To The Moon

Hello. The largest first tranche of any IPO lock-up in US market history came free last night: 911.5 million $SpaceX(SPCX)$ insider shares became eligible to sell, taking the freely tradable count from about 639 million to roughly 1.55 billion in one step.

The shares rose 6.14 per cent.

The day the selling actually happened wasn't last night. It was the day before. On Wednesday the stock fell 13.61 per cent on about 200 million shares, the heaviest turnover since 18 June; last night volume hit 251 million and set another high, yet the price went up. Heavier volume than Wednesday, and the price went the other way — the people who wanted to sell had mostly finished on Wednesday.

Someone went back through three comparable cases: in each, the first unlock day marked a local bottom. The day the shoe hits the floor tends not to be the painful one. The day before is.

Wednesday's two sides are easy to see. On the short side, positions as of that close ran to about 36 per cent of the pre-unlock float, sitting on more than US$9 billion of paper profit. On the retail side, net buying in the first hour came to about US$22.7 million — the third-highest of the 37 opening hours this stock has had since listing, and more than three times the average. One side cashing out, the other buying it up.

Last night the banks took over: JPMorgan lifted its target from US$225 to US$240, Morgan Stanley reiterated US$300, and at least six houses stayed at buy. But that settles only one thing: the unlock won't break it. What it's worth, nobody can say — the US$135 issue price broke on Wednesday and has not been recovered since.

Don't read that bounce as the space sector turning, though: $Leverage Shares 2X Short SPCX Daily ETF(SSPC)$ fell 12.16 per cent the same day and didn't follow. What cleared was SpaceX's own float, not the valuation of the theme.

And this is only step one of nine. The largest tranche sits in June 2027, Musk's roughly 6.4 billion class A shares. That's one tranche down, eight still to come.

The same day, good results got sold a second time. $SanDisk Corp.(SNDK)$ fell another 6.81 per cent, with the US$14 billion buyback counting for nothing; $Western Digital(WDC)$ fell another 13.03 per cent, $SK hynix(SKHY)$ 4.97 per cent, and $Tradr 2X Long SNDK Daily ETF(SNXX)$ 13.53 per cent.

The difference is whether a thing has a date on it. The lock-up is written down: which day, how many shares, with the next eight stages already scheduled. The market finished pricing it a day early and cleared it in one go. Guidance has no date. If you don't trust next quarter today, you can go on not trusting it the quarter after, which is why one good set of results gets sold twice, then a third time.

And last night handed that doubt two fresh reasons to renew, both on the demand side. Nvidia is reported to be planning to cut the HBM configuration on Rubin Ultra to cope with the shortage of high-end HBM. And Musk's Terafab chip mega-plant has been launched with an initial US$16.8 billion and its own power generation, while Tesla is already recruiting memory-processing engineers. Micron fell 1.31 per cent on the day, having traded higher earlier on.

The last piece was about a price that wouldn't honour long-term contracts. This one is about customers working out how to go around them.

The opposite view exists too, and from the same firm. Goldman's Korea trading desk thinks the market has turned "too pessimistic" on memory fundamentals — the reverse of the Goldman note a day earlier that said expectations had run too far ahead of reality. Micron's management is talking as well, putting memory at more than half of total system value. Neither view need be wrong, but both are about the endgame, and what the market is pricing now is next quarter.

The bears haven't left either. They have changed target. Michael Burry was disclosed to have bought Oracle puts again and to have added a short position in $NEBIUS(NBIS)$; $NEBIUS(NBIS)$ fell 13.29 per cent on the day, among the hardest-hit AI names. Data-centre results from CoreWeave and Nebius come next.

At the other end, nothing has changed: $Microsoft(MSFT)$ rose 2.54 per cent to a record high in a down market, and $Advanced Micro Devices(AMD)$ rose 1.50 per cent on the acquisition of Toronto AI-chip start-up Taalas, pushing further into inference — a day after guidance knocked it down 7 per cent and took expectations with it. $NVIDIA(NVDA)$ fell 0.10 per cent and barely moved.

Bank of America reckons dispersion across US single stocks has reached its highest in nearly 35 years, and that with the exception of Nvidia, which has yet to report, all six of the other mega-caps moved more after results than options had implied beforehand. The market isn't rising or falling as a whole. It is pricing names one at a time.

The indices drifted lower through all this: $Dow Jones(.DJI)$ fell 464 points, or 0.85 per cent, $S&P 500(.SPX)$ 0.18 per cent and $Invesco QQQ(QQQ)$ 0.37 per cent, while Treasury yields and crude both jumped.

July payrolls land tonight at 20:30 Singapore time. The market expects about 80,000 jobs, up from 57,000 in June, with unemployment holding at 4.2 per cent — but ADP private payrolls added only 44,000, job openings are falling and the employment component of the ISM services survey has dropped to 47.4. On the other side, Warsh's line is that September is on the table if inflation data over the coming weeks come in strong. So a number that is too good is also bad news.

Payrolls get their answer at half past eight tonight. When memory will stop worrying people, nobody can say.

The above is personal analysis, not investment advice.