South Korea’s SK Group Chairman Chey Tae-won issued a sobering assessment at the APEC CEO Summit 2025, highlighting a rapidly escalating crisis in the global supply chain for artificial intelligence infrastructure.
The core issue centers on the explosive growth of AI data centers, which now consumes resources at a rate far exceeding the world’s ability to produce essential components. Chips, servers, and advanced cooling systems have emerged as critical chokepoints, threatening to derail the next phase of technological advancement unless decisive, collaborative action is taken.
The Scale of the AI Infrastructure Surge
The expansion of AI-driven computing has reached unprecedented levels. Global data center capacity dedicated to artificial intelligence is on track to triple within the next five years, requiring investments that rival the annual economic output of major nations. Leading technology companies are committing more than a trillion dollars over the coming three years to construct facilities optimized for machine learning and large-scale inference. This surge places extraordinary pressure on every layer of the supply chain, from raw materials to finished systems.
Semiconductor and Memory Constraints at the Core
At the heart of the bottleneck lies the production of advanced semiconductors and specialized memory. High-bandwidth memory, essential for training complex AI models, faces severe capacity limitations despite aggressive expansion by leading manufacturers. Foundry output for cutting-edge AI accelerators remains constrained by wafer availability and process yield challenges. Lead times for these components now extend well beyond a year, creating delays that cascade through the entire ecosystem of server assembly and deployment.
Server and Cooling Systems Under Equal Strain
Beyond silicon, the physical infrastructure required to house and operate AI hardware presents equally daunting obstacles. Server chassis, power supplies, and networking equipment suffer from component shortages that delay assembly by several months. Cooling technology, increasingly vital as processor density rises, has become a particularly acute pain point. Modern AI racks generate heat loads equivalent to dozens of households, rendering traditional air-cooling inadequate. Advanced liquid and immersion cooling solutions face shortages of specialized materials and manufacturing capacity, further complicating large-scale deployments.
Energy Infrastructure Emerges as the Ultimate Limit
The crisis extends far beyond hardware into the realm of power generation and distribution. A single large-scale AI data center can consume a gigawatt of electricity, comparable to the needs of a mid-sized city. Grid operators in key technology hubs report growing risks of capacity shortfalls within the next few years. Transmission constraints and permitting delays for new power infrastructure compound the challenge, creating a situation where computing demand may soon outstrip available energy regardless of hardware availability.
The Perils of Protectionist Responses
Current geopolitical tensions exacerbate these supply challenges through restrictive trade policies and export controls. Measures intended to secure domestic technology advantages often disrupt established global production networks. Critical materials flow across multiple borders before reaching final assembly, making unilateral restrictions counterproductive. Attempts to create fully independent national AI ecosystems risk fragmenting the very supply chains that enable rapid innovation and scale.
A Call for Coordinated Global Action
The solution, as presented at APEC, requires shifting from competition to cooperation on infrastructure development. Standardized efficiency metrics for data centers could optimize resource use across regions. Shared research into next-generation cooling technologies might accelerate breakthroughs currently stalled by duplicated efforts. Coordinated investment in renewable energy microgrids tailored for computing loads could alleviate pressure on existing power systems. Perhaps most crucially, agreements on critical mineral allocation would prevent destructive cycles of hoarding and retaliation.
Industry Investment Signals Long-Term Commitment
Major conglomerates are already moving to address portions of the crisis within their control. SK Group has committed tens of billions toward developing AI-specific power solutions, including advanced battery storage and modular nuclear technologies. Similar initiatives from other global players suggest recognition that the current trajectory is unsustainable without fundamental changes to how computing infrastructure is planned and provisioned.
The Broader Implications for Global Technology Leadership
The outcome of this supply chain crisis will shape the future distribution of technological capability worldwide. Nations and companies that successfully navigate these constraints stand to gain significant advantages in the coming era of advanced artificial intelligence. Conversely, those trapped in prolonged shortages risk falling behind in both economic competitiveness and strategic influence. The window for establishing resilient, collaborative frameworks grows narrower with each passing month of accelerating demand.
Toward a Sustainable AI Infrastructure Paradigm
The warnings delivered at APEC 2025 underscore a fundamental truth about the current technological moment: artificial intelligence has transitioned from software innovation to physical infrastructure challenge. The algorithms may advance rapidly, but their real-world impact now depends on steel, silicon, coolant, and kilowatts. Building the computing foundation for tomorrow’s AI applications demands the same level of international coordination once reserved for space exploration or global telecommunications networks. The choices made in the coming years will determine whether the AI revolution powers broadly shared progress or sputters amid self-imposed limitations.