字号·· | 护眼
彭博社

摩根士丹利警告:高风险AI债务交易面临“更高门槛”Morgan Stanley Warns ‘Higher Bar’ Is Set for Risky AI Debt Deals

点「原文对照」整页切到原文,或双击某段只看那段的原文。

摩根士丹利纽约办公室的标识。摄影师:迈克尔·纳格尔/彭博社Signage at the Morgan Stanley office in New York. Photographer: Michael Nagle/Bloomberg

摩根士丹利表示,由于借贷成本上升以及潜在的建设延误抑制了活动,与人工智能相关的杠杆债务发行预计将保持“低迷,新交易门槛更高”。

Issuance of leveraged debt tied to artificial intelligence is expected to remain “subdued, with a higher bar for new deals” as higher borrowing costs and potential construction delays weigh on activity, according to Morgan Stanley.

该银行最新报告显示,10月AI相关杠杆融资发行总额仅为50亿美元。这50亿美元全部来自单一借款人Volta Infrastructure Holdings Ltd.,该公司通过定价收益率超过11%的期限贷款筹集了资金,彭博新闻此前报道过。

AI-related leveraged finance issuance totaled just $5 billion in October, according to a new report from the bank. All $5 billion came from a single borrower, Volta Infrastructure Holdings Ltd., which raised the money through term loans that priced at a yield of more than 11%, Bloomberg News previously reported.

该银行预计,该市场的发行量在年底前将保持有限。摩根士丹利策略师写道,美国中期选举后,依赖高收益债券和杠杆贷款的公司在为新建设项目争取电力和审批方面可能面临更多障碍。这可能改变这些公司投资的时间和规模。

The bank expects issuance in that market to remain limited through year-end. After the US midterm elections, companies that rely on high-yield bonds and leveraged loans could face more hurdles securing power and approvals for new construction, Morgan Stanley strategists wrote. That could change when and how much those companies invest.

摩根士丹利表示,为了完成交易,开发商可能需要向贷款人支付更多费用以承担建设风险,在再融资前偿还更多本金,或锁定信用更好的租户。投资者更加仔细地审查新建设项目,信用评级较低的租户支持的项目可能尤其难以融资。

To get deals done, developers may have to pay lenders more to take on construction risk, repay more principal before refinancing or line up tenants with stronger credit, Morgan Stanley said. Projects backed by lower-rated tenants could be particularly difficult to finance as investors scrutinize new construction more closely.

摩根士丹利还预计,数据中心建设债券以及所谓“新云”公司发行的债务将出现更多波动。新云公司是CoreWeave和Nebius等专业提供商,出租AI芯片使用权。这些公司高度依赖外部资本,运营历史有限。随着投资者通过由更强大公司支持的交易获得更多融资芯片的机会,贷款人可能会对保护较少的交易变得更加挑剔。

Morgan Stanley also expects more volatility in data center construction bonds and in debt issued by so-called neoclouds, specialized providers such as CoreWeave and Nebius that rent out access to AI chips. These companies rely heavily on outside capital and have limited operating histories. As investors get more opportunities to finance chips through deals backed by stronger companies, lenders could become choosier about deals with fewer protections.

据摩根士丹利称,截至2026年,人工智能和数据中心债券发行额达到6460亿美元,约为2025年全年筹集的2160亿美元的三倍。该银行对最大云服务提供商2027年合并资本支出预测在中期选举后保持不变,并预计这些公司将在明年引领人工智能相关融资需求的增加。

AI and data center debt issuance reached $646 billion so far in 2026, about triple the $216 billion raised in all of 2025, according to Morgan Stanley. The bank’s combined 2027 capital spending forecast for the biggest cloud providers held steady after the midterms, and it expects those companies to lead an increase in AI-related financing needs next year.

阅读媒体原文 ↗