How to avoid the pitfalls of AI? This is a major challenge for pension funds and sovereign wealth funds.

The AI boom is eroding the diversified investment logic upon which institutional investors rely. From stocks and private equity to corporate bonds and infrastructure, AI risks have permeated almost all asset classes, plunging pension funds and sovereign wealth funds managing hundreds of billions of dollars into unprecedented difficulties.

According to a Bloomberg report on Monday, Monte Tarbox, chief investment officer of the New York City Retirement System (NYCRS), recently rejected a fundraising offer from a private equity fund because the product's AI holdings were too heavy. "When that day comes, we don't want to be left hanging, with people asking, 'Why did you say yes to everything?'" he said. This decision reflects the common predicament of almost all major institutional investors on Wall Street today.

Lisa Shalett, chief investment officer of Morgan Stanley Wealth Management, warned that AI is exacerbating the diversification challenges already faced by asset allocators. In an environment of persistently sticky inflation, the traditional 60/40 stock-bond portfolio has repeatedly failed, and the deep penetration of AI is further worsening the positive correlation between stocks and credit.

Goldman Sachs estimates that AI infrastructure-related companies (including chipmakers and hyperscale cloud computing companies) account for about 40% of the total market capitalization of the S&P 500; according to Apollo Global Management, AI-related issuances account for nearly half of the total investment-grade bond issuances this year and 87% of venture capital funding.

AI risks are everywhere; decentralization is an illusion.

AI's penetration into the market is multi-dimensional. In the public market, the stocks of just three chipmakers account for more than a quarter of the weighting in emerging market benchmark indices. In the private market, private equity has almost become synonymous with AI startups. In the bond market, the largest issuers are all hyperscale cloud computing companies. In the infrastructure sector, almost every major project points to data centers. Even US Treasury bonds can be included in the AI trading narrative—the additional electricity demand drives up inflation expectations, thereby affecting yield trends.

The presence of companies across multiple asset classes further concentrates risk. Alphabet, for example, has a market capitalization of $4 trillion, outstanding debt of $130 billion, and infrastructure connecting over 30 data centers globally. Meanwhile, many companies have so-called "revolving financing" arrangements, holding shares in each other, resulting in highly intertwined risk exposures.

Thomas Salopek, Head of Cross-Asset System Strategy Research at JPMorgan Chase, developed four models simulating institutional investor asset allocation. The results show that these portfolios have exhibited a significant positive correlation with AI-related risk factors in recent years. In an Invesco survey of 90 sovereign wealth funds, over half of the respondents listed market concentration as the primary risk associated with AI-related investments.

Quantitative AI exposure: various institutions showcase their unique strengths

Faced with this challenge, institutional investors are exploring their own ways of coping, but the primary challenge is how to define and measure AI exposure. Unlike conventional classifications such as asset class, industry, or region, there is currently no unified standard for AI exposure—it can be a pure chip manufacturer or an ordinary company that simply uses AI technology.

The Los Angeles County Employees Retirement Association (LACERA) specifically discussed the issue of AI holdings in the private equity market at its board meeting in May. Chief Investment Officer Jonathan Grabel subsequently led a comprehensive review, combining internal bottom-up analysis with a top-down approach using MSCI, estimating that 8% to 19% of the pension fund's holdings were AI-related. "We don't want to pursue perfection at the expense of insight," he said.

Elo Mutual Pension Insurance Co., a Finnish company managing €36 billion in assets, is using AI to track AI, employing AI-driven tools to monitor the sensitivity of its public holdings to AI adoption trends. Kari Vatanen, Head of Asset Allocation and Alternative Investments, acknowledges that limiting AI exposure for diversification comes at a cost—potentially harming performance if the AI narrative continues to dominate market returns for years to come. A Bloomberg Intelligence index tracking global AI-related stocks shows an annualized excess return of 11 percentage points over the past two years.

After conducting factor analysis on about 50 large pension funds, financial analysis firm Markov Processes International found that these funds have an average excess AI exposure close to zero, but some individual funds show significant deviations. For example, the excess AI exposure of the California Public Employees Retirement System (CalPERS), the largest public pension fund in the United States, has been rising in recent years, mainly due to investments in large private companies.

The "total portfolio approach" has emerged as a new problem-solving strategy.

The need to quantify AI risk is revitalizing an investment framework called the "Total Portfolio Approach" (TPA). This framework breaks down asset class barriers, comparing all investments on the same dimension to seek the overall optimal allocation. CalPERS has recently officially adopted this approach.

In June, Aware Super, Australia's A$240 billion investment firm, completed a multi-year internal investment platform revamp codenamed "Project Odin," integrating third-party data and analytics tools, including those from BlackRock. Michael Clavin, Head of Liquidity and Markets, stated that the platform is naturally suited for tracking AI exposure across the entire fund. "True diversification isn't just about geographical or sectoral diversification; it's about truly understanding the correlations between assets and themes," he said.

Marsh Investments and Retirement (formerly Mercer, managing $846 billion in assets) spent six months developing a thematic exposure tracking system that uses proxy AI to analyze the revenue and profit sources of public and private companies. It officially launched in February of this year. Andrew McDougall, Chief Investment Officer for the US, stated, "The key is that you have to know what you hold before you know where you're going. Institutions with good systems today can roughly assess the impact before making investments; while those with lagging systems may only discover the problem when it's too late."

For NYCRS' Tarbox, the challenge lies in keeping pace with the rapid and far-reaching evolution of a technology. He stated that even the most sophisticated portfolio mapping can quickly become obsolete, and AI's risk profiles must be constantly redrawn. "We not only need to improve the resolution of our microscopes, but also accelerate the frequency of our assessments," he said. "Almost every fund we commit to investing in inevitably brings more exposure to AI."

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.