To The Moon

Hello. Three companies reported on Tuesday. Put the three side by side and no order makes sense.

$Fabrinet(FN)$ beat on revenue by about 3.1 per cent and on earnings per share by about 7.5 per cent, both records — and closed down 19.38 per cent at US$482.59.

$Home Depot(HD)$ beat on both lines, posted net sales of about US$47.9 billion and its best comparable sales since 2022, and reaffirmed full-year guidance — and closed down 0.12 per cent.

$BIDU-SW(09888)$ missed on both revenue and profit, with earnings per share about 22.7 per cent short — and closed down 12.73 per cent.

Full marks fell 19 per cent, a pass fell 0.12 per cent, a fail fell 12.7 per cent. The line between results and share price snapped that day.

What snapped was not the numerator. It was the denominator.

The 30-year Treasury yield sat at 5.28 per cent, having touched 5.31 per cent intraday on Monday, its highest since 2007. A long-dated yield is the ruler you use to mark down money that only arrives later — move the ruler and the further away the money is, the deeper the mark-down. And once a risk-free yield is above 5 per cent, bonds start competing with equities for the same money.

US stocks fell for a third straight session: $Invesco QQQ(QQQ)$ down 1.69 per cent, $NASDAQ(.IXIC)$ Composite 1.33 per cent, $SPDR S&P 500 ETF Trust(SPY)$ 0.68 per cent and $Dow Jones(.DJI)$ 0.22 per cent.

Fabrinet came off worst not because the results were bad but because the quality of them got questioned. Coverage put the fall down to three things: gross margin, free cash flow and guidance. Its gross margin is 12.06 per cent — contract manufacturing is thin work however much volume goes through it. Lift the denominator and the first question becomes what that growth is worth in margin. Coherent beat on both lines a few days earlier and closed down 7.99 per cent the next day. Same question.

Baidu gives the same answer from the other end. Its AI cloud infrastructure business grew 50 per cent year on year, but the drag from advertising ran deeper, and one write-up simply called the results the price of going all in on AI. The part that was growing sits in the future, the part that was shrinking sits in the present, and the market only counted the second.

Memory turned over completely. $SanDisk Corp.(SNDK)$ closed down 9.01 per cent at US$1,625.78, handing back the whole of Monday's 8.88 per cent; $Tradr 2X Long SNDK Daily ETF(SNXX)$ fell 17.84 per cent while $Tradr 2X Short SNDK Daily ETF(SNDQ)$ rose 18.00 per cent. $SK hynix(SKHY)$ closed down 9.20 per cent and $Micron Technology(MU)$ 7.02 per cent.

The fuse was lit in Asia, where Kioxia fell about 7.5 per cent in Japan. What widened the fall was a single sentence: one house set out a comparison the same day of what is over-owned and what nobody has bought, and concluded that Nvidia remains under-owned while SanDisk is a crowded trade. Arriving after a month of gains in memory, that handed leveraged money its reason to cut.

$Direxion Daily Semiconductors Bull 3x Shares(SOXL)$ fell 14.80 per cent and $Direxion Daily Semiconductors Bear 3x Shares(SOXS)$ rose 15.20 per cent, almost mirror images. This was not an accident at one company but a whole board being marked down together: $Marvell Technology(MRVL)$ fell 7.82 per cent, Intel 6.58 per cent and $Advanced Micro Devices(AMD)$ 4.27 per cent.

One concession is due here. If everything came down to the discount rate, Micron would not make sense: the same day Bank of America raised its target to US$1,550 and another house upgraded it on the grounds that "inventories don't lie", and the shares still fell 7.02 per cent.

The two sides are working different problems. The sell side works the numerator — inventories, contract prices, the tightness in HBM. The market works the denominator. Micron's earnings peak is pinned somewhere around 2027, and its next results are not due until 22 September, with nothing in between to verify the numerator. At times like that the denominator is louder.

The only real gain of the day was Apple, up 1.45 per cent at US$310.03. It handed in no results that day. What it handed in was an agreement: the EU core technology commission cut to 5 per cent, the fee on alternative payment channels reset to 20 per cent, the rules on third-party app stores loosened, and a long-running antitrust dispute with Brussels settled.

A fee you can put a number on, in exchange for a fine you cannot. When the denominator is rising, what the market least wants to pay for is a figure nobody can put down.

$Meta Platforms, Inc.(META)$ is the mirror of that sentence. It closed down 4.45 per cent, the worst of the Magnificent Seven. Twenty-nine states are suing over the way Facebook and Instagram are designed and the harm done to young users; the trial opened on Tuesday and company executives were on the stand from day one. Coverage has put potential damages as high as US$1.4 trillion — the scale the plaintiffs are asking for, not a verdict, but enough for the market to reprice legal costs.

The same day the investor Steve Eisman questioned its AI spending, pointing at expenses up about 55 per cent year on year. Legal costs and capital expenditure rising together, against a valuation resting on advertising profit from 2027 onwards.

The same arithmetic reached other places. $NEBIUS(NBIS)$ closed down 7.60 per cent even though its own news that day was all good: the city council in Vineland, New Jersey approved the second phase of its data centre and the project cleared its permits. The AI neoclouds are the most direct casualties of a long-dated yield — asset-heavy, expanding on debt and long contracts, with profit landing two or three years out.

$NVIDIA(NVDA)$ closed down 2.34 per cent, a week from results after the close on 26 August. The bond investor Jeffrey Gundlach compared its US$500 billion AI financing arrangement to "bonds collateralised by bananas", and its guarantee on that Ohio data centre for OpenAI was cut on Monday, from US$$250 billion to U$$120 billion.

And the buildout is still running on borrowed money. Alphabet is preparing a record A$5 billion Australian bond to fund its AI spending, timed to a moment when the long end is pinned at its highest since 2007.

The week is not done reporting: Target pre-market on Wednesday, Kuaishou after the Hong Kong close on Wednesday, Alibaba pre-market in the US on Thursday, Yangtze Optical on Friday.

The minutes of the July policy meeting come at 2am Singapore time on Thursday — that 29 July meeting was the fifth in a row with no move, leaving the range at 3.50 to 3.75 per cent, though three members argued for a 25 basis point rise. If the minutes set out the case for a rise more fully, the long end has further to go.

The best of the three report cards fell 19.38 per cent. While the denominator is still moving, a report card can only do so much.

The above is personal analysis, not investment advice.

💬 【Talking Point】

$Apple(AAPL)$ cut its EU core technology commission to 5 per cent and settled a long-running antitrust case — the only Magnificent Seven name up on the day. Do you read that as buying certainty cheaply, or as giving margin away under pressure?

💰 【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!