• Gundlach warns that investors may believe they own diversified portfolios while depending heavily on the same AI-linked companies, infrastructure spending and financing assumptions.
  • He has already reduced his recommended equity allocation from 40% to 30% and favors an equal-weighted Fortune 500 approach to reduce dependence on the largest AI-linked stocks.
  • Recent pressure on AI-related financing and cash flow gives that warning added relevance.

Jeffrey Gundlach, the billionaire founder of DoubleLine Capital, is warning that investors may be underestimating the concentration risk embedded in their portfolios, cautioning that many who believe they own diversified holdings are actually making a concentrated bet on the same AI-linked companies, infrastructure spending and financing assumptions.

The warning comes as the market tests whether massive AI spending will produce sufficient revenue and cash flow. Reuters reported on October 27 that Amazon (AMZN), Microsoft (MSFT) and Alphabet (GOOGL) were expected to show accelerating cloud growth, but investors were increasingly concerned about financing costs. Rising bond yields at CoreWeave (CRWV) and record costs to insure Meta (META)’s debt highlighted those concerns.

Gundlach’s response has been to recommend less equity exposure and a more evenly weighted stock allocation—not to short AI or abandon stocks entirely. He has already reduced his recommended equity allocation from 40% to 30%. In a September interview, he favored an equal-weighted Fortune 500 approach to reduce dependence on the largest AI-linked stocks. He warned that eventual losers in the AI investment race could trigger a significant drawdown in risk assets, while explicitly saying, “I’m not selling everything. I’m not short anything.”

An important qualification: equal weighting reduces exposure to the largest companies, but does not eliminate AI exposure. MarketWatch noted that DoubleLine’s relevant equal-weight ETF still held AI-linked businesses, including Applied Materials (AMAT), Marvell (MRVL) and Palo Alto Networks (PANW).

The Financing Test

The immediate market test is whether AI spending produces sufficient revenue and cash flow. Reuters reported nearly $1.7 trillion in combined backlog and planned spending of as much as $600 billion this year across Amazon, Microsoft and Alphabet. These are spending plans and contracted business—not realized profits. Their forthcoming earnings were expected to test the investment case.

Meta offered a concrete example of infrastructure spending weighing on cash generation. Third-quarter free cash flow was negative $2.12 billion, compared with positive $10.63 billion a year earlier.

The risk extends beyond high stock valuations. AI investment generates business for chipmakers, data-center operators and infrastructure suppliers, but those suppliers’ revenues depend on customers continuing to finance enormous capital budgets. Reuters estimated roughly $800 billion in 2026 capital spending across the broader hyperscaler group, supporting earnings throughout the supply chain. That broader figure covers more companies than the $600 billion three-company estimate above.

A Portfolio-Fragility Argument

Gundlach’s criticism has developed from concerns about financing into an explicit portfolio-allocation stance. In August, he questioned using rapidly evolving AI chips as collateral for long-term debt: the equipment could lose economic value before the borrowing is repaid. In September, he described moving farther away from the AI trade’s “epicenter.”

The historical parallel is an investment boom in which genuine technological progress does not guarantee attractive returns for every investor. Reuters Breakingviews has emphasized a related accounting issue: equipment suppliers recognize sales immediately, while buyers spread infrastructure costs over years through depreciation. That timing can make current profits look stronger than the eventual cash economics. This is an analytical argument, not proof that reported earnings are improper.

Two closely connected developments illustrate the stakes. Reuters reported October 26 that the mainland STAR50 index had fallen 40% in less than four months, erasing about $480 billion in market value amid valuation, demand and export-policy concerns. It is a parallel case, not evidence that U.S. markets must follow the same path. Separately, Reuters reported disappointing reactions to Meta and Lam Research (LRCX) results and said coming earnings would be judged on capital-spending guidance and returns from AI outlays—not simply growth headlines.

Policy and Public Backdrop

Two policy issues directly affect the investment outlook. U.S. chip-export policy: Reuters reported in March that officials were considering tying large overseas AI-chip shipments to security assurances or investment in U.S. data centers. That report described a proposal, not a finalized rule; the retrieved evidence does not establish its subsequent status. China’s technology-export controls: Reuters reported today that proposed Chinese rules would generally allow overseas access to Chinese AI services while subjecting some transfers of underlying models, technology and overseas deployments to approval. The draft, released October 16, covers 23 AI-related technologies; a temporary registration requirement for certain transfers took effect immediately.

Meanwhile, Reuters reports growing protests over pressure on power grids, electricity bills and the environment. That backlash has made data-center expansion an election issue as well as a commercial one. Rising AI-related bond yields and debt-insurance costs show that creditors are demanding more compensation for risk. Sustained financing pressure could constrain expansion even where demand for computing remains strong.

The central public debate is whether today’s infrastructure boom is laying the foundation for durable productivity gains—or building capacity whose eventual returns will not justify its cost. Recent reporting documents both strong demand indicators and worsening financing concerns; neither alone settles that question.

In the short term, the most informative signals are cloud revenue growth, free cash flow, capital-spending guidance and AI borrowers’ financing costs. The immediate question is whether strong demand can offset the mounting cost of building and funding capacity. Over the longer term, three outcomes remain plausible: successful monetization, a selective shakeout, or a broader retrenchment.

Gundlach’s warning is therefore best read as a portfolio-fragility argument, not a verified prediction of an imminent crash: AI can remain economically important while concentrated exposure to its financing and spending cycle proves costly. His documented position does not establish when—or whether—a broad correction will occur.

Correction: An earlier version misstated the timing of the Federal Reserve’s September rate increase; it was a quarter-point increase, and the Fed held rates unchanged on October 28. The article also clarified that the $600 billion spending estimate refers to three companies, while the $800 billion figure covers the broader hyperscaler group.