The relentless advance of the artificial intelligence boom is dismantling the diversification playbook that institutional investors have long depended on. From equities and private equity to corporate bonds and infrastructure, AI-driven risk has woven itself into nearly every asset class, leaving pension funds and sovereign wealth managers who control hundreds of billions of dollars facing a quandary with no easy escape route.

As reported recently, Monte Tarbox, chief investment officer of the New York City Retirement Systems, made a pointed decision by turning down a fundraising offer from a private equity vehicle. His reasoning was direct: the product carried an excessively heavy AI allocation. "When that day comes, we don't want to be left standing there with people asking, 'Why did you say yes to everything?'" he remarked. This single choice highlights a predicament now shared by most major institutional investors across Wall Street.

Lisa Shalett, chief investment officer at Morgan Stanley Wealth Management, cautions that AI is amplifying the diversification challenges allocators already face. With inflation remaining persistently sticky, the conventional 60/40 equity-bond portfolio has repeatedly faltered, and the deep penetration of AI is further worsening the positive correlation between stocks and credit.

Goldman Sachs estimates that AI infrastructure-linked companies, including chipmakers and hyperscale cloud operators, now account for roughly 40% of the total market capitalization of the S&P 500. Data from Apollo Global Management indicates that AI-related issuance represents nearly half of this year's investment-grade bond supply and constitutes a staggering 87% of venture capital funding. AI risk is everywhere, making diversification seem like an illusion. The infiltration of AI into markets is multi-dimensional. In public markets, just three chipmaker stocks alone make up over a quarter of the weighting in the emerging markets benchmark index. In the private sphere, private equity has become virtually synonymous with AI startups. In the bond market, the largest issuers are hyperscale cloud companies. In infrastructure, nearly every major project points to data centers. Even US Treasuries can be woven into the AI trade narrative—additional power demand pushes up inflation expectations, which in turn influences yield movements.

The same cluster of companies spanning multiple asset classes further concentrates risk. Take Alphabet Inc as an example: its equity market value reaches $4 trillion, it carries $130 billion in outstanding debt, and its infrastructure connects over 30 data centers worldwide. Meanwhile, several firms are entangled in so-called "circular financing" arrangements, holding stakes in one another with highly interwoven risk exposure.

Thomas Salopek, head of cross-asset systematic strategy research at JPMorgan, constructed four models simulating institutional investor allocations. The results show that in recent years these portfolios have exhibited significant positive correlation with AI-related risk factors. In a survey of 90 sovereign wealth funds conducted by Invesco, over half of respondents cited market concentration as the leading risk tied to AI investments.

Where Institutions Start to Quantify the Problem

Faced with this challenge, institutional investors are each exploring their own methods of response, but the primary hurdle is defining and measuring AI exposure. Unlike conventional classifications such as asset class, industry, or geography, AI exposure currently lacks a unified standard—it can encompass anything from a pure-play chipmaker to an ordinary enterprise that merely adopts AI technology.

The Los Angeles County Employees Retirement Association dedicated a board meeting in May to discussing private market AI holdings. Chief investment officer Jonathan Grabel then spearheaded a comprehensive review, blending internal bottom-up analysis with MSCI's top-down approach, to estimate that 8% to 19% of the pension fund's holdings are AI-related. "We don't want to let the pursuit of perfection stand in the way of insight," he said.

Finland's Elo Mutual Pension Insurance Co., managing €36 billion in assets, has portfolio managers using AI to track AI. They employ AI-driven tools to monitor how sensitive public holdings are to AI adoption trends. Kari Vatanen, head of asset allocation and alternative investments, candidly acknowledges the trade-off: limiting AI exposure for the sake of diversification carries a cost—if the AI narrative continues to dominate market returns for several more years, this approach could hurt performance. An index tracking global AI-related stocks compiled by Bloomberg Intelligence shows an annualized excess return of 11 percentage points over the past two years or so.

Factor analysis conducted by the financial analytics firm Markov Processes International on roughly 50 large pension funds found that average excess AI exposure is close to zero. However, individual funds show notable deviations—the United States' largest public pension fund, the California Public Employees' Retirement System, has seen its excess AI exposure climb steadily in recent years, largely driven by investments in large unlisted companies.

A Fresh Framework Emerges to Rethink AI Risk

The need to quantify AI risk is breathing new life into an investment framework known as the Total Portfolio Approach. This method breaks down asset class barriers, placing all investments on a single axis for comparison to pursue optimal overall allocation. CalPERS has recently formally adopted this framework.

Australia's Aware Super, managing A$240 billion, completed its multi-year internal investment platform overhaul codenamed "Project Odin" this June, integrating third-party data and analytical tools from providers including BlackRock. Michael Clavin, head of liquidity and markets, states that the platform is naturally suited for tracking AI exposure at the total fund level. "True diversification isn't just about geography or industry—it's genuinely understanding the correlation between assets and themes," he said.

Marsh Investments and Retirement, formerly Mercer with $846 billion in assets under management, spent six months developing a thematic exposure tracking system that uses agentic AI to parse revenue and profit sources of public and private companies. The system went live this February. Andrew McDougall, chief investment officer for the US region, explains: "The key is that you have to know what you hold before you can know where you're going. Institutions with good systems today can gauge the impact before making an investment; those with lagging systems may only discover problems when it's too late."

For Tarbox at NYCRS, the challenge is keeping pace with a technology whose impact is both swift and profound. He notes that even the most sophisticated portfolio mapping can quickly become outdated, requiring the risk profile of AI to be continuously redrawn. "We not only need to increase the resolution of the microscope, but also accelerate the frequency of our assessments," he said. "Almost every fund we commit to inevitably brings more AI exposure."