Artificial Intelligence, Semiconductors, and the Chip Wars: Reviewing the Geopolitics of AI in the Military and in International Security

Military forces worldwide are integrating AI into operations, using autonomous drones, robotic systems, and decision-support tools to process vast amounts of battlefield data at unprecedented speeds. In an environment where the tempo of combat is accelerating, AI is seen as essential for maintaining a strategic advantage and necessary tool for electronic warfare, cyber operations, and autonomous defense systems. Currently, the military sector is seeing a rapid influx of AI investment. For example, the U.S. military is advancing AI through key projects like Maven, which uses machine learning to analyze surveillance data for improved targeting; and Replicator, a large-scale initiative to rapidly deploy autonomous systems like drones to counter emerging threats. Other notable efforts include the AI Rapid Capabilities Cell, the Thunderforge Initiative, and AI-driven enhancements to Navy and Air Force operations – all aiming to boost decision-making, readiness, and combat effectiveness. China is similarly accelerating its military AI programs. The Chinese government prioritizes AI as a national security imperative, investing in AI-driven surveillance, cyber warfare, and autonomous weapons. Companies such as Huawei and SenseTime develop AI with both civilian and military applications, highlighting the global competition to use AI in the armed forces.

Unlike traditional military innovations, AI advancements largely originate in the commercial sector. Companies like OpenAI and Google DeepMind drive AI breakthroughs, with rapid spillover effects into defense applications. This trend is reshaping global security strategies as consumer AI advancements influence military capabilities. Because developing and running an AI is much more expensive than current IT cloud applications, a key driver of this trend is the commercial AI sector’s struggle to monetize its technology.¹ Companies like OpenAI have explored subscription-based revenue models but struggle to get a return for their investments through public commercial applications.² Other companies, such as Palantir with its program Gotham,³ or Google’s involvement in the project Thunderforge,⁴ are increasingly positioning themselves as AI suppliers for governments, armed forces, and security agencies. Additionally, startup companies are using this development to promote their products. One example is Anduril Industries – a defense technology company specialized in autonomous systems – and its product Lattice, an off-the-shelf AI-powered operating system that integrates ready-to-deploy sensor networks and defense drones that are autonomously managed for autonomous perimeter monitoring and protection.

While much attention is currently focused on high-density and high-performance cutting-edge semiconductor nodes below 7 nm, chips with larger structures – such as those above the size of 28 nm – remain critically important for military forces worldwide. Many military systems prioritize reliability, radiation tolerance, and long-term availability over sheer computational power. Chips built on such mature nodes are often more robust against cosmic radiation and electronic warfare attacks, making them ideal for satellites, radar systems, and secure communications equipment. Moreover, military platforms frequently require decades of operational life, favoring technologies that are easier to manufacture, maintain, and reproduce without the rapid obsolescence seen in cutting-edge consumer electronics. Thus, ensuring secure and sovereign access to mature node semiconductor technologies is just as vital for national defense as investing in the latest chip innovations.

The Global Semiconductor Backbone

As of 2025, China accounts for approximately 28%⁵ of the global production of semiconductors manufactured at 28nm and larger process nodes, with forecasts suggesting a rise to 39% by 2027.⁶ Other leading producers in this segment include Taiwan (primarily TSMC), the United States (notably GlobalFoundries), and South Korea (Samsung). China’s rapid expansion in mature-node production is largely driven by strategic investment initiatives and the restricted access to cutting-edge semiconductor manufacturing technologies.

Artificial intelligence relies on specialized semiconductor chips that can handle complex, high-speed computations while sustaining continuous, large-scale workloads. At the heart of this ecosystem are graphics processing units (GPUs). Originally developed for graphics rendering and computer gaming, GPUs are now crucial to AI due to their massively parallel architecture. Capable of holding up to thousands of processing cores on a single chip, current GPUs can execute computational operations for deep neural network computations with high efficiency. Equally important are tensor processing units (TPUs), custom-designed by Google, which use systolic arrays, a specialized hardware architecture built for high-throughput, parallel computation, that are useful for many AI operations. Other hardware approaches for AI computation include application-specific integrated circuits (ASICs), which are purpose-built for narrow AI functions such as image recognition or natural language processing. These chips trade flexibility for performance and power efficiency, making them ideal for large-scale AI inference applications in both cloud and edge environments. Some approaches, like neuromorphic chips, even try to model biological neural systems, use neural networks and event-driven processing to reduce energy consumption drastically, offering a potential breakthrough in low-power, always-on AI systems.

Along with this computation-focused hardware, high-bandwidth memory (HBM) is another critical component of AI systems because it enables rapid data transfer between the calculating chips and the memory to temporarily store and exchange results. Unlike traditional technical approaches, HBM stacks memory vertically and connects it with through-silicon vias (TSVs). While it does require correspondingly high-precision production processes, TSVs offer significantly higher bandwidth and energy efficiency, an important trade-off for AI accelerators like GPUs and TPUs.

The production of all these chips relies on a deeply interconnected and highly specialized global supply chain. The United States remains dominant in chip architecture and design, led by companies like NVIDIA, AMD, and Intel, which rely on Electronic Design Automation (EDA) tools developed by Cadence and Synopsys. However, the actual manufacturing of advanced nodes – sub-5nm processes – is concentrated in Taiwan, with TSMC producing over 90% of the world’s most advanced AI chips. These chips are fabricated using extreme ultraviolet (EUV) lithography machines, which are manufactured almost exclusively by ASML in the Netherlands. Without ASML’s machines, which cost upwards of $200 million each, modern chip production would be impossible. But Germany also has its role with optical systems developed and produced by ZEISS, a company that provides high-precision lithography optics that enable the extreme ultraviolet (EUV) lithography processes necessary for manufacturing AI processors.⁷ Despite its leading position at the production of “conventional” chips – 28nm and above – China’s advancement in artificial intelligence development and production has been significantly impeded by comprehensive export restrictions imposed by the United States and its allies.⁸ These measures encompass limitations on the transfer of semiconductor manufacturing equipment, particularly for chips production below 14nm. Furthermore, access to essential EDA software and critical intellectual property has been curtailed, hindering China’s ability to design and produce cutting-edge AI hardware domestically.

The entire semiconductor ecosystem would not work without critical production materials, including rare earth elements, cobalt, and gallium, sourced largely from China, Ukraine, and several African nations, making supply chains vulnerable to geopolitical tensions.⁹ Finally, the back-end packaging and testing, often carried out in Southeast Asia (Malaysia, Vietnam, Philippines), is another vital step. During this stage the semiconductor die (the small piece of silicon that contains the electronic circuits) is enclosed in a protective casing. This shields the chip from environmental factors and establishes the necessary electrical connections between the microchip and the larger electronic system, such as a circuit board. Advanced packaging technologies, like 3D stacking and chiplet integration, are increasingly critical to improve the performance of chips.

AI as a Key Battleground in International Power Competition

These developments have resulted in a situation in which AI has rapidly become a strategic asset in the global struggle for technological dominance, highlighting the national dependencies on the access to cutting-edge semiconductors. This competition affects technological advancements because small performance gains in chip efficiency can significantly reduce costs, enhance capabilities, and reshape the competitive landscape. It also affects access to specific machines, production materials and intellectual property itself.

Against this backdrop, the U.S.-China AI rivalry is intensifying and both states are investing vast amounts of money. For example, the U.S. Stargate Initiative, a $500 billion infrastructure partnership between government and industry, is aiming to build AI data centers across the U.S.¹⁰ China is also heavily investing in AI through major state and private initiatives, including a $138 billion government-backed tech fund¹¹ and a $8.2 billion national AI investment fund to accelerate innovation. Other Chinese tech giants like Alibaba and Tencent are also scaling up, with Alibaba pledging $50 billion over three years and Tencent increasing AI-related capital expenditures to $10.7 billion in 2024.¹²

Besides fostering domestic AI development, both countries are increasingly resorting to other measures to prevent competitors from accessing technology. The U.S. has imposed several export controls on AI chips,¹³ restricting China’s access to critical hardware. In response, China has ramped up domestic semiconductor investments but remains generations behind in AI chip manufacturing. Despite this gap, China dominates the production of larger, non-AI chips, giving it leverage over global supply chains.

Europe faces its own challenges. While firms like ASML provide critical lithography technology, the continent lacks sufficient AI chip manufacturing infrastructure, creating strategic dependencies on the U.S. and Taiwan. Against this background, European policymakers are increasingly seeing AI and semiconductor technology as critical infrastructure and aiming to extend or establish funding for research and production opportunities.¹⁴

Russia is also recognizing AI’s strategic importance. Reports suggest growing AI collaboration between Russia and China to counter U.S. advancements. Russia is investing in AI-powered battlefield automation and cyber operations, signaling AI’s increasing role in future armed conflicts.

A recent example of AI competition volatility was DeepSeek, a Chinese AI model that briefly shook global markets. Its claim to achieve performance levels comparable to Western AI models while using significantly less computing power temporarily dropped NVIDIA’s stock and unsettled U.S. tech circles. If true, this would challenge the assumption that AI superiority depends on massive computational infrastructure. However, skepticism remains. Independent validation is lacking, and some analysts suspect DeepSeek’s announcement was strategically timed to create uncertainty in Western AI markets. Even if only partially valid, the episode highlights the fragile and dynamic character of the AI technology race. Governments and private enterprises heavily invest in computing infrastructure, yet software-driven efficiency breakthroughs like new computational approaches or methods that reduce the size of AI models while maintaining comparable performance could disrupt the current focus on hardware supremacy, weakening export controls as a geopolitical tool. Future AI leadership will depend not just on superior chips, but also on software-driven adaptability.

Further Perspectives

Although not the scope of this chapter, several other aspects will likely influence (or indeed already are influencing) the discussion of AI and its security relevant implications in the near future.

For one, a growing challenge for AI infrastructure is the massive power consumption of AI data centers. Training large models like GPT-4 or custom vision systems can consume energy on the scale of megawatts,¹⁵ equivalent to the needs of a small town. This has led to calls to develop new solutions to power future AI clusters. These AI facilities often require advanced liquid cooling systems, thermal management solutions, and redundant power architectures, further increasing their complexity and environmental impact.

Unlike traditional cloud computing, AI server farms must be highly centralized due to latency and bandwidth demands, creating potential single points of failure. These centralized AI hubs are increasingly viewed through a national security lens – they are vulnerable to cyberattacks, supply chain sabotage, grid outages, and physical strikes. Their dense compute loads also produce high thermal footprints that distinguishes them from regular data centers.¹⁶ While this introduces new vulnerabilities, in terms of future arms control it may also be used as a basis for verification measures.

Finally, also the global demand for submarine cable infrastructure is rising sharply, driven by the data needs of artificial intelligence applications. AI systems, particularly large-scale model training, require vast bandwidth and low-latency connections. Submarine cables provide this, reinforcing their already critical role in supporting global internet traffic.¹⁷ U.S. tech giants such as Google, Meta, Microsoft, and Amazon now dominate this sector, owning or heavily investing in a significant share of global submarine cable capacity.¹⁸ Projects like Meta’s Project Waterworth, aiming to build the world’s longest undersea cable, illustrate how these companies are strengthening their global AI infrastructure.¹⁹ While this ownership secures their data transfer needs, it also raises questions about digital sovereignty and centralized control over critical communications infrastructure.²⁰

Policy Recommendations for Decision-­Makers

The dependencies on AI and the semiconductor race require proactive political decisions. Although Germany and Europe currently play a minor role in the global semiconductor supply chain, some bargaining chips exist. The following recommendations outline key steps to remain at the technological forefront, maintaining supply chain stability, and navigating geopolitical AI challenges.

Adapting to U.S.-China AI competition through domestic semiconductor production. With the growing tensions and competition between U.S. and China over semiconductors and AI, Europe can significantly reduce dependencies by further expanding domestic semiconductor research, development and manufacturing. Some important steps have already been taken by programs like the EU AI Continent Action Plan²¹ or the EU Chips Act,²² but these alone will not be sufficient – a criticism voiced recently by the European Court of Auditors.²³ As the current Trump administration aims to gain and sustain AI world leadership²⁴ by – among other measures – trying to bring semiconductor production facilities to the U.S.,²⁵ its role as reliable future supplier is questionable. China too is indirectly restricting the access to semiconductor production capabilities by extending export control measures to rare earth minerals²⁶ that are – among other high-tech use cases – needed to manufacture chip production machinery. So, while no formal export controls on semiconductors have been announced, these measures can be seen as part of a strategic consolidation of the semiconductor sector in China as it aims to politically leverage its capabilities in this area. Nevertheless, given that the establishment of domestic capabilities in Europe is a mid- to long term project, strategically using the existing interdependencies of valuable European companies like ASML or ZEISS can help to stabilize current supply chains.

Protect European semiconductor and AI assets. Given the highlighted importance of European companies like ASML or ZEISS, appropriate measures should be used or put into place to protect these assets. For instance, at the request from the U.S. government, ASML halted shipments of advanced chip-making equipment to China in January 2024. While ASML officially attributed the decision to changes in export license requirements, reports suggest that U.S. pressure played a significant role in this action.²⁷ Additionally, in January 2025, the Dutch government aligned its export control policies with those of the U.S., requiring ASML to obtain licenses from the Dutch government for certain chip-making tools. This move effectively tightened restrictions on ASML’s exports to China, reflecting the influence of U.S. policy on its allies.²⁸ Such influence, be it via political pressure or economic influence, should be prevented by clear political statements and potential protective laws. To maintain a leading position within the global semiconductor supply chains, coordinated measures among European countries to deliberately control the proliferation of technology should be considered.

Recognizing AI and Semiconductors as Critical Infrastructure. Production facilities for semiconductors, necessary machinery, or the processing of production materials should be recognized as part of national critical infrastructures. As Germany and the EU have already recognized semiconductors and AI as key technologies of strategic security importance,²⁹ existing research, development and production facilities need to implement appropriate protection measures, especially – but not exclusively – in terms of cybersecurity to protect infrastructure from cyberattacks and espionage. Europe and Germany have developed strong guidelines, legal obligations, and practical recommendations for this approach through their regulation frameworks for critical infrastructure. Semiconductor production as well as AI server systems should be explicitly included into these frameworks, to recognize their importance and the potentially widespread and massive impact that damage to or failure of these systems could have on European societies.

Given the current pace of AI technology development, its security-related dependencies, and its already emerging impact on global politics and international tensions, decisions are needed to establish, keep, and sustain Germany and Europe as stakeholders in these processes.