One US$500 Billion Deal: The Fee Collectors Rose, the Payers Fell

Hello. In the last piece Nvidia was out raising up to US$500 billion for AI infrastructure and the market had started asking where the money would come from. Last night we got half an answer: the money is there, but the people putting it up and the people spending it went in opposite directions.

The providers all rose: KKR up 6.88 per cent, Apollo 6.26 per cent, Brookfield 4.77 per cent, Blackstone 3.89 per cent and BlackRock 1.54 per cent.

The spenders fell for a second day: Alphabet down 3.61 per cent, its fourth decline in five sessions; $Amazon.com(AMZN)$ down 2.09 per cent, $Broadcom(AVGO)$ 1.5 per cent, $Apple(AAPL)$ 1.09 per cent and $Microsoft(MSFT)$ 0.44 per cent. $NVIDIA(NVDA)$ itself fell 0.02 per cent, which is to say it didn't move.

Someone at KKR explained the delight in one line: "think of it as a revenue stream, then securitise it." Six institutions are each building a compute-financing platform, Nvidia brings the customers, and it may choose to take a guarantee exposure of up to 25 per cent on each loan. For firms that manage money for a living, this is a new pipe that pays fees for decades.

The bull case is real: a one-gigawatt data centre costs about US$50 billion, and a buyer like OpenAI does not even have an investment-grade rating yet, so that gap has to be bridged by something like this. BlackRock's Larry Fink compared it to first encountering mortgage-backed securities in the 1970s.

The bear case is equally real, and it points at one thing: what the collateral is worth when the loan matures. Chip generations have gone from every two years to every year, and each new one marks down the last. The model has not been through a full cycle. And what was signed is a memorandum — no borrowers, rates, sites or start dates have been disclosed.

One place is already flashing. Five-year credit default swaps on Nvidia are up about 90 per cent this year and, even after narrowing 5 basis points to 72.11 last night, sit near their highs. The share price didn't budge; the cost of insuring it has nearly doubled.

In the last piece I wrote that the names due to report this week had been sold in advance. They were sold early: all three turned around, and all three in the same way.

$CoreWeave, Inc.(CRWV)$ rose 2.42 per cent in the session and jumped 15.37 per cent after hours. Revenue of US$2.575 billion came in US$36 million below the US$2.611 billion expected, but it was up 112 per cent year on year, the loss of US$1.03 a share beat the US$1.24 expected, and — the part that mattered — the order backlog passed US$100 billion, at about US$104 billion.

$SUPER MICRO COMPUTER INC(SMCI)$ rose just 0.45 per cent in the session and 7.72 per cent after hours. Revenue of US$11.12 billion missed the US$11.782 billion expected by more than US$600 million, yet earnings of US$1.70 a share came in well above the US$0.98 expected, with more than US$60 billion of new orders alongside.

$Lumentum(LITE)$ rose 0.87 per cent in the session and 3.56 per cent after hours. Revenue was US$1.01 billion, up 109 per cent, and guidance for the current quarter has a midpoint of about US$1.255 billion, some 8 per cent above expectations. The same statement carried a GAAP net loss of nearly US$7.2 billion — a one-off, non-cash charge from converting part of its convertible notes into equity, not a business turning bad.

Two of the three missed on revenue and the third only just beat. All three went up anyway, on the order book and the guidance.

What can be securitised was never how much was sold last quarter; it is the part signed on paper and not yet delivered. Compute is turning from selling equipment into signing long contracts, and a long contract is the thing you can take to a lender. On the same day, Anthropic and the miner Riot agreed a 20-year compute purchase: a base contract of about US$9.1 billion locking in 191 megawatts, rising to as much as US$16.1 billion if all extension options are exercised. Riot rose 4.33 per cent.

One name was the exception. Sea rose 14.56 per cent on neither an order book nor guidance, but on the quarter itself: pre-market revenue of US$7.805 billion beat the US$7.275 billion expected, and earnings of US$0.86 a share beat US$0.79. Among last night's results, it was the only one that did not have to talk about what it will collect later.

Memory is going the same way. $SK hynix(SKHY)$ rose 4.70 per cent, $SanDisk Corp.(SNDK)$ 2.68 per cent, $Micron Technology(MU)$ 0.87 per cent and $Direxion Daily Semiconductors Bull 3x Shares(SOXL)$ 2.31 per cent, clawing back part of the previous day's losses, with long-term memory contracts reported to be spreading from the leaders down to second- and third-tier makers. At the same time, Korean brokers cut their targets on Samsung and SK Hynix, the deepest by 30 per cent. Upstream conditions keep getting confirmed and the shares keep going sideways, and this is what is missing: nobody can say how long those contracts hold.

A batch of 13F filings also landed, putting the other half of the question — where the money is going — on the table. Berkshire bought about US$23.5 billion of listed stock in the second quarter against US$3.7 billion of selling, close to US$20 billion net, ending 14 straight quarters of net selling: the most conservative money in the market has started buying again. The successor has started to leave a mark too, with three new positions already about US$1.2 billion in the money. Citi's holdings passed US$300 billion, with notable additions in Micron and AMD. The share price did not play along: Berkshire's B shares fell 2.46 per cent.

Space cooled off, with $SpaceX(SPCX)$ down 3.93 per cent and one broker calling Rocket Lab's roughly 36 times forward sales a valuation trap. The indices went on hugging their highs: the S&P 500 fell 0.32 per cent and the Nasdaq-100 ETF (QQQ) 0.34 per cent.

July CPI comes at 20:30 Singapore time tonight. The Dow Jones survey of economists looks for 3.4 per cent year on year, down from 3.5 per cent, and 2.5 per cent for the core, down from 2.6 per cent. The same evening, Nebius reports pre-market and Coherent after the close.

Last night the market accepted the order book. Tonight the question is whether inflation plays along.

A note on sources: the moves for KKR, Apollo, Brookfield, Blackstone, BlackRock, Amazon and Broadcom, and the Nvidia CDS quote, are cited from market reports rather than this piece's own price feed; Lumentum's US$7.2 billion loss is a one-off, non-cash item. The above is personal analysis, not investment advice.

💬 【Talking Point】

Berkshire just ended 14 straight quarters of net selling. Is the most conservative money in the market early, or late?

💰 【Bounty】

Drop your view in the comments and there are coins in it for you! 🎁

🔔 Better shared than saved — tag a friend and split the coins!

# Indices Pause at Highs Ahead of CPI — Is 3.4% the Pass Mark?

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Report

Comment7

  • Top
  • Latest
  • Investing Leon
    ·22 minutes ago
    Interesting that the “fee collectors” rallied while the actual AI spenders fell. The market clearly likes the financing opportunity, but it still hasn’t decided whether all this AI capex will generate enough returns.
    Reply
    Report
  • 北极篂
    ·08-12 17:51
    所以我的判断是,伯克希尔这次更像是一个“风险偏好开始改善”的信号,而不是全面看多的号角。
    Reply
    Report
  • 北极篂
    ·08-12 17:51
    我反而更关注另一个变化:资本市场正在从“讲故事”逐渐转向“看现金流”。 AI基础设施可以有巨额订单,但最终必须证明这些订单能够产生收入、利润和自由现金流,否则融资成本和信用风险迟早会成为估值压力。
    Reply
    Report
  • 北极篂
    ·08-12 17:51
    但我不会把这理解成“市场已经到底”或者新一轮牛市确认。伯克希尔买入的时间点和具体标的,才是关键。如果资金明显流向金融、科技等具备长期现金流的行业,更像是在利用市场波动寻找价值,而不是追逐AI情绪。
    Reply
    Report
  • 北极篂
    ·08-12 17:50
    巴菲特体系一直强调安全边际和现金储备,因此当全球市场估值已经不低、AI资本开支又处于历史高位时,伯克希尔重新增加股票配置,至少说明他们看到了一些值得承担风险的价格。
    Reply
    Report
  • 北极篂
    ·08-12 17:50
    我认为,伯克希尔结束连续14个季度净卖出,这个信号的意义其实比单纯“买了235亿美元股票”更值得关注。
    Reply
    Report
  • 靖润
    ·08-12 16:25
    今晚CPI只是一道开胃菜,这5000亿美元的交易才是真正的主菜。市场已经在定价了:钱到位了,但花出去的效率开始被质疑。


    数据不会说谎: KKR涨6.88%、阿波罗涨6.26%、Brookfield涨4.77%、黑石涨3.89%、贝莱德涨1.54%。提供资金的人股价在上涨,说明市场认为这笔钱借出去能产生稳定的利息收益。另一边,Alphabet跌3.61%、亚马逊跌2.09%、博通跌1.5%、苹果跌1.09%、微软跌0.44%。花钱的人股价在下跌,说明市场开始质疑这些资本开支的回报效率。


    英伟达自己跌0.02%,横着不动,说明市场还没决定它是收钱的还是花钱的,它在中间。资金供给方在涨,技术支出方在跌,这个分化已经足够清晰,不需要等任何数据来确认方向。市场正在对“循环融资”结构进行定价:借钱的赚利息,花钱的被质疑,夹在中间的观望。这是对AI基建效率的第一次系统性重新定价,不会在一天内完成。


    CPI今晚是次要变量。 如果温和,分化趋势不会变,只是上涨的时间会提前。如果超预期,分化加剧,因为高利率环境下资本开支效率会更受审视。短期内,KKR、阿波罗、Brookfield、黑石这类资产不会出现大幅调整,Alphabet、亚马逊、博通、苹果、微软这些科技巨头可能还会面临持续的压力。


    CPI数据只是影响节奏,不会改变这个分化趋势。利率预期转向的方向已经足够清晰,剩下的只是时间问题。伯克希尔连续14个季度净卖出后开始买进,这是最保守的钱在进场。它们不是来追涨的,是来收息的。当最保守的钱开始进场,说明这个方向已经没有太多争议了。
    Reply
    Report