英伟达与高盛、阿波罗全球、KKR、布鲁克菲尔德和贝莱德等华尔街主要金融机构达成协议，将建立一个5000亿美元或更多规模的资本池，用于资助客户购买英伟达芯片[1]。这一融资方案将通过创建资产支持证券供投资者购买的方式实现，英伟达将为任何项目抵押品的剩余价值提供25%的担保[1]。值得注意的是，各方签署的是谅解备忘录而非确定的信贷额度，每个具体项目仍需单独进行评估[1]。
这项安排背后存在多项风险因素。英伟达芯片的价值存在不确定性，应在3至5年内完全折旧，尽管英伟达主张折旧周期可达10年[1]。与此同时，包括谷歌、Meta、亚马逊、微软和甲骨文在内的超大规模云计算公司今年的投资约达7500亿美元，预计明年将超过1万亿美元[1]。这些公司的人工智能相关支出增速已严重滞后于投资增速，预计明年将出现现金赤字并高度杠杆化[1]。此外，人工智能相关债券发行已达约3500亿美元，而AI公司的表外负债估计超过1.6万亿美元，远超资产负债表上的债务[1]。中国正在战略性瞄准推理芯片领域，其芯片价格已显著低于美国同类产品，这也可能对融资项目构成挑战[1]。
Nvidia has reached an agreement with major Wall Street firms including Goldman Sachs, Apollo Global, KKR, Brookfield, and BlackRock to establish a capital pool of $500 billion or more to finance customers' purchases of Nvidia chips.[1] The arrangement, formalized through a memorandum of understanding rather than a formal credit facility, will create asset-backed securities for investor purchase, with each project subject to individual evaluation.[1] Nvidia will guarantee that the residual value of collateral in any transaction retains at least 25% of its initial worth.[1]
The financing structure faces significant headwinds as AI chip valuations remain uncertain and subject to rapid depreciation.[1] Nvidia estimates chips should fully depreciate within 3 to 5 years, though the company has suggested a lifespan extending to 10 years.[1] Meanwhile, major cloud infrastructure operators—Google, Meta, Amazon, Microsoft, and Oracle—are projected to spend approximately $750 billion on AI-related investments this year, with outlays expected to exceed $1 trillion next year.[1] These hyperscale companies are anticipated to run cash deficits and become highly leveraged in the coming year.[1] The broader AI financing market has already seen roughly $350 billion in bond issuances, while off-balance-sheet liabilities of AI companies are estimated to exceed $1.6 trillion, surpassing the debt recorded on their financial statements.[1] Adding to competitive pressures, China has strategically focused on inference chip development, with Chinese chip prices now substantially undercutting comparable American products.[1]