
이미지: The Decoder
Summary
- NVIDIA has signed a letter of intent with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to raise more than $500 billion for AI infrastructure
- To reassure investors, NVIDIA agreed to directly guarantee up to 25% of the residual value of its own chips installed in the projects
- NVIDIA's stock fell about 1.4% after the news broke, and figures including Michael Burry have previously flagged GPU depreciation risk
- 조달 목표액
- 500억달러 이상(수년에 걸친 누적 목표)
- 참여 금융사
- 아폴로, 블랙록, 블랙스톤, 브룩필드, 골드만삭스, KKR
- 잔존가치 보증 비율
- 프로젝트별 최대 25%
- 주가 반응
- 발표 후 약 1.4% 하락, 시가총액 700억달러 이상 감소
- 별도 협상 건
- 오하이오 소재 10기가와트 데이터센터(오픈AI 임대분) 보증 협상 중
NVIDIA is now guaranteeing the value of the very chips it manufactures. The company has signed a letter of intent with six major financial firms—Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—to bring in more than $500 billion in third-party capital for building data centers, semiconductor fabs, and power plants. Shortly after the Financial Times first reported the deal, NVIDIA's stock dropped about 1.4%, wiping out more than $70 billion in market capitalization.
CEO Jensen Huang described the initiative on X as "a shift from one-off projects to a repeatable financing platform." The idea is to treat data centers as productive infrastructure, akin to power grids or transportation networks, opening up capital access for AI companies that have compute demand but struggle to secure funding at the necessary scale. The $500 billion figure he cited is neither NVIDIA revenue nor a single fund, but a cumulative target spanning several years; specific terms, individual participation sizes, and implementation timelines have not been disclosed.
The residual-value guarantee is the core of the arrangement. One reason financial institutions have been reluctant to invest in data centers is the difficulty of knowing what GPUs will be worth as assets a few years down the line. NVIDIA has agreed to absorb part of that uncertainty. If the resale or recycling value of installed chips falls short of expectations by the time financing concludes, NVIDIA will cover up to 25% of the shortfall on a per-deal basis. Core underwriting decisions—such as customer creditworthiness, demand, and utilization rates—remain with the financing firms, but the manufacturer itself now bears the risk of the chips' own depreciation. Huang emphasized that this guarantee ratio is "significantly lower" than in other computing finance arrangements. NVIDIA is also reportedly negotiating separate guarantees for a 10-gigawatt data center in Ohio and a facility being leased to OpenAI.
The move reads as an implicit response to criticism raised by investor Michael Burry, who has called hyperscalers' GPU depreciation practices "one of the more common frauds of modern times," arguing that GPUs become obsolete far too quickly to justify useful lives of five to seven years. The Bank of England has also warned that a downturn in the AI sector could spread risk across the broader financial system. Data center construction costs have already ballooned into the trillions, with the bulk going toward GPU purchases. For financiers, who guarantees the future value of those chips has become a decisive factor in whether deals get done.
NVIDIA's latest move extends a broader trend of the company expanding from GPU supplier to infrastructure finance architect. As the capital required to build a single data center outstrips what individual companies can carry on their balance sheets, the next thing to watch is whether chipmakers directly providing credit enhancement becomes an industry standard.



