Artificial Intelligence and Semiconductors—Key Developments Since 2025

Since the publication of the previous CNTR Monitor, several developments have reinforced and, in some areas, intensified these structural dynamics. The past year has been characterized by three overarching trends: the geopoliticization of critical raw materials, new bottlenecks in AI hardware supply chains, and the continued fragmentation of global technology governance. At the same time, the rapid expansion of AI infrastructure has created new strategic vulnerabilities related to energy supply, data center capacity, and the geographic concentration of advanced semiconductor manufacturing.

Critical Minerals as Strategic Leverage

One of the most notable developments is the increasing use of critical minerals as instruments of geopolitical leverage. In recent years, China, which controls a dominant share of global rare earth mining and processing capacity,¹ has expanded export controls and licensing requirements for several rare earth elements and related processing technologies that are essential for semiconductor manufacturing, advanced electronics, and defense systems. In late 2025, a conflict arose that illustrates this impact. The Dutch government intervened in Nexperia, a semiconductor company headquartered in the Netherlands, amid concerns that the firm’s Chinese owner, Wingtech, might transfer strategic technology and production capabilities from Netherlands to China, raising national and European economic security concerns. In response, the Chinese Ministry of Commerce imposed export restrictions on Nexperia chips manufactured or packaged in China, causing almost immediate disruptions to global semiconductor supply chains. The dispute had a direct effect on other countries, including Germany, where many manufacturers rely on Nexperia components for automotive and industrial electronics.

This development illustrates that strategic competition over semiconductor technologies extends well beyond chip fabrication itself. Control over upstream inputs is becoming an increasingly important component of technological power. As a result, supply chain resilience in the semiconductor sector must now also account for vulnerabilities in the extraction, processing, and trade of critical minerals.

Emerging Bottlenecks in AI Hardware

The rapid expansion of AI infrastructure has created new structural bottlenecks in the semiconductor ecosystem. While earlier policy debates focused primarily on the supply of advanced logic chips, the recent surge in AI data center construction has revealed shortages in other critical components. For instance, modern computing systems rely on random-access memory (RAM) as fast, temporary working memory enabling processors to access and manipulate data during computation. The dominant form of RAM today is dynamic random-access memory (DRAM). Artificial intelligence systems require a specialized high-performance variant of this known as high-bandwidth memory (HBM), which provides the exceptionally high data transfer rates needed by AI accelerators. As demand for HBM has surged, it has placed increasing pressure on the broader DRAM ecosystem and manufacturing capacity. These developments have also affected global memory supply chains more generally, including those for NAND flash technologies, which provide the nonvolatile, long-term data storage used, for example, in USB sticks or memory cards. Consequently, tightening conditions in memory markets are increasingly affecting a wide range of digital products, including smartphones, personal computers, storage systems, and industrial electronics. In practice, this means that the strategic competition over AI hardware is now increasingly extending beyond the production of leading-edge processors to encompass memory technologies, advanced packaging, and specialized semiconductor manufacturing processes, with growing repercussions for other IT sectors.

Fragmentation of Technology ­Governance

Another notable development is the increasing volatility of technology governance regimes. Export control frameworks targeting advanced AI chips and semiconductor manufacturing equipment remain a central instrument of geopolitical competition, particularly in the context of the U.S.–China technology rivalry. However, these regulatory frameworks have become more dynamic and politically contested.

Changes to export licensing regimes, tightening restrictions on semiconductor equipment, and more comprehensive investment screening mechanisms have created a policy environment characterized by frequent adjustments and regulatory uncertainty. For governments and companies alike, this complicates long-term planning for semiconductor manufacturing investments, AI infrastructure development, and technology partnerships.

At the same time, concerns have emerged over the effectiveness of Western export controls as a tool for constraining China’s economic AI developments and ambitions. In the United States, in particular, policymakers and analysts have increasingly acknowledged that export restrictions have not fully prevented Chinese firms from advancing their AI capabilities and competing effectively with U.S. companies, raising doubts about the long-term efficacy of their containment-oriented approaches. More broadly, recent debates have centered on whether export controls can sustainably preserve technological advantages or whether they may instead risk accelerating domestic innovation and technological self-sufficiency in the very states they target. Moreover, differences in regulatory approaches among major technology powers have complicated efforts to establish coordinated governance frameworks, increasing the risk of a more fragmented global technology landscape characterized by competing standards, regulatory regimes, and technology ecosystems.

The AI Data Center Boom

Number of data centers (2025)²

Energy consumption

According to the International Energy Agency, the global energy consumption of data centers in 2026—at more than 500 TWh—will be more than double that of 2020 and is expected to reach nearly 1,000 TWh by 2030.³

In 2024, electricity consumption by data centers was estimated at approximately 415 TWh, which equals about 1.5% of global electricity consumption.⁴

Investment plans	$700 billion Alphabet, Microsoft, Meta and Amazon are expected to spend nearly this amount combined in 2026 on their AI build-outs.5	$500 billion The U.S. public-private initiative Stargate Project (in partnership with OpenAI, SoftBank, Oracle, and MGX) plans to invest this amount over 4 years.6
$295 billion has been announced as China’s investment in state-run AI data centers over five years, to create a nationwide “National Computing Network”.7	$30 billion The EU is putting seven large data centers out to bid and plans to invest €10 billion to attract an additional €20 billion in private investment.8	$13.4 billion is the total value of data center projects in Russia at various stages of planning; however, Russia has suspended 38 projects worth $2.26 billion combined.9

How reliable are the numbers?

The funding and investment figures presented here are only directly comparable to a limited extent and should be understood primarily as an indicator of the strategic importance that countries attach to AI technologies. In the global AI race, such announcements are often driven by industrial, economic, and geopolitical considerations. In particular, leading AI nations such as the U.S. and China also use investments strategically to secure and expand existing technological, economic, and infrastructural competitive advantages or positions of dominance. Furthermore, the measures differ significantly in terms of their structure—such as direct government investments, funding programs, and public-private partnerships—as well as their objectives, for example, the expansion of data centers, research funding, or the development of broader AI ecosystems.

Meanwhile, there is evidence to suggest that Chinese companies have responded strategically by stockpiling AI-related hardware and semiconductor components in anticipation of future restrictions. This accumulation of inventories has helped them to mitigate the immediate impact of supply constraints, while reducing the short-term effectiveness of tightening export control measures.

The Rise of AI Compute as Strategic ­Infrastructure

A further development is the growing recognition that AI compute infrastructure is itself becoming a strategic asset. The rapid expansion of large-scale AI systems has led to unprecedented investment in hyperscale data centers, specialized AI computing clusters, high-performance networking technologies, and even submarine cables.

These infrastructures are increasingly comparable to other forms of strategic infrastructure, such as energy systems or telecommunications networks. For states, control over large-scale AI compute infrastructure provides several strategic advantages. It enables the domestic development and deployment of advanced AI systems and reduces dependence on foreign technology providers, thereby strengthening the country’s ability to capture the economic benefits associated with AI innovation. Access to large-scale compute capacity has become an important determinant of technological sovereignty, as it shapes who can develop, deploy, and commercialize frontier AI applications. As a result, compute infrastructure is increasingly viewed not only as an economic asset but also as a strategic capability, prompting governments to pay greater attention not only to semiconductor production but also to the geographic location, ownership, and resilience of AI data center infrastructure. In addition, the use of AI is expanding rapidly throughout society, making access to reliable AI services more and more important for the functioning of public services and digital end-user infrastructures. As a result, disruptions affecting access to AI capabilities or compute infrastructure may have consequences that extend beyond individual firms, increasingly affecting broader societal resilience.

The growth of AI infrastructure is also placing new pressures on energy systems and physical infrastructure. Training and operating advanced AI models requires vast computing resources, which, in turn, depend on large and reliable supplies of electricity. The rapid construction of hyperscale data centers—particularly in the United States, China, and parts of the Middle East—has already begun to affect regional electricity demand and grid planning. This development introduces a new dimension to AI geopolitics: access to abundant, reliable, and affordable energy may become an important determinant of future AI capabilities. In several countries, policymakers are now increasingly linking AI strategy with broader debates on energy security, grid expansion, and industrial electricity pricing.

Conclusion

The geographic concentration of key segments of the semiconductor supply chain remains a major structural vulnerability. In particular, the production of the most advanced chips continues to be heavily concentrated in East Asia, especially Taiwan. While global investment in semiconductor manufacturing capacity has increased significantly, new facilities will take years to reach full operational capacity and are unlikely to fully offset existing concentration risks in the near term.

For Europe, these developments underscore the persistence—and in some areas the deepening—of structural dependencies within the semiconductor ecosystem. Despite policy initiatives aimed at strengthening domestic manufacturing capabilities, Europe remains heavily reliant on external suppliers for both advanced AI chips and critical raw materials. At the same time, European industry is directly exposed to supply disruptions arising from geopolitical tensions and export restrictions in upstream segments of the semiconductor supply chain. These dependencies create not only economic vulnerabilities but also broader security risks, as disruptions to semiconductor supply chains can affect critical industries, digital infrastructures, and key sectors of modern economies. The ability of states to influence access to critical minerals, semiconductor components, manufacturing capacity, or AI infrastructure increasingly creates opportunities for geopolitical leverage.

Taken together, these developments reinforce the central conclusion of the previous CNTR Monitor:¹⁰ the competition over artificial intelligence capabilities is increasingly shaped by control over semiconductor ecosystems and their associated supply chains. This competition spans the entire value chain—from critical minerals and semiconductor manufacturing equipment to advanced chips, memory technologies, energy-intensive AI computing infrastructure, and hyperscale data center capacity. Consequently, resilience within semiconductor ecosystems is becoming an increasingly important element of economic security, technological sovereignty, and the protection of critical infrastructures.

At the same time, some analysts have raised concerns about the sustainability of the current levels of AI-related investment. Should expectations regarding future AI adoption or profitability fail to materialize, a correction in infrastructure investment could follow, with implications for semiconductor demand, data center construction, and capital allocation across the technology sector. Nevertheless, even if current investment dynamics were to slow, the strategic importance of semiconductors and AI-related technologies—and the dependencies associated with them—would likely remain unchanged.