Hello everyone. When last night's US session closed, $S&P 500(.SPX)$ and $Dow Jones(.DJI)$ were both at records, $Dow Jones(.DJI)$ above 54,000 for the first time. Not long after the bell, $Advanced Micro Devices(AMD)$ was down 8.68 per cent.
It wasn't alone. $SpaceX(SPCX)$ was up 9.43 per cent at the close and dived 8.04 per cent after hours, at one point down almost 9 per cent. The two traced almost the same roller coaster: lifted in the session, knocked down after the bell.
And neither paper was a bad one. Parts of both were strong.
The same night, the other end of the market was having a party. $Palantir Technologies Inc.(PLTR)$ closed up 29.45 per cent, leaving those short of it about US$3 billion down on paper in a single day. Memory and optical names were bid up too — and not one of those reported overnight.
Two markets in one night, and both of them real. The real question sits in the gap between them.
The papers themselves were strong. AMD's second-quarter revenue was a record US$11.5 billion, up 50 per cent year on year; data centre revenue came in at US$6.7 billion, up 107 per cent, taking its share of the total from 42 per cent a year ago to 58 per cent; gross margin was 56 per cent. On the call, Lisa Su said demand for the new platform had "exceeded our initial forecast", and admitted server supply right now is "tight". SpaceX posted revenue of about US$7.8 billion, up 92 per cent and more than a tenth above what the market wanted; adjusted EBITDA of US$3.5 billion, up around 192 per cent; an operating loss of US$1.26 billion in the AI business, far less than the US$2.39 billion expected; and capital expenditure of US$18.4 billion, near enough the US$18.5 billion pencilled in.
What knocked them down wasn't how much they earned; it was how high the bar had been set. AMD guided the third quarter to about US$13 billion, a midpoint up 41 per cent year on year — not weak, just not enough to feed the most aggressive expectations. SpaceX's drop got put down to a rich valuation, the first tranche of lock-up expiries on Thursday, and the pace of the cash burn, rather than to the quarter it had just delivered.
The results were good enough. The share price had already spent them. AMD's 7.00 per cent before the close came with the whole chip complex — the Philadelphia Semiconductor Index rose 6.55 per cent that day, Intel 10.84 per cent. By the time the print landed, "solid" was already in the price, and worth nothing.
$Arista Networks(ANET)$ handed in a paper the same night, and not a "not bad" one: it beat and raised guidance sharply, with management saying flatly that "demand is not the problem, supply is the main bottleneck". Once the price has spent the good news, only something like that buys it back. "Not bad" won't.
And most of the night's hardest moves belonged to names that haven't sat the exam yet. The whole memory complex got bid up: $Micron Technology(MU)$ +7.62%, $SanDisk Corp.(SNDK)$ +10.84%, $SK hynix(SKHY)$ +8.17%, and $Direxion Daily Semiconductors Bull 3x Shares(SOXL)$ up 19.87%. $Philadelphia Semiconductor Index(SOX)$ Index rose 6.55%, the optical networking sub-sector 33.7%, Applied Optoelectronics 19.44% and Marvell 12.81%, whose call options suddenly traded in unusual size. None of the memory names moved on results — Micron put out a new AI memory platform, Bank of America relabelled the recent pullback a buying point, and SK Hynix picked up initiations from two houses, both at Outperform with a US$200 target.
In the last piece I wrote that the market only honours money that has already arrived, that a receipt beats an IOU. Last night put a discount on that. Palantir's receipt still got snapped up, but the ones AMD and SpaceX handed in lost marks, while the holders of nothing but IOUs got chased.
It wasn't sorting on who had the better results. It was sorting on whose results weren't in the price yet.
Only one name won at both ends. $Palantir Technologies Inc.(PLTR)$ closed up 29.45 per cent on revenue of US$1.935 billion, up 93 per cent, with US commercial revenue up 149 per cent to US$764 million. Nobody had dared put a price on that number in advance, so it all got paid at once — in the last piece I put the after-hours move at 14.25 per cent, and the regular session paid double. The bill went to the short side: a roughly 30 per cent move left shorts about US$3 billion down on paper, Michael Burry has closed half of a short, and Duan Yongping, having built a position early in the year, is still writing puts on top of it. Snap closed up 14.88 per cent the same day.
On the call, SpaceX said its chips are going "all in" on Nvidia, with a target of 20 gigawatts of power capacity by the end of next year. On SpaceX's books that sentence means years of money going out. On Nvidia's it is an order. Nvidia's 2.56 per cent was a regular-session close, and the purchase commitment only came out after the bell — the entry is on the books, the price hasn't moved on it yet. Cook's line on the shortage was the same machine: one company's cost, another's revenue. This is the second time.
The macro really was the main driver overnight. Bessent said a deal to reopen the Strait of Hormuz could land within a day or two, WTI fell about 5.8 per cent to US$75.72 a barrel, the 10-year Treasury yield dropped 6 basis points to 4.62 per cent, risk appetite and rates loosened together, and that is where the green at the close came from.
But a macro lift lifts everything, and it doesn't explain $Amazon.com(AMZN)$ falling 2.32 per cent and $Meta Platforms, Inc.(META)$ 0.39 per cent on the same day, nor $Advanced Micro Devices(AMD)$ up 7.00 per cent in the session and down 8.68 per cent after it. The macro can raise the water. It can't produce a split like that.
The new high sits on a thin base, though. Volume ran 7 per cent below the five-day average, Goldman's sentiment reading is 4 out of 10, and the VIX didn't fall — it rose 4.04 per cent to 16.50, though 16.50 is still a low reading. Last month's deleveraging washed positioning out, so the book is light and it moves on very little; Goldman itself puts the rally down to clean positioning, a valuation discount and returning earnings visibility, not to stronger conviction. One trader said flat out that a five-day rip like this doesn't hold. A light book can carry an index to a record. It can also hand the whole thing back on one set of results.
$SanDisk Corp.(SNDK)$ and $Western Digital(WDC)$ hand in their papers after Wednesday's close — the memory complex has run hard without sitting a single exam. On Thursday, as many as roughly 911.5 million SpaceX insider and employee shares come free, a fifth of what is eligible.
What last night bought was everything that hadn't sat the exam. Over the next two days, the papers get collected.
The above is personal analysis, not investment advice.
Comments