Big Tech's $236B AI Debt Binge Losing Investor Appeal

Hyperscalers—the giant AI companies racing to build data centers as rapidly as possible—have turned to the corporate bond market in unprecedented volume to fund their expansion. Morgan Stanley's analysis shows that as of May 31, AI-related global debt issuance had reached $236 billion（約38兆円）, a four-fold increase from one year prior. The bank forecasts this trajectory will accelerate to $570 billion（約91兆円） by the end of 2026, reflecting the scale of capital outlays ahead. Meta Platforms, Alphabet, Amazon, and Microsoft alone are projected to spend $700 billion（約110兆円） on capital expenditure in the current year, with the vast majority directed toward AI infrastructure.

Initially, this debt found eager buyers. Investors rushed to acquire bonds from these established, cash-generative technology giants. But that appetite has contracted sharply. According to Apollo Global Management, cover ratios for hyperscaler bonds—the ratio of asset demand to supply—have fallen from 5x in February to 2x in July. This decline signals that borrowing costs will rise: if these companies wish to continue issuing bonds, they will have to offer higher yields to compensate for weaker demand. The trend extends beyond traditional hyperscalers. SpaceX, which went public in June with stock initially bidding above $225 before settling near its $150 IPO price, illustrates broader investor hesitation around AI-adjacent debt. SpaceX's bonds trade at a 1.62-percentage-point premium over Treasuries, placing them in the non-investment-grade ("junk") category, despite the company carrying $29 billion（約4.6兆円） in long-term debt.

Bond investors' growing skepticism is noteworthy because they tend to prioritize capital preservation and focus on near-term cash flows rather than speculative upside. Changes in bond demand, pricing, and credit spreads often signal shifts in a company's financial health before those shifts become apparent in equity markets. The article suggests this dynamic warrants attention: while the AI revolution remains young and these massive investments could eventually deliver substantial returns, the tightening of debt market appetite may be an early warning that investors are questioning whether the scale and timeline of AI infrastructure buildouts can justify the financing costs being incurred.

Hyperscalers are caught in an escalating capital cycle: the sheer scale of AI data center buildouts has forced them away from relying purely on cash flows and into the bond market for funding. Morgan Stanley's projection of $570 billion（約91兆円） in AI bond issuance by end of 2026 underscores the magnitude of this shift. What makes this moment significant is the divergence between equity and bond investor sentiment—stock buyers initially rushed to snap up the offerings (as evidenced by early SpaceX stock enthusiasm), but bond investors, who prioritize capital preservation and near-term cash flow, are signaling caution through lower demand and wider credit spreads.

This caution is already visible in concrete metrics: the cover ratio collapse from 5x to 2x in just six months suggests a sharp erosion in appetite. SpaceX, while not a traditional hyperscaler, illustrates the broader pattern—its $29 billion（約4.6兆円） long-term debt now trades at a 1.62-percentage-point premium over Treasuries, placing it in junk territory despite the stock's initial rally. Bond investors' skepticism matters because they are typically early detectors of financial stress; equity investors who ignore these signals may be exposed to a repricing if the data center investments fail to generate expected returns quickly enough to service the mounting debt.