[Edaily Marketin YunJi Kim Reporter] During the era of massive railroad and telecommunications network construction in the United States, companies poured enormous amounts of capital into expanding railroad and fiber-optic networks in anticipation of future demand. However, it took several years for actual demand to catch up with the pace of investment, and in the meantime, a significant number of companies that were unable to recoup their investments went bankrupt.

The perspective of major U.S. pension funds regarding the recent boom in artificial intelligence (AI) infrastructure investment is not much different. While AI may drive economic growth in the long term, concerns are spreading that the current scale of investment is outpacing the actual ability to generate returns.

According to the global investment industry on the 22nd, voices warning against overinvestment in AI infrastructure are growing louder among U.S. pension funds. The concern is that as AI-related investments expand across various asset classes—including stocks, private equity (PE), and infrastructure—investment risks could become concentrated in specific industries. Some pension funds are already taking concrete steps in asset allocation, such as rejecting investment proposals from asset managers.

Notably, the Texas Teachers’ Retirement System (TRS), which manages $225.3 billion (approximately 306 trillion won) in assets, recently held a meeting of its investment committee to discuss the scale of AI infrastructure investments and their future profit-generating potential as key topics. Currently, capital invested in AI accounts for approximately 3.5% of U.S. annual gross domestic product (GDP), a level that exceeds the investment levels seen during the construction of canals, railroads, power grids, highways, and telecommunications networks in the past. TRS maintains that, given this massive influx of capital, it is necessary to assess whether actual returns can justify the scale of investment.

TRS is concerned that it may take a considerable amount of time for AI infrastructure, which has received such massive funding, to generate returns commensurate with the scale of the investment. Even if the AI industry grows in the long term, it cannot be ruled out that it will take 10 to 20 years for actual demand to catch up. In this scenario, there is concern that related companies will be burdened with financial strain for an extended period without being able to recoup their investments, and that investors who funded these companies could also be exposed to the risk of losses.

Some pension funds are adjusting their actual asset allocations to mitigate AI investment risks. The New York City Retirement System (NYCRS), which manages $327 billion (approximately 445 trillion won) in assets, recently rejected investment proposals from some asset management firms. This decision was based on the assessment that even if investments are spread across different asset classes—such as stocks, private equity, and infrastructure—losses could occur simultaneously across multiple asset classes if the AI industry’s growth slows. They explain that they are deciding whether to invest by considering not only the risk of individual assets but also the proportion of AI-related investments within the overall portfolio.

The fact that pension funds have begun reviewing their AI investment allocations is partly due to the expansion of such investments from publicly traded tech stocks to private companies. As of the end of August last year, the Texas Teachers’ Retirement System (TRS) held stocks in companies such as NVIDIA, Microsoft, Apple, and TSMC among its major equity holdings. Of these, NVIDIA accounted for 2.45% of the fund’s total equity holdings. Investments in unlisted AI companies are also continuing. Global institutional investors, such as the Ontario Teachers’ Pension Plan (OTPP) in Canada, participated in a $65 billion funding round for the U.S. AI company Anthropic last May. As AI-related investments expand across various asset classes, the need for pension funds to manage investment risks not only at the level of individual assets but also at the portfolio level is growing.

However, it is not easy for pension funds to scale back AI investments altogether. Given the rapid pace at which AI technology is spreading, they cannot afford to miss out on investment opportunities arising from the growth of related industries. According to the Federal Reserve Bank of St. Louis, generative AI has achieved a 53% adoption rate in about three years since the launch of ChatGPT. This rate exceeds the initial adoption speeds of the internet and personal computers (PCs). Analysts note that as AI is adopted, the competitive landscape in existing industries could rapidly shift; therefore, pension funds must secure investment opportunities in these sectors while simultaneously assessing the competitiveness of their existing investment assets.

Consequently, pension funds are focusing less on whether to invest in AI and more on the scale of investment and the level of risk exposure they can tolerate. An official in the global capital markets industry stated, “Even as the AI industry grows rapidly, not every company will generate returns commensurate with the scale of investment,” adding, “A consensus is forming that we must evaluate both the profitability of individual investment assets and the proportion of AI investments within the overall portfolio.” The official added, “We are in a situation where we must manage the risks associated with AI investments while also preparing for existing investment assets that could lose their competitiveness due to technological changes.”