JPMorgan 분석에 따르면 AI 기업들의 채권 발행이 올해 신규 장기 미국 국채 차입의 68%에 해당합니다.

The AI infrastructure boom has a scale problem the bond market cannot ignore. Michael Cembalest, chairman of market and investment strategy at JPMorgan, estimates the five major hyperscalers plus Nvidia have issued roughly $320 billion of debt so far in 2026, including special purpose vehicles where the companies ultimately stand behind data center lease obligations. In 10-year-equivalent terms, Cembalest puts the long-duration component at roughly $303 billion—equivalent to around 68% of new long-duration Treasury borrowing this year.

Six companies—Oracle, Microsoft, Amazon, Alphabet, Meta Platforms, and Nvidia—now compete directly with the U.S. Treasury for the same pool of long-duration bond demand.

The figure gains credibility from independent analysis by Goldman Sachs, which estimates roughly $300 billion of AI-related issuance already completed this year, with senior hyperscaler and chip supply slowing sharply into year-end. The two bank analyses converge on the same magnitude.

The stress emerges when looking forward. Goldman credit trader Jeffrey Papai identifies 2027 as the real stress test, with roughly $340 billion of senior hyperscaler and chip issuance projected for the coming year, before another substantial layer of data center and structured chip financing is added on top.

The structural composition of the pipeline drives the concern more than headline size. Limited hyperscaler maturities mean far less natural recycling of investor capital than in sectors like banking, leaving markets dependent on fresh demand for fresh supply. A market financed by reinvested proceeds absorbs shocks; a market financed by net new inflows depends on appetite that can shift rapidly. If AI issuers absorb a meaningful share of long-duration capacity, the marginal buyer of a 30-year corporate bond and the marginal buyer of a long Treasury become increasingly the same investor—determining the clearing price for mortgages and long yields alongside AI capex decisions.

The counterargument rests on the enormous cash flows behind these balance sheets. Bank of America noted the big five hyperscalers' debt-to-cash ratio improved from 0.94 to 0.75. By 2029, Bank of America projected operating cash flow reaching $1.1 trillion, a 95% increase. For most of these issuers, this represents balance-sheet optimization by cash-rich businesses.

Oracle stands as the exception. Bank of America projects Oracle running negative free cash flow until 2029, meaning capex will exceed cash from operations, leaving little debt capacity. Oracle's fiscal 2026 capital expenditures reached $55.7 billion against operating cash flow of roughly $32 billion, with the company planning to raise around $40 billion in debt and equity during fiscal 2027. Oracle shares are down 17.91% year to date as of Wednesday's close.

Law360 reported on January 14, 2026 that Oracle faced a lawsuit by a pension plan over AI-linked debt disclosures, with the complaint alleging deficiencies in disclosure tied to the AI buildout.

The aggregate picture remains manageable, though dispersion across issuers is wide. Oracle is where the risk concentrates, and hyperscaler spreads versus Treasuries are the key metric to watch as 2027 supply approaches, according to Goldman Sachs.