A Political Economic Perspective on the Global Generative AI Adventure

This study provides a political-economic and thermodynamic critique of the global generative artificial intelligence (AI) and Large Language Model (LLM) adventure. Driven by financial capital and geopolitical anxiety, the LLM frenzy has been widely hailed as the engine of a "fourth industrial revolution". However, grounded in the first law of thermodynamics and the reality that probability-based token prediction faces insurmountable "hallucinations" before reaching the zero-tolerance, high-precision demands of real-world physical manufacturing, this study constructs a "Silicon Ledger" and "Carbon Ledger" analytical framework. Through an "Energy Matryoshka" model, the research evaluates the asymmetric resource consumption and development trajectories of the United States and China between 2023 and 2026. The findings reveal that: The U.S. Model (Capital-Silicon Heavy Industry): Dominated by machine computing power consumption (accounting for 68% of its total energy), the U.S. relies on elite-driven R&D and massive hardware clusters. Its primary losses are financial and capital-based, facing hundreds of billions of dollars in fixed-asset write-downs and short-term hardware depreciation if the productivity myth collapses. The China Model (Engineer-Carbon Intensive Collaboration): Constrained by semiconductor blockades, China mobilized a "people's war" model featuring 200,000 core engineers and millions of support personnel, where human and social logistics energy consumption accounts for 77% of its total ledger. Consequently, China’s total societal energy expenditure for the LLM ecosystem (approx. 44 billion kWh/year) surpasses that of the U.S. (approx. 36.6 billion kWh/year). China's losses are characterized by irreversible intellectual mismatch and opportunity costs, squandering a critical demographic window of top-tier talent on parameter-tweak involution. Ultimately, the study argues that LLMs represent a pseudo-technology wave that has diverted both superpowers from foundational hard technologies. To mitigate these losses after the bubble's deflation, strategic course corrections are urgently required: shifting idle high-performance computing clusters toward zero-tolerance industrial domains (such as computational materials, chip EDA simulation, and smart grid control) and channeling carbon-based intellectual resources back into core operating systems, industrial software, and physical manufacturing hardware.