies lies in investing in R&D to develop viable synthetic alternatives to REEs and bring their production costs down to competitive levels. Such efforts would align with the EU’s broader competitiveness and environmental agenda, given the severe ecological impact of REE extraction and processing. Ultimately, a technological leap forward may be the only sustainable way to break free from Europe’s two-level dependency in the AI sector.

That said, while thinking about the more distant future, EU capitals must first focus on minimizing potential damage to their lagging AI sector now, while securing the few remaining advantages they still hold. Only then will thinking about the long term make any sense.