Banks are now trying to offload the AI data center debt they just underwrote

本文为原文前 6,000 字符的节选翻译，完整内容请查看原文。

JPMorgan and MUFG are part of the group that has spent more than six months trying to distribute $38 billion in construction debt tied to a Texas and Wisconsin data center project leased to Oracle.

摩根大通（JPMorgan）和三菱日联金融集团（MUFG）所在的财团已耗时超过六个月，试图分销与甲骨文（Oracle）租赁的德克萨斯州和威斯康星州数据中心项目相关的380亿美元建设债务。

JPMorgan, Morgan Stanley, SMBC and MUFG are quietly shopping pieces of the AI debt pile they helped build, a sign the lenders closest to the money have started hedging their own bet.

摩根大通、摩根士丹利（Morgan Stanley）、三井住友金融集团（SMBC）和三菱日联金融集团正在悄悄兜售他们参与构建的人工智能债务组合，这表明最接近资金的贷款机构已开始对冲自己的押注。

Six months ago, banks were racing each other to underwrite data center loans tied to Oracle, Meta and the rest of the AI buildout. Now some of the same banks are trying to get that debt off their books. JPMorgan, Morgan Stanley, SMBC and MUFG have been exploring significant risk transfers, according to a report from the Financial Times. These are deals that let a bank keep a loan on paper while shifting its riskiest slice to outside investors. The banks are doing this because they are bumping up against internal limits on how much AI-linked exposure any one lender can hold.

六个月前，各家银行还在竞相承销与甲骨文、Meta及其他人工智能基础设施建设相关的数据中心贷款。如今，其中一些银行正试图将这些债务从资产负债表中剥离。据《金融时报》报道，摩根大通、摩根士丹利、三井住友和三菱日联一直在探索大规模的风险转移。这些交易允许银行在账面上保留贷款，同时将风险最高的部分转移给外部投资者。银行这样做是因为它们已触及内部限制，即单一贷款机构所能持有的与人工智能相关的风险敞口上限。

One case shows how stuck this has gotten. A group of banks including JPMorgan and MUFG has spent more than six months trying to distribute $38 billion in construction debt tied to a Texas and Wisconsin data center project leased to Oracle, Bloomberg has reported. Some lenders are now willing to sell pieces of it to non-bank buyers at a discount, just to move it. Six months is a long time to sit on debt nobody underwrote to hold forever.

一个案例显示了这种情况陷入了何种僵局。据彭博社报道，包括摩根大通和三菱日联在内的一组银行已花费超过六个月的时间，试图分销与甲骨文租赁的德克萨斯州和威斯康星州数据中心项目相关的380亿美元建设债务。一些贷款机构现在甚至愿意折价将部分债务出售给非银行买家，只为将其脱手。对于本非长期持有而承销的债务来说，持有六个月已经是非常长的时间了。

Matthew Moniot, co-head of credit risk sharing at Man Group, told the Financial Times the deal sizes involved are “out of scale to anything we’ve thought about, ever.” That is not a hedge-fund manager talking down a trade he wants no part of. It is someone whose job is pricing exactly this kind of risk, saying the numbers no longer fit any model he has used before.

英仕曼集团（Man Group）信贷风险分担联席主管马修·莫尼奥特（Matthew Moniot）对《金融时报》表示，所涉及的交易规模“超出了我们以往所能想象的任何范畴”。这并非一位对冲基金经理在贬低他不想参与的交易，而是负责为这类风险定价的专业人士在表示，这些数字已不再符合他以往使用的任何模型。

Oracle is the clearest tell. Its five-year credit default swaps, the cost of insuring against the company defaulting, climbed to around 151 basis points in early October, levels last seen in 2009, according to market data cited by Bloomberg. S&P had already cut Oracle’s credit rating to BBB-, a single notch above junk, back in July. The company is reportedly targeting more than $40 billion in additional debt and equity this fiscal year. That is on top of the $18 billion in bonds it sold in a single day last September, and the money is needed to keep funding its Texas, Wisconsin and New Mexico sites.

甲骨文的情况最能说明问题。据彭博社引用的市场数据显示，10月初，其五年期信用违约掉期（即为公司违约投保的成本）攀升至约151个基点，达到2009年以来的最高水平。标普早在7月份就将甲骨文的信用评级下调至BBB-，仅比垃圾级高出一级。据报道，该公司本财年计划额外筹集超过400亿美元的债务和股权资金。这还不包括去年9月其在一天内发行的180亿美元债券，而这些资金是维持其德克萨斯州、威斯康星州和新墨西哥州项目运营所必需的。

Oracle sued Wisconsin’s Public Service Commission over a rule that could force it to post more than $7 billion in collateral for its Port Washington data center campus. The fight helped push S&P to cut Oracle’s credit rating to BBB-, one notch above junk, as the company’s AI spending outpaces its cash flow. - Oracle Wisconsin data center collateral dispute - utilities regulator $7 billion lawsuit

甲骨文就一项可能迫使其为华盛顿港（Port Washington）数据中心园区提供超过70亿美元抵押品的规定，起诉了威斯康星州公共服务委员会。由于该公司在人工智能方面的支出超过了现金流，这场纠纷促使标普将甲骨文的信用评级下调至BBB-（仅比垃圾级高一级）。——甲骨文威斯康星州数据中心抵押品纠纷——公用事业监管机构70亿美元诉讼案

Demand for that paper has thinned out, too. Bonds from Alphabet, Amazon, Meta, Microsoft and Oracle now trade at a premium to similarly rated non-AI corporate debt. Order books for their offerings have shrunk from roughly five times the amount on sale back in February to under two times by July, Bloomberg has reported. That is the difference between a bond that sells itself and one a bank has to work to place.

对这些债券的需求也已减弱。Alphabet、亚马逊、Meta、微软和甲骨文的债券目前的交易价格高于同评级的非人工智能企业债务。据彭博社报道，这些公司发行的债券订单簿已从2月份时约五倍的认购倍数缩减至7月份时的不到两倍。这就是“自动售出”的债券与银行必须费力推销的债券之间的区别。

None of this is happening in isolation. The 10-year Treasury yield hit 5.304% this month, its highest since 2002, after a $39 billion auction cleared at 5.30%, the highest auction yield since 2000. The Federal Reserve’s own September meeting minutes named AI-related corporate borrowing as a contributor to that move. Capital Economics argues the bigger driver is plain old inflation and government deficits, not tech company bond sales. Whoever is right about the cause, the effect is the same for anyone trying to finance a data center right now: the risk-free rate they are borrowing against just got a lot more expensive.

这一切并非孤立发生。本月，10年期美国国债收益率达到5.304%，创下2002年以来的最高水平，此前一次390亿美元的国债拍卖以5.30%的收益率成交，这是自2000年以来的最高拍卖收益率。美联储9月份的会议纪要将与人工智能相关的企业借贷列为推动这一走势的因素之一。凯投宏观（Capital Economics）则认为，更大的驱动因素是传统的通货膨胀和政府赤字，而非科技公司的债券销售。无论谁对原因的判断正确，对于目前任何试图为数据中心融资的人来说，结果都是一样的：他们所借贷的无风险利率已经变得昂贵得多。

Frankly, the mechanics here matter more than the headline number. Banks do not sell down debt they are confident in. They sell down debt when they are worried about concentration, which is exactly what “approaching internal limits” means in plain English: too much of this bank’s balance sheet is now one bet on one sector holding up. Spreading $120 billion of AI infrastructure spending across special purpose vehicles was supposed to make that risk easier to swallow. That is the structure Oracle, Meta, xAI and CoreWeave have all used to keep debt off their own balance sheets. Instead the risk is turning up at the banks that financed the SPVs in the first place.

坦率地说，这里的运作机制比头条数字更重要。银行不会出售它们有信心的债务。它们出售债务是因为担心集中度风险，这正是“接近内部限额”的通俗含义：该银行资产负债表中有太多的部分押注于单一行业的持续增长。将1200亿美元的人工智能基础设施支出分散到特殊目的载体（SPV）中，本应使这种风险更容易被消化。这就是甲骨文、Meta、xAI和CoreWeave用来将债务排除在自身资产负债表之外的结构。然而，风险最终却出现在了最初为这些SPV提供融资的银行身上。

The investors who still want in have gotten pickier about where. Capital is rotating toward data centers leased to investment-grade hyperscalers like Microsoft and Google, and away from weaker credits, Bloomberg has reported. That is exactly why CoreWeave-leased paper prices at 7.00% to 9.875%, against the 5.70% to 7.875% that hyperscaler-leased bonds command, per bond market data compiled by CCIR. A smaller, less capitalized AI infrastructure company trying to raise debt next quarter is now competing for the same nervous pool of lenders as Oracle. Except without Oracle’s balance sheet or its existing customer base to point to.

据彭博社报道，那些仍想入场的投资者变得更加挑剔。资本正转向租赁给微软和谷歌等投资级超大规模云服务商的数据中心，而远离信用较弱的借款方。根据CCIR汇编的债券市场数据，这正是CoreWeave租赁项目的债券定价在7.00%至9.875%之间，而超大规模云服务商租赁债券的定价在5.70%至7.875%之间的原因。一家规模较小、资本较少的人工智能基础设施公司如果想在下个季度筹集债务，现在必须与甲骨文争夺同一批紧张的贷款人。不同的是，它没有甲骨文那样的资产负债表或现有的客户群作为支撑。

That’s the part of this story that does not show up in a single day’s stock move. The AI boom still runs on the assumption that credit stays open indefinitely at reasonable terms. Six months of a bank struggling to sell $38 billion in loans is the clearest evidence yet that assumption is being tested in real time, not in some theoretical downturn some years away.

这是这个故事中无法在单日股价波动中体现的部分。人工智能热潮仍然建立在信贷将以合理条款无限期开放的假设之上。一家银行在六个月内难以售出380亿美元贷款，这是该假设正在实时受到考验的最明确证据，而不是在几年后的某个理论上的经济衰退中。

Also read: Trump-linked stablecoin USD1 adds $427 million in a single week • IREN shares sink 20% in ten days as a scathing infrastructure report collides with Wall Street’s AI bull case • Micron’s Board Authorizes a Record $35 Billion Stock Buyback

另请阅读：特朗普关联稳定币USD1单周增加4.27亿美元 • IREN股价十天内下跌20%，一份严厉的基础设施报告与华尔街的人工智能看涨观点发生碰撞 • 美光科技董事会批准创纪录的350亿美元股票回购计划

Oracle’s five-year CDS spread hit 203 basis points, an 18-year high, after S&P cut its rating to BBB-, while Meta’s latest bond drew a third less investor demand than its sale a year earlier. - bond investors concerned about AI infrastructure debt

在标普将甲骨文的评级下调至BBB-后，其五年期CDS利差达到203个基点，创下18年新高；与此同时，Meta最新发行的债券吸引的投资者需求比一年前减少了三分之一。——债券投资者对人工智能基础设施债务感到担忧