The Institute for Democratic Labor Studies, affiliated with the Korean Confederation of Trade Unions (KCTU), released an issue paper titled “AI Investment Fever: How Should We View It?” on October 1, warning that investment in artificial intelligence (AI) infrastructure is growing rapidly, fueled by debt and leverage, and that if returns fall short of expectations, losses could spread throughout the financial markets.
Research Fellow Ryu Seung-min, who authored the issue paper, noted that competition in AI investment has intensified rapidly since the launch of ChatGPT in late 2022. According to data from Stanford University’s Human-Centered AI Institute (HAI) cited in the report, global corporate AI investment reached $581.7 billion in 2025—an increase of approximately 130% from the previous year—with private investment accounting for $344.7 billion of that total. McKinsey projected that AI-related capital expenditures from 2025 to 2030 would reach $5.2 trillion under a base-case scenario and $7.9 trillion if demand accelerates. Morgan Stanley estimated global data center capital expenditures for 2026–2028 at approximately $3.21 trillion. The report analyzed that this pace of investment is extremely rapid, even when compared to past booms such as those in canals, railroads, electrification, and the dot-com boom.
The problem lies in revenue. Venture capital firm Sequoia Capital estimated that, to justify current investment trends, the AI industry would need to generate $600 billion in annual revenue from end users, but actual revenue falls significantly short of this figure. A 2025 report by the Massachusetts Institute of Technology (MIT) Project NANDA found that despite companies investing $30 billion to $40 billion in generative AI, approximately 95% of the organizations surveyed showed no measurable improvement in profit or loss. However, the researcher added that this study should be interpreted with caution, as it is not a representative sample of all companies.
The researchers identified three key financial structures underpinning the expansion of investment. The first is a circular financial system in which chip manufacturers, data center operators, and AI model companies invest in one another and, in turn, purchase each other’s products and services. This makes it difficult to distinguish between independent end-demand and demand generated by internal value-chain funding, which can cause capital expenditures to outpace actual demand. The report explains that this resembles “supplier financing” seen during the dot-com bubble of the late 1990s, when telecommunications equipment manufacturers lent funds to their clients to purchase equipment, only to suffer losses when those clients went bankrupt.
Second are loans secured by graphics processing units (GPUs), off-balance-sheet financing using special purpose vehicles (SPVs), and the securitization of data center-related loans and cash flows. If the value of GPUs declines faster than expected, the value of the collateral could plummet all at once, and SPV borrowing obscures the actual leverage. Securitized products are sold to insurers, pension funds, and asset management firms, spreading risk rather than eliminating it. Third is the expanding role of private credit, which is subject to weaker regulations and disclosure requirements than banks.
The AI value chain is concentrated among a small number of companies. According to the report, NVIDIA accounts for approximately 90% of the GPU market, TSMC accounts for approximately 90% of the cutting-edge process market, and SK Hynix and Samsung account for approximately 80% of high-bandwidth memory (HBM) sales. The research institute pointed out that since the Korean economy is closely linked to the AI value chain through semiconductors, a domestic shock could be significant if AI investments turn out to be a bubble and a market correction begins in earnest.
Power and water consumption by data centers, as well as environmental concerns, are also growing. In the United States, protests against the expansion of data centers took place simultaneously in 42 states this past July. The researcher argued that the debate over AI investment must go beyond the question of whether it is a bubble to examine who benefits and who bears the economic, environmental, and social costs, and that the scale and direction of investment, as well as its negative impacts, must be reviewed to establish necessary regulations. The researcher noted that the views expressed in this publication may differ from the official position of the Korean Confederation of Trade Unions (KCTU).
