S&P, 하이퍼스케일러 AI 지출이 1조 3천억 달러를 넘어 현금 흐름이 마이너스라고 전망

세계 최대 하이퍼스케일러들의 총 자본 지출이 2027년까지 1조 3천억 달러를 초과할 것으로 예상되며, 여섯 기업 모두 2026년과 2027년에 자유 영업 현금 흐름이 마이너스로 전환되고 2029년 이전에는 회복이 예상되지 않는다고 S&P Global Ratings가 2026년 8월 27일 발표한 AI 구축 자금 조달에 관한 새로운 연구에서 밝혔습니다.

보고서 “S&P Global Ratings’ View On Artificial Intelligence And Hyperscalers”는 Alphabet (GOOG ), Amazon (AMZN ), Microsoft (MSFT ), Meta, Oracle (ORCL ) 및 SpaceX의 자금 조달 모델을 조사합니다. 핵심 발견은 지출 규모보다 누가 비용을 부담하는가에 초점이 맞춰져 있으며, 영업 현금 흐름이 더 이상 비용을 충당하지 못해 이 여섯 기업이 건설을 지속하기 위해 부채, 주식 발행, 임대 약정 및 기타 금융 조치를 더 많이 활용하고 있다는 것입니다.

“As AI infrastructure investment accelerates, the focus is expanding beyond the scale of spending to the funding models, financial commitments and long-term implications that accompany it,” said Naveen Sarma, Managing Director and Sector Lead at S&P Global Ratings. “Understanding how these investments are financed and managed will be increasingly important in assessing credit quality across the sector.”

현금 흐름에서 신용으로의 전환

The financing mix is where the report does its real work. S&P flags the growing use of joint ventures, special purpose vehicles, residual value guarantees and similar structures, and says they are increasing the complexity of credit analysis. A residual value guarantee, in this context, is a commitment by the hyperscaler to make a lessor or financing partner whole if servers, GPUs or data-center assets are worth less than expected at the end of a lease term — an obligation that sits off the headline debt figure but behaves like debt if hardware values fall.

That structure matters for anyone holding hyperscaler bonds or tracking the companies leasing capacity to them. Contractual commitments and other debt-like obligations are among the items S&P says it is now monitoring, alongside monetization of AI investments, demand durability and overcapacity risk. The more of the buildout that is funded through vehicles and guarantees rather than retained cash, the more the credit exposure disperses across lessors, lenders and joint-venture partners rather than sitting visibly on one balance sheet.

S&P’s models generally assume a 2028 inflection point, with revenues accelerating and capital expenditure growth moderating as monetization improves. Until that point, the agency’s projection leaves a multi-year window in which the six largest buyers of compute are cash-consumers rather than cash-generators on a free operating basis: the period in which the equipment vendors, landlords and power providers on the other side of their purchase orders book the revenue. Securities.io has tracked that flow-through at the site level, from Digital Realty’s 50-megawatt Singapore allocation to Bitdeer’s $400 million of contracted AI capacity inside a 350-megawatt Malaysia build-out, and at the equipment layer in coverage of Vertiv’s cooling and power exposure to the same order cycle.

현금 흐름표가 이미 보여주는 내용

The negative-cash-flow projection is a forecast, but the direction is visible in current filings. Microsoft, one of the six named in the report, spent $115.9 billion on additions to property and equipment in its fiscal year ended 2026년 6월 30일, against $182.9 billion in net cash from operations (up from $64.6 billion of additions a year earlier), per its 2026년 7월 29일 earnings release. That 80% jump in a single year still left Microsoft cash-flow positive after capex; S&P’s projection implies that even companies of that scale stop self-funding as the 2026–2027 tranches land, and that the smaller balance sheets in the group get there sooner.

Microsoft’s filing also illustrates the demand side of S&P’s 2028 thesis: commercial remaining performance obligation, contracted business not yet recognized as revenue, rose 84% year over year to $678 billion as of 2026년 6월 30일, and Azure revenue surpassed $100 billion for the fiscal year. Those contracted backlogs are the revenue base the agency’s models are counting on when they assume monetization catches up with construction.

보고서 자체의 한계

S&P states the report does not constitute a rating action. No hyperscaler’s credit rating changed on its publication, and none of the six names carries a new outlook as a result of it. The projections are the agency’s own model assumptions, including the 2028 inflection and the 2029 cash-flow recovery, and the report itself lists the conditions that would stress them: AI monetization arriving slower than modeled, demand proving less durable, or overcapacity emerging if buildout outruns usage.

For bondholders and the financing counterparties now carrying residual value guarantees and lease commitments, the operative dates in the analysis are 2027 (when cumulative capex crosses $1.3 trillion), 2028, the modeled revenue inflection, and 2029, the first year S&P projects positive free operating cash flow returns across the group.