“Monetization Remains Elusive Despite Astronomical Spending” The Dark Side of the AI Investment Boom, as Redemption Shockwaves Spread Through Private Credit Markets

Surging redemption requests force even Blackstone into defensive measures against capital outflows
Debt burdens accumulated through AI data center and semiconductor investments come into focus
Heightened uncertainty over monetization fuels growing tension in private credit markets

Blackstone, one of the largest investment firms in the United States, has imposed restrictions on redemption requests for the first time in a massive lending fund offered to individual investors. While the move is officially framed as a liquidity-management measure in response to a surge in withdrawal requests, markets are increasingly focused on the darker side of the artificial intelligence (AI) investment frenzy underlying the development. As enormous sums continue to pour into the race for data centers and semiconductor capacity, concerns are mounting that revenue generation may fail to keep pace with the accelerating scale of investment. The result has been a growing sense of caution across private credit markets closely tied to AI infrastructure financing.

Redemption Requests Surge 10%, Prompting Blackstone to Impose Limits

According to Reuters on June 8, Blackstone disclosed in a recent investor letter that its flagship private credit fund, Blackstone Private Credit Fund (BCRED), received redemption requests equivalent to roughly 10% of fund shares during the second quarter but capped withdrawals at 5% in accordance with fund rules. The $84.8 billion BCRED fund has never before restricted redemptions.

Private lending funds, which lack deposit-gathering functions comparable to banks and therefore cannot maintain large pools of liquidity for withdrawals, are inherently vulnerable to large-scale redemption requests. Because these vehicles invest in long-term, illiquid loan assets, they cannot simply call in loans whenever investors seek to redeem their capital. This creates a fundamental liquidity mismatch between assets and liabilities. During favorable periods, such funds can reliably deliver annual returns of around 10% to investors. However, once redemption pressure intensifies beyond manageable levels, a vicious cycle can emerge. Managers are often forced to sell their highest-quality assets or the most easily recoverable loans first, shrinking both profitability and fund size while leaving the vehicle increasingly exposed to future redemption pressure.

Blackstone’s efforts to contain investor anxiety reflect precisely this risk. During the first quarter, the firm fully honored redemption requests equal to 7.9% of BCRED shares. To accommodate withdrawals exceeding regulatory limits, Blackstone temporarily raised the redemption threshold from 5% to 7% and covered an additional 0.9% through capital contributions from the firm and its executives. The company deployed not only corporate capital but also $150 million from the personal wealth of 25 senior executives to reassure investors. However, after second-quarter redemption requests climbed to 10% of total fund shares, Blackstone ultimately activated measures designed to prevent further capital flight. For investors, this means that although withdrawal requests have surged, only roughly half of the requested amount will actually be redeemed under fund rules.

Similar pressures have emerged elsewhere. Cliffwater’s private credit fund for retail investors, CCLFX, recently informed investors that it would again cap redemptions at the 5% limit during the second quarter following similar restrictions in the previous quarter. The fund received redemption requests totaling 17% of assets. Anxiety within private lending markets is also spreading into private equity (PE) funds that invest in unlisted companies. On June 4, European investment giant Partners Group Holding announced partial redemption restrictions after withdrawal requests in its U.S. retail PE fund exceeded the 5% limit. The firm had already imposed redemption restrictions on another PE fund just one day earlier.

Big Tech’s $800 Billion Annual Spending Spree and the Hidden Fixed-Cost Leverage Buried Beyond Financial Statements

Private lending funds operate by deploying investor capital into corporate loans, with returns distributed through dividend-like payments. The sector has expanded rapidly in recent years, particularly among wealthy investors seeking high-yield lending opportunities in a high-interest-rate environment. However, concerns regarding deteriorating corporate loan quality and, in particular, uncertainty surrounding the business outlook of AI companies have intensified this year, raising questions about the overall health of the private credit market.

A major source of growing market unease lies in the enormous debt burden accumulated through AI infrastructure investment. America’s technology giants are pouring unprecedented sums into data centers and semiconductors in their race for AI leadership. According to the Financial Times, total AI-related capital expenditures by the four major hyperscalers—Alphabet, Meta, Microsoft, and Amazon—are expected to exceed $700 billion this year. Including Oracle and Apple, total infrastructure investment could reach as much as $805 billion in 2026. This represents nearly 3% of U.S. gross domestic product, an extraordinary figure for a single technology investment cycle.

The problem is that much of this spending has been financed through borrowing and bond issuance. During the previous year, the five largest hyperscalers—Alphabet, Meta, Microsoft, Amazon, and Oracle—issued a combined $121 billion in corporate bonds, 4.3 times the five-year historical average of $28 billion. Morgan Stanley estimates that the broader technology sector will need to absorb approximately $1.5 trillion in additional debt over the next three years. Amazon has already accelerated debt-financed fundraising, issuing a record $16.8 billion in corporate bonds in March.

An even greater credit risk resides in obligations financed through special-purpose vehicles (SPVs) that remain largely outside corporate balance sheets. Oracle, Meta, xAI, and CoreWeave have collectively accumulated more than $118.6 billion in AI-related liabilities through off-balance-sheet financing structures. Oracle established a $66 billion SPV to construct and subsequently lease data centers, while Meta raised $30 billion through an SPV known as “Beignet Investor.” A substantial portion of this debt is ultimately absorbed by private credit markets. Private lending funds have become major providers of capital for data center construction, server procurement, and semiconductor purchases, creating increasingly intertwined interests between the AI investment boom and the private credit ecosystem.

Capital Floods In Before AI Profits Materialize, Raising Fears of a Replay of the 1873 Financial Crisis

Against this backdrop, many observers warn that delays in monetizing AI inference services could expose vulnerabilities across the broader financial system. Financing structures built around leases and SPVs function as fixed-cost leverage mechanisms whose expenses do not decline immediately when demand weakens, potentially magnifying downside risks. The ultimate question confronting markets is when this enormous wave of investment will finally translate into sustainable, recurring profits. While Big Tech companies continue escalating the scale of investment in pursuit of market dominance, soaring electricity costs, AI chip prices, data center construction expenses, and financing costs are simultaneously increasing pressure to generate returns.

Yet the AI investment boom shows little sign of slowing. Investors and technology giants appear fully aware of the risks associated with a potential bubble collapse, but remain gripped by fear of missing out (FOMO), believing that falling behind in this infrastructure arms race could result in permanent competitive disadvantage. Demand remains strong even in cases where immediate use cases are unclear. The near doubling over the past year of “construction in progress” assets on the balance sheets of Google and Meta suggests that data centers are being built and semiconductors acquired faster than they can be brought online, leaving growing pools of idle infrastructure.

Perhaps the most concerning issue is the possibility that investment spending could overtake operating cash generation. Operating cash flow for the five major hyperscalers stood at $380 billion in 2023 and is projected to reach $750 billion by 2026. Over the same period, however, capital expenditures are expected to surge to $820 billion. Based on the upper end of market consensus estimates, investment spending could exceed operating cash generation by approximately $70 billion in 2026, effectively pushing the sector into a period of negative cash-flow investment.

The parallels with the 19th-century railroad investment boom are striking. At the time, U.S. financial markets poured enormous sums into railroad construction. Railroads ultimately became a foundational pillar of American industrialization, but the financial system collapsed before the investments could be fully absorbed. The failure of Jay Cooke & Company, the largest underwriter of railroad bonds, became the trigger for the Panic of 1873. The deterioration of railroad investments continued thereafter, and by 1879 more than half of all U.S. railroad bonds were unable even to make interest payments.

Today’s AI investment frenzy exhibits similar vulnerabilities. Should the pace of model advancement at OpenAI and Anthropic, data center deployment schedules, or growth in inference demand fall short of expectations, investor sentiment could cool rapidly. While some analysts argue that AI investments are supported by greater equity capital than the railroad bubble and therefore possess stronger shock-absorption capacity, the useful life of semiconductors—the sector’s core asset—is far shorter than that of railroad infrastructure. As chip performance cycles accelerate, asset values may deteriorate more quickly, increasing the risk of collateral impairment and weakening loan quality. Policymakers also possess less flexibility than in previous crises. If excessive AI investment ultimately triggers broader financial-market stress, it would be far more difficult for the U.S. government and the Federal Reserve to deploy the massive liquidity backstops that helped stabilize global markets during the 2008 financial crisis.