In recent weeks, investors have gotten rattled by a potential slowdown in corporate spending on AI, but that is unlikely to put much of a dent in the boom in AI capital spending. The race is on.
Even though it’s still early days, the sheer scale of the equity and debt issuance funding the infrastructure needed for AI should prompt investors to rethink how to manage risk and apply foundational concepts such as diversification, according to Jared Gross, head of institutional portfolio strategy at J.P. Morgan Asset Management. Through July of this year, more than $800 billion of capital was invested in AI on top of almost $500 billion in 2025.
The portfolios of pensions, endowments, and other institutions may look diversified at the top level between U.S. and international stocks, private equity, private credit, infrastructure, real estate, and venture capital. But J.P. Morgan Asset Management argues that the amount of capital spending has already embedded many of the same AI-related risks into almost every asset class. JPMAM’s baseline scenario forecasts that spending will peak around $1.2 trillion before the end of the decade. With “sustained demand,” the firm expects capital spending to peak around $1.4 trillion, with $1 trillion spent annually for many years after that.
“The vast size of the capital invested has, I think, overwhelmed some traditional models of portfolio construction and diversification,” said Gross during an interview on the asset allocation implications of AI. Gross co-chairs the Strategic Investment Advisory Group at J.P. Morgan Asset Management, which spent about six months on research behind a report called, “AI-conscious asset allocation: Portfolio construction in a capital investment boom” and which was released today.
JPMAM sees AI investments, benefits, and risks unfolding in at least three phases. Companies that make the picks and shovels during the buildout; core AI companies in the second phase, and companies that stand to benefit.
“But diversifying along traditional asset class boundaries may not be the best way to control those risks and target the best opportunities, said Gross. “AI has crept into many asset classes that we used to regard as distinct and fundamentally diversifying relative to one another, and it’s very hard, particularly in the private markets, to tease out exactly how much AI exposure is in any one of those particular investments.” (Institutions, however, are getting more transparency into private equity and other funds, particularly after the decline in the software sector that began in February. Gross called that a “wake-up” call to GPs.)
Technology concentration has been a worry for years as the largest tech companies have grown to become a larger and larger share of the S&P 500 and other indices. But the AI boom has compounded the problem and tucked it into far broader set of investments.
As Gross pointed out, AI represents a huge swath of U.S. large-cap equities and venture and growth capital. And the same goes for China, but the country is a much smaller part of investors’ portfolio. Private credit funds have exposure through financing of AI data centers and debt from the hyperscalers such as Meta and Amazon is a growing part of investment grade credit. Still, J.P. Morgan Asset Management points out in the report that core real estate funds focused on multi-family, retail, and office, have little exposure to AI, as do many infrastructure strategies focused on power distribution, And, of course, government bonds, and short-duration fixed income have “negligible exposure” but also offer little return.
The boom in AI has often been compared to infamous booms and busts like the Internet tech build-out of the late 1990s that collapsed in 2000. Critics point out that investors often lost money on companies that went bankrupt, even if the technology itself transformed the economy in the decades since.
The JPMAM report includes a detailed historical analysis of five past booms, including railroads, oil and gas, and even the U.S. highway system.
One thing that is unique to the development of AI, Gross said, and one reason it may be difficult to guard against the decline in diversification, is that all booms are stopped eventually by the realities of financial discipline. Companies overbuild, returns decline, firms go bankrupt, and companies get restructured and sold off. Investors may lose a lot of money, but the excess does get burned off.
But with AI, the hyperscalers are financing much of what’s needed for AI to work. These companies are financially healthy, with plenty of cash. Gross explained that they’ve issued debt and follow-on equity, and are guaranteeing some of the off-balance sheet obligations. “At some point, organic revenue has to arrive,” said Gross.
As Gross notes in the study, “For now, we see little evidence of vulnerability among major players, nor any subsequent rationalization or reorganization. Public hyperscalers remain profitable despite this large capital spend and significant debt issuance, although they have less financial flexibility than they once did.
Perhaps ironically, the ability of these firms to “stay in the game” longer could create the conditions for overinvestment in costly physical infrastructure that may have a hard time generating adequate returns.”
Gross is focused for now on investors’ understanding of where “you’re taking on positive AI exposure, where you can take on unintended exposure, which could be good or bad, and where you find defensive… non-AI exposures that will balance some of that uncertainty.”
