AI infrastructure growth relies on more leverage & is increasingly opaque

As Nvidia joins forces with major Wall Street financial players to mobilize over $500 billion of third-party capital for artificial intelligence infrastructure, the intricate web of financing supporting the AI boom is coming under increasing scrutiny. Leading cloud service providers, known as hyperscalers, are ramping up spending on massive data centers and related technology, turning to bond markets, joint ventures, leasing structures, and a variety of other mechanisms to fund what has become an unprecedented global buildout of AI infrastructure.

In recent months, leverage has become an integral part of this expansion. Investors, including hedge funds and others, are borrowing through prime brokerage facilities and using derivatives to amplify their exposure to the AI sector. However, last month’s collapse of the AI-focused hedge fund Situational Awareness after losing bets on heavily leveraged AI equities has intensified concerns about the true extent of this borrowing, where risks reside in the system, and how rapidly such leverage can unwind in turbulent markets. The precipitous drop in that fund’s assets from $45 billion to $10 billion after a tech selloff - forcing a fire sale to Citadel at a discount - served as a reminder of the hazards of concentrated, leveraged bets, despite recent recoveries in names like SK Hynix and CoreWeave.

On the corporate side, Nvidia CEO Jensen Huang described to CNBC this week how Nvidia’s chips are now being viewed as “investable infrastructure assets,” with the company’s planned AI infrastructure platforms involving prominent financial partners such as Apollo, Blackstone, BlackRock, Brookfield, KKR, and Goldman Sachs. In an industry where tech giants are increasingly relying on off-balance-sheet vehicles - such as joint ventures and leases - to borrow for new data centers, equipment, and research facilities, the result is a vast buildup of future commitments that remain largely invisible to investors until lease payments begin. According to Goldman Sachs analysts, combined lease commitments across hyperscalers for data centers and related facilities have surged to $1.5 trillion, up dramatically from $200 billion five years ago. About $1 trillion of these obligations are for “uncommenced” leases that are not yet included in companies’ financial disclosures but will require substantial outlays in years ahead. Goldman cautioned in an August 6 research note that such hidden leverage can understate tech companies’ true borrowing and eventual liquidity needs as these commitments are recognized.

Industry observers note that this new wave of debt financing and less-visible leverage is raising questions about whether the commercial returns from AI infrastructure will justify such outlays. Lotfi Karoui, a multi-asset credit strategist at PIMCO, said that adjusted for inflation, the current AI capital spending cycle is set to be the largest investment surge since the construction of 19th-century railways. Despite consensus forecasts suggesting hyperscaler capital spending may exceed $1 trillion annually beginning in 2027, Karoui described the ultimate scale as “deeply uncertain.” He added that borrowers are now diversifying into euro, sterling, yen, Swiss franc, and Canadian dollar debt markets, with recent outperformance of euro-denominated bonds from major tech issuers like Amazon and Alphabet suggesting possible “demand fatigue” in the dollar bond market.

While concerns about leverage are rising, some analysts believe stretched earnings expectations present a more immediate risk. Sahil Mahtani, director at Ninety One’s Investment Institute, warned that the main vulnerability tied to the AI trade is not the scale of borrowing, but rather the high and rising forecasts for future profits. He noted that U.S. equity markets, already concentrated in a few dominant tech stocks, could face sharp swings if investor sentiment falters or if companies shift from share buybacks to new stock issuance just as lofty AI-fueled valuations are in question.

As the industry digests the fallout from the Situational Awareness fund and contemplates the risks of further leveraged trades, trade groups argue that borrowing by hedge funds remains a standard market tool and, for now, is not an inherent systemic risk. A spokesperson for the Alternative Investment Management Association, which represents hedge funds globally, noted that the Situational Awareness episode does not warrant a new regulatory review. The spokesperson differentiated past leverage disruptions in the market, such as the Archegos Capital Management and UK gilt market incidents, from the unique, contained circumstances surrounding this latest fund collapse.

As the AI infrastructure investment cycle accelerates at a record-breaking pace, investors and regulators alike are watching closely to see whether the revenue streams generated by AI can ultimately justify the hundreds of billions being poured into the sector and whether the methods of financing will stand up to future market volatility.