A critical moment for capital has arrived? AI narrative shifts from technology to financing, major players face cash flow crises, and warning signs emerge in the bond market

The narrative around AI investment is undergoing a transformation. It started as a technology story, then evolved into a capital expenditure story, and now is becoming a financing story.

Recently, Greg Jensen of Bridgewater summarized the current situation in one sentence: “We are entering a critical moment for capital.”

Morgan Stanley predicts that total AI infrastructure expenditure will reach 3.2 trillion USD by 2028, with approximately 1.75 trillion USD needing to be raised through credit markets. Sources of funding have expanded from traditional investment-grade bonds to leveraged loans, private credit, and securitized products. The money is available, but it won't come cheap.

The bond market has already started to react. According to Apollo data, AI-related bond issuance now accounts for 40% of long-duration supply. Credit spreads for Hyperscalers have widened significantly this year, while the overall investment-grade market has remained almost unchanged.

The “new Bond King”, Gundlach, put it bluntly, issuing long-term bonds collateralized by GPUs is akin to “making 30-year ABS with bananas.”

Cash flow is moving in the wrong direction

Capital expenditure forecasts continue to be revised upward, while free cash flow forecasts are consistently declining.

Morgan Stanley has drastically lowered 2027 free cash flow forecasts for major Hyperscalers. Oracle stands out in particular, with its 2027 free cash flow forecast now close to -40 billion USD.

The scale of spending commitments is equally staggering. Hyperscalers' purchase commitments have exploded to 982 billion USD. Morgan Stanley points out that a large amount of actual capital expenditure now exists off-balance-sheet. This means that simply looking at the balance sheet would seriously underestimate the real capital pressure.

The credit market is already pricing in risk

Investors are demanding higher risk compensation.

Morgan Stanley data shows that credit spreads for Hyperscalers have “widened significantly this year, far exceeding the overall investment-grade market”—spreads for high-quality Hyperscalers (HQ Hyperscalers) have widened by about 25 basis points, regular Hyperscalers by about 22 basis points, while the overall investment-grade market has seen no change.

Oracle is the most closely watched case. Credit default swaps (CDS) for other Hyperscalers are generally in the 30 to 80 basis points range, while Oracle's CDS soared from about 40 basis points in mid-2025 to nearly 190 basis points at its peak in April 2026, and currently hovers around 180 basis points. Meta’s CDS rose modestly to around 75 basis points. The credit market has made it clear which company concerns it the most.

Balance sheets remain strong, but the issue lies ahead

Morgan Stanley’s Q1 2026 data show that Hyperscalers’ total leverage is only 1.3x, net leverage is 0.5x, cash-to-debt ratio is as high as 128%, with a median rating of AA-. In comparison, overall non-financial investment-grade total leverage is 2.4x, with a rating of BBB.

But the problem is in the future. Morgan Stanley has sharply raised its forecast for cloud capital expenditure growth in 2027 from 14% to 29%, and Hyperscalers are “reaffirming their confidence in investment returns.” With forecasts for expenditures doubling, the financing gap is widening accordingly.

“New Bond King” Gundlach: Using GPUs as collateral is like making ABS with bananas

Regarding the emerging financing schemes in the market that use AI assets as collateral, renowned bond market investor Jeff Gundlach has issued sharp criticism.

He warned about a 50 billion USD fund consortium plan, saying it “will likely not stand the test of time.” On social media, he asked: “Using collateral with unknown lifecycle for long-term debt? Why not do a 30-year ABS deal backed by bananas in a warehouse?” He added that these are “brand-new, engineered bananas with unknown lifespans.”

Gundlach’s analogy points directly to the core problem: GPUs depreciate rapidly, with technological iteration cycles far shorter than the lifespan of debt. Using such assets as collateral for long-term financing introduces high uncertainty over the actual value of the collateral.