As AI technology evolves rapidly, whether safety governance can keep pace with technological development has become a major concern for industries worldwide. Anthropic CEO Dario Amodei recently urged the global AI industry to slow the pace of capability gains in frontier models. OpenAI CEO Sam Altman and xAI founder Elon Musk publicly echoed his call. Tai-Chin Chiang, Director of the Japan Center Sub Committee at the Chung-Hua Institution for Economic Research (CIER), said rapid advances in AI are creating multiple challenges related to safety, energy, and supply chains that warrant continued attention.
AI Leaders Warn That Frontier Models Are Advancing Too Quickly for Safety Validation to Keep Pace
In his article We Must Pace the Frontier, Amodei argued that AI capabilities for self-improvement and autonomous agency have advanced rapidly, while safety alignment and external validation may not be keeping pace. He therefore called for moderating the rate of capability gains in frontier models to allow more time to strengthen safety mechanisms and third-party evaluations. This does not mean halting AI training or development. Rather, the objective is to maintain an appropriate balance between technological progress and risk management.
Chiang said rapid gains in computational efficiency also demonstrate that AI is developing at a fundamentally different pace than in the past. He noted that processing one token required approximately 42,000 joules in 2014. By 2024, after NVIDIA introduced its Blackwell architecture, that figure had fallen to approximately 0.4 joules. Semiconductor process nodes once took much longer to advance, but industry cycles have now shortened dramatically. The speed of development has itself become a source of risk.
Sharp Decline in Token Energy Consumption and Shorter Process Cycles Drive AI-Related Energy and Supply Chain Demand
The impact of rapidly expanding AI demand has extended beyond semiconductors to the broader industrial supply chain. Chiang said massive computing requirements are first driving up electricity demand, which in turn is increasing demand for power-related products such as fuses, rectifiers, and passive components. As computing density rises, thermal management is becoming a critical bottleneck and accelerating the development of related technologies. Faster data transmission is also fueling demand for silicon photonics, while advanced packaging faces challenges such as higher costs and greater process complexity. In other words, AI is not driving growth in a single industry. Its effects span energy, electronic components, thermal management, optical communications, and advanced packaging, creating a highly interconnected industrial ecosystem.
Chiang indicated that AI is already having a far-reaching impact on industry and everyday life, making the term “Fourth Industrial Revolution” an appropriate description. However, the opportunities and risks associated with AI are expanding in tandem. In addition to the technology itself, energy supply, cybersecurity, supply chain resilience, and governance mechanisms must all be upgraded accordingly.
Beyond Being a Performance Provider: Taiwan Must Co-Shape Rules and Risk Governance
For Taiwan, the rapid expansion of AI supply chains presents significant opportunities for industrial development. However, Taiwan cannot focus solely on hardware manufacturing and performance improvements. Chiang cautioned that Taiwan’s long-term development could be constrained if it serves only as a “computing performance supplier” without actively participating in discussions on AI rules, standards, and risk governance.
Overall, global competition in AI continues to intensify, and simply slowing development is unlikely to alter the industry’s direction in the short term. The more important question is whether energy supply, supply chains, safety validation, and institutional development can keep pace with rapidly advancing technology. For Taiwan, the next major industrial policy challenge will be to seize AI opportunities while strengthening its capabilities in energy, critical components, cybersecurity, and governance.
Author: CIER Editorial Team
Date: September 16, 2026