The entire AI value chain (also known as the AI technology stack or life cycle), from data and algorithms to computing infrastructure required for training and deployment, is critically dependent on semiconductors. Different kinds of chips like CPUs, GPUs, FPGAs and specialised ASICs, form the substrate that enables the creation and operation of AI systems. As AI systems become more sophisticated and ubiquitous, efforts to create robust governance frameworks to ensure their safe, ethical and responsible development and deployment have emerged and accelerated. Multilateral efforts like the OECD’s AI Principles and the Global Partnership on AI, etc., are important initiatives. However, any serious effort to govern AI must also grapple with the complex geopolitical and geo-economic dynamics of semiconductors.
Last week, the US Department of Commerce announced its intentions to review concerns surrounding China’s access to RISC-V, an open-source instruction set architecture standard popular with chipmakers and designers developing chips for multiple use cases, including AI workloads. That China is exploiting the open, collaborative nature of RISC-V development to advance its own semiconductor industry and AI capabilities, potentially eroding US leadership and posing national security risks is a concern echoed by many US policymakers. This is just the latest development at the centre of a high-stakes geopolitical battle for technological supremacy; how the US, China, EU, Japan, South Korea, India, and other countries seek to shape the evolution of this critical industry and control or gain access to chips will have profound implications for the trajectory of global AI development and governance.
The Semiconductor Global Value Chain: Complex Interdependencies and Chokepoints
The semiconductor global value chain (GVC) is highly complex and globalised, with design, fabrication, packaging and testing often done in different countries. However, it is also heavily concentrated: over 90% of the world’s most advanced (<10nm) chips are manufactured in Taiwan by TSMC and South Korea by Samsung. The US still leads in chip design and semiconductor manufacturing equipment. This geographic concentration creates major risks and vulnerabilities because disruptions or restrictions at certain points can have cascading effects across the entire AI compute hardware ecosystem. This also means that countries and firms that can control these points in the GVC can utilise them as leverage in geopolitical and geoeconomic manoeuvres. Case in point: the US’ stringent export controls targeting advanced chips and chipmaking equipment have restricted China’s ability to harness high-performance computing. China, therefore, must work to build up its capacity in almost every stage of the semiconductor GVC and has begun undertaking massive efforts to accelerate its indigenous semiconductor ecosystem in a bid to achieve greater self-sufficiency. Other major powers like the EU, Japan, and India are also ramping up investments and incentives to localise more parts of the GVC and reduce external dependencies.
Considerations Affecting AI Governance
The semiconductor chokepoint in the AI value chain creates both challenges and opportunities for global efforts towards AI governance. The industry’s high capital intensity and economies of scale lead to concentration and the rise of dominant firms with outsized influence. This raises risks around resilience, security, and single points of failure. Aside from market concentration risks like these, there are a few ways in which nation-states’ AI governance efforts could be influenced by access/restriction to semiconductors and computing power.
Firstly, countries with access to advanced chips can push the boundaries of AI research and development, while those facing restrictions may struggle to keep pace. This technological disparity can affect a nation’s ability to shape AI governance frameworks and influence global standards that are relevant to its social, economic, and cultural context. Advanced chips like GPUs and TPUs are critical for training large AI models that drive cutting-edge research. Nations with limited access to these chips will face challenges in developing state-of-the-art AI capabilities. As AI governance norms and standards emerge from real-world use cases and applications, countries at the forefront of AI research and development will have a significant influence in setting those standards based on their experiences and priorities. In contrast, nations lagging in AI capabilities due to semiconductor access restrictions may find it difficult to ensure their social, cultural and economic interests are adequately represented in global AI governance frameworks
Second, nations with a strong semiconductor sector and access to advanced chips can gain a competitive advantage in downstream AI-driven industries. Conversely, countries facing restrictions will not have a similar advantage as homegrown alternatives may end up losing out to foreign competitors, whose governance priorities may not be in alignment.
Third, as mentioned earlier, control over advanced semiconductors has become a geopolitical issue amidst export restrictions. Export controls and trade restrictions can be blunt instruments that cause collateral damage, erode international trust, and hamper global multilateral AI governance efforts. While AI models and algorithms are more difficult to regulate due to their intangible nature, chips provide a tangible point of control. Governments can track and restrict access to chips more easily than software, enabling them to enforce AI governance measures that align with their values. However, as more intrusive options like the recently proposed on-chip governance mechanisms (or hardware-enabled governance mechanisms) are tabled, their implications on sovereignty and ownership rights amplify strain on diplomatic relations and hinder international cooperation on AI governance. The proposals seek to ensure that the design of advanced chips used in AI workloads incorporates elements that allow the monitoring of the workload being run on them. Depending on whether they exceed performance or touch use-case thresholds, the chips can lower their performance. Amidst this trust deficit, multilateral efforts may take a backseat as nations pursue their own interests rather than working towards unified global frameworks.
Fourth, countries with easier access to advanced chips, and thriving downstream industries may adopt more permissive regulations to foster innovation while those facing restrictions may end up implementing stricter controls to mitigate perceived risks, making a shared understanding of regulatory principles in multilateral initiatives for AI governance more difficult to achieve.
Lastly, restrictions on semiconductor access can limit the flow of talent and knowledge across borders, hindering international collaboration on AI research and governance efforts. This can slow down the development and adoption of best practices and standards for responsible AI development. AI startups and research organizations are already struggling to compete for top talent against tech giants which can lure recruits with the promise of vastly more computing power.
Any effective global AI governance framework must account for the myriad geopolitical and economic dynamics of semiconductors in a world critically dependent on chips. This requires that multilateral governance initiatives like the GPAI or OECD attempt to stymie unilateral actions that fragment the global semiconductor ecosystem. Policymakers will need to navigate this complex landscape with both technical savvy and diplomatic finesse in order to avoid a zero-sum “AI arms race” from taking hold.
Satya S Sahu is a Research Analyst with the High-Tech Geopolitics Programme at the Takshashila Institution, Bangalore. His areas of interest are semiconductor geopolitics, AI governance, and technology policy.
