Is the critical moment for capital approaching? The AI narrative is shifting from technology to financing, with giants facing cash flow crises and the bond market sounding the alarm

Author: Long Yue
The narrative around AI investment is undergoing a transformation. It initially began as a technology story, then evolved into a capital expenditure story, and is now becoming a financing story.

Recently, Greg Jensen from Bridgewater Associates summarized the current situation in one sentence: "We are entering a critical moment for capital."

Morgan Stanley predicts that total spending on AI infrastructure will reach $3.2 trillion by 2028, with approximately $1.75 trillion needing to be raised through the credit markets. The sources of funds have expanded from traditional investment-grade bonds to leveraged loans, private credit, and securitized products. Money is available, but it won't be cheap.

The bond market has already begun to react. According to Apollo, AI-related bond issuance has accounted for 40% of long-duration supply. The credit spreads of hyperscalers have widened significantly this year, while the overall investment-grade market has barely moved.

The "new bond king" Gundlach bluntly stated, issuing long-term bonds backed by GPUs is akin to "using bananas to create a 30-year ABS."

Cash Flow is Heading in the Wrong Direction

Capital expenditure forecasts have been repeatedly raised, while free cash flow forecasts continue to decline.

Morgan Stanley has significantly lowered the free cash flow forecasts for major hyperscalers for 2027. Among them, Oracle's situation is the most pronounced, with its 2027 free cash flow forecast nearing -$40 billion.

The scale of spending commitments is equally staggering. The procurement commitments of hyperscalers have exploded to $982 billion. Morgan Stanley points out that real capital expenditures are now largely off-balance-sheet. This means that merely looking at the balance sheet will severely underestimate the actual funding pressure.

The Credit Market Has Started Pricing Risk

Investors are demanding higher risk compensation.

Morgan Stanley data shows that the credit spreads related to hyperscalers have "widened significantly this year, far exceeding the overall investment-grade market"—the spreads for high-quality hyperscalers have widened by about 25 basis points, while those for ordinary hyperscalers have widened by about 22 basis points, and the overall investment-grade market spread has changed by 0.

Oracle is the most closely watched case. The credit default swaps (CDS) of other hyperscalers are generally in the range of 30 to 80 basis points, while Oracle's CDS surged from about 40 basis points in mid-2025 to a peak of nearly 190 basis points in April 2026, and is currently hovering around 180 basis points. Meta's CDS has risen moderately to about 75 basis points. The credit market has clearly identified the one it is most concerned about.

Strong Balance Sheets, but the Problem Lies Ahead

Morgan Stanley's Q1 2026 data shows that the total leverage ratio of hyperscalers is only 1.3 times, with a net leverage ratio of 0.5 times, a cash/debt ratio as high as 128%, and a median rating of AA-. In contrast, the total leverage ratio for non-financial investment-grade entities is 2.4 times, with a rating of BBB.

But the problem lies in the future. Morgan Stanley has significantly raised its forecast for cloud computing capital expenditure growth in 2027 from 14% to 29%, and hyperscalers are also "reaffirming confidence in investment returns." As spending forecasts double, the financing gap widens accordingly.

"New Bond King" Gundlach: Using GPUs as Collateral is Like Using Bananas for ABS

Regarding the financing schemes backed by AI assets that have emerged in the market, renowned bond market investor Jeff Gundlach offered sharp criticism.

He warned against a $50 billion fund consortium plan, stating that it "is likely to fail the test of time." He questioned on social media: "Using assets with unknown lifecycles as collateral for long-term debt? Why not do a 30-year ABS deal backed by bananas in the warehouse?" He added that these are still "brand new engineering bananas with unknown lifecycles."

Gundlach's analogy points to the core issue: GPUs depreciate rapidly, and the technology iteration cycle is much shorter than the debt maturity period. When using such assets as collateral for long-term financing, the actual value of the collateral is highly uncertain.