CPI Came In Bang On. What Rallied Was Nebius, Not Meta

Hello. The figure this market had spent two days sitting still for landed last night, and all four parts of it came in on the nose: July CPI was 3.4 per cent year on year and 0.1 per cent on the month, with the core at 2.5 per cent and 0.2 per cent. Traders trimmed their bets on a September rate rise, with the odds easing to about 33 per cent.

The gate opened. The water did not run towards the mega-caps.

Of the Magnificent Seven, only $NVIDIA(NVDA)$ rose, up 3.03 per cent. $Meta Platforms, Inc.(META)$ fell 3.38 per cent, $Microsoft(MSFT)$ 2.26 per cent, $Amazon.com(AMZN)$ 1.83 per cent, $Tesla Motors(TSLA)$ 1.59 per cent and $Apple(AAPL)$ 0.87 per cent, while $Alphabet(GOOG)$ barely moved. The indices duly twitched: $S&P 500(.SPX)$ rose 0.26 per cent, $NASDAQ(.IXIC)$ Composite 0.54 per cent, and $Dow Jones(.DJI)$ fell 0.04 per cent.

What did get bought was a different set. $NEBIUS(NBIS)$ surged 34.14 per cent, $CoreWeave, Inc.(CRWV)$ rose 19.28 per cent, $SUPER MICRO COMPUTER INC(SMCI)$ 19.02 per cent and $Lumentum(LITE)$ 13.63 per cent.

Memory was strong across the board too: $SK hynix(SKHY)$ rose 9.01 per cent, $SanDisk Corp.(SNDK)$ 5.76 per cent and $Micron Technology(MU)$ 4.92 per cent, taking Micron's market value back above US$1 trillion. $CSOP SK Hynix Daily (2x) Leveraged Product(07709)$ rose 14.28 per cent and $Direxion Daily Semiconductors Bull 3x Shares(SOXL)$ 6.89 per cent.

What these names have in common is that they have just reported.

And what they handed in has changed. $NEBIUS(NBIS)$ posted second-quarter revenue of US$582 million, up 454 per cent, with an adjusted net loss of US$33.2 million, 64 per cent narrower than a year ago — and the sell-side language moved with it, from watching the order backlog to watching the cash flow break out.

Super Micro's 19 per cent was not about revenue, which at US$11.12 billion still missed the US$11.782 billion expected. It was about a stronger gross-margin outlook — and gross margin was exactly what the market had been worried about going in. CoreWeave simply banked more of the previous night's after-hours move in the regular session.

A day earlier this market was content to pay for an order book. Last night it asked for one thing more: the order book had to turn into cash flow.

The reverse case makes it clearer. Three more reported after the close last night and were sold straight after.

$Coherent(COHR)$ closed up 8.24 per cent and fell 2.71 per cent after hours to US$346, having beaten on both revenue and profit with guidance that was not weak either. Cerebras closed up 11.63 per cent, raised its full-year outlook, and fell 16.43 per cent after hours.

$Cisco(CSCO)$ is the clearest case. It beat on revenue and profit, booked US$4 billion of AI infrastructure orders in the quarter and US$9.3 billion for the financial year, guided fiscal 2027 revenue above Wall Street forecasts and put current-quarter revenue at US$18.0 billion to US$18.2 billion. Not one line of it was bad. The shares still fell 4.07 per cent after hours, to US$118.84.

There is no ready explanation for that 4 per cent. The one thing on view is that the share price had already had its run before the report — up 2.86 per cent on the day, up 60 per cent for the year, on 40 times earnings, and closing within 5 per cent of its 52-week high.

The names that were bought look like the mirror image. Nebius and $SUPER MICRO COMPUTER INC(SMCI)$ went in as the doubted ones, and the gross margin the market feared most at Super Micro was exactly where it produced a better signal.

So the dividing line last night was not whether a company beat. $Cisco(CSCO)$ sits on the wrong side of it twice over: its order book jumped, and an order book on its own is no longer enough; its share price had already run, and that part has to be paid back. This is the same thing that happened to AMD on 4 August — record revenue of US$11.5 billion and data centre revenue doubled year on year, the shares up 7.00 per cent at the close and then down 8.68 per cent after hours, because "solid" was already in the price.

A correction to my own wording first. I had Nvidia's 25 per cent as a choice "to take a guarantee exposure of up to 25 per cent on each loan". The accurate version is that it guarantees these chips will retain at least 25 per cent of their residual value over the lease term. That promise is the foundation of the whole US$500 billion structure — packaging GPU leases into securities is a bet that AI chips will not depreciate the way ordinary technology products do.

The bear case comes down to one point. Loans in this market are repaid in full over three to five years, and that convention itself assumes the underlying asset is worth close to nothing afterwards. There is also no secondary market. Cars and commercial aircraft have mature channels after a default; once a chip is technologically obsolete, there is almost no reliable buyer.

The bull evidence isn't thin either. The A100 was launched six years ago and is still in use; GPU lease rates have a decade of price history behind them; and Nvidia keeps updating Cuda, deliberately extending the life of older cards. The financing structure borrows its tranching straight from CLOs.

$CoreWeave, Inc.(CRWV)$ supplied a piece of corroboration: it warned that if it had to give up Nvidia's chips, it would face difficulties. The tighter the ecosystem lock, the steadier the residual value may be — and the more the value of these assets depends on a single supplier.

The 13F filings filled in a gap. Of Berkshire's nearly US$20 billion of net buying in the second quarter, about US$10 billion went into $Alphabet(GOOG)$ — the most conservative money in the market putting its single largest bet on Alphabet, which barely moved last night.

Citi's holdings passed US$300 billion, with notable additions in $Micron Technology(MU)$ and $Advanced Micro Devices(AMD)$.

In Hong Kong, $TENCENT(00700)$ reported: second-quarter revenue of RMB204.79 billion, up 11 per cent, and adjusted net profit of RMB68.42 billion, up 9 per cent, with AI lifting marketing services revenue by 22 per cent; on a GAAP basis, though, net profit rose just 0.7 per cent, short of expectations. Capital expenditure rose to RMB52.78 billion and free cash flow turned negative at RMB13.8 billion, though it was still positive at RMB37.6 billion excluding prepayments for compute.

The shares fell 3.03 per cent in today's Hong Kong session, while SMIC rose 2.67 per cent and Hua Hong 2.45 per cent — here too the money went round the platform and bought capacity instead.

$SpaceX(SPCX)$ rose 9.65 per cent after Musk said AI would become its largest business in September, a date far more aggressive than the market had assumed.

Gold futures rose 2.02 per cent to US$4,471 last night and silver 1.48 per cent to US$65.73.

$Applied Optoelectronics(AAOI)$ rose 4.29 per cent last night and reports after the close tonight, with the market looking for revenue of about US$9 billion, up some 23 per cent year on year. It is the most direct test of the story that the AI giants are building their own capacity — all those expansion announcements have to turn into equipment orders before they count.

CPI has opened the gate on rates. But last night the money wanted two things: cash that has already arrived, and good news the share price has not already spent.

The above is personal analysis, not investment advice.

💬 【Talking Point】

Nvidia guarantees these chips keep at least 25 per cent of their residual value. Cars and aircraft have resale markets after a default; obsolete GPUs have almost none. Is that guarantee the foundation of the deal, or its weakest point?

💰 【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!

# CRWV, NBIS Surge Post-Earnings — Has AI Compute Hit Its Inflection Point?

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

Comment1

  • Top
  • Latest
  • 靖润
    ·08-14 16:21
    你问的这个问题,恰好戳在这笔5000亿交易最核心的矛盾点上。


    汽车和飞机都有成熟的二手车市场,残值是市场交易出来的,不是厂商承诺出来的。英伟达说“芯片至少保留25%剩余价值”,本质上是在自己兜底风险,帮客户把折旧从算力成本里剥离出来,让AI芯片变成可融资资产。它的真实意图是降低客户采购的门槛,加速算力基础设施的扩张,而不是在给客户提供安全垫。


    这个保证的定价基础是算力需求持续增长,而GPU的供给不会过剩。如果下一代GPU性能提升足够大,上一代产品的残值就会被市场重新定价。真正的风险在于技术迭代,如果下一代GPU的性能代差足够大,上一代产品的残值就会被市场重新定价,英伟达自己也兜不住底。在市场消化供给冲击的时候,英伟达只能在回购二手芯片和价格补贴之间选择一个,撑住残值预期的成本会挤占其他业务的利润空间。


    它是交易的坚实基础,因为残值保证让客户敢于下长期订单;也是交易的潜在弱点,因为一旦技术迭代速度超过市场需求的消化速度,这个保证就会变成一项负债。当前的方向已经足够清晰,但技术迭代速度仍然是一个需要在后续财报中持续观察的关键变量。
    Reply
    Report