The receipts came in. The market still sent an invoice.
Issue 16 ended with a simple split. Nvidia proved the build. Warsh proved the rent.
This week both sides of that sentence got louder.
Broadcom reported $29.6 billion of quarterly revenue. AI semiconductor revenue hit $16.7 billion, up 221%. The company guided Q4 AI semis to $21.7 billion. Custom silicon and networking — the layer underneath Nvidia — is not a rumor. Somebody is paying for it.
Rocket Lab flew Owl Around The World on September 2 after wind delays. Electron mission 94. Fifteenth launch of 2026. Another Synspective StriX in a 575-kilometer orbit. One hundred percent success across that customer relationship, with sixteen more missions booked before 2030. The stock finished the week basically flat.
Planet reported a record quarter. Then the stock fell about 9% on the week because next quarter’s guide missed what the market had already priced in.
Friday, the August jobs report printed 162,000 payrolls against a forecast near 56,000. Unemployment held at 4.1%. The 10-year moved back toward roughly 4.8%. September hike odds moved back toward 60%. Brent gained about 7.6% on the week.
The S&P finished the week up 0.1%. The Nasdaq up 0.4%.
That is not the story.
Chips and memory got paid. The companies writing the capex checks got marked for duration.
The story is the sort:
Paid scarcity got paid. Timing got marked down. Time still costs money.
Where my head is at.
The map before the positions.
Scarcity got a bid. Timing got a haircut.
| Marker | Level | Read |
|---|---|---|
| S&P / Nasdaq | +0.1% / +0.4% | indexes hid the sort |
| MU | $1,016.59 / +9.0% | memory scarcity bid |
| NVDA | $230.34 / +5.9% | build still on |
| AVGO AI semis | $16.7B / +221% | custom silicon is real |
| PL | ~$18.12 / −9.3% | beat, then Q3 guide |
| RKLB | ~$64.26 / flat | flew, not celebrated |
| ASTS | $61.99 / +6.8% | sentiment, not a receipt |
| Jobs | +162k / 4.1% | hike odds ~60% |
| Brent | +7.6% week | inflation still in the room |
This was not a generic technology rally. Memory and compute got paid. Cloud mega-caps did not. Space split between execution and guidance. Nuclear manufacturing worked quietly.
That is a selective market. I prefer it.
Selective markets tell you who is actually getting paid.
Do not turn confirmation into a shopping list.
The dangerous candle.
Micron rose almost 9% on the week and more than 6% on Friday. Seagate jumped with it. Broadcom and Nvidia gave the fundamental cover.
That is the trade I most want to prevent: buying the belief after the belief already printed in the quote.
Hold. September 30 is the next memory receipt. Friday was not.
The build keeps clearing. The stock already knows.
$230 after last week’s giveback is not a reason to add. It is a reason to stay seated.
Core stays Core.
More ways to monetize the same scarcity. Same size limit.
$152.85 to $157.79. Another NNSA lithium-processing design award. Investor Day September 29.
This is what I want from Core: not a louder story, a business finding new invoices for capacity that already exists.
If the reconstructed weight is still ~18%, do nothing. If a fresh reconciliation eventually shows it near 20%, the conversation becomes a modest trim — not another add.
Thesis strong. Concentration real. Both can be true.
The other half of the surplus.
The build they are funding is working. Their stocks still got charged for duration.
Hold. A lagging mega-cap is not a broken cloud thesis.
Best operating receipt of the week. Size unchanged.
Wind delayed the mission. Then they flew it. That is process.
Sixteen more Synspective launches on the books is backlog, not a press release. Flat tape after a clean mission is healthier than a 15% celebration because a rocket did its job.
Operational thesis up. Position-size decision unchanged. If a Growth dollar is ever earned, RKLB belongs high on the underwriting list. I will not invent the dollar before CPI.
Proof arrived. Timing got punished.
Old stance: do not add until proof.
New stance: proof arrived; wait for the market to digest the Q3 guide.
Record quarter. Real defense mix. Real backlog. The next-quarter softness is partly timing: Planet said the Swedish satellite handover landed in Q2 rather than Q3.
Cheaper is not the same as cheap. If the stock stabilizes instead of retracing the whole print, it earns deeper work.
Not Monday. Labor Day is closed. CPI is the weather.
A bounce is not a receipt.
+6.8% after last week’s stairs. Hold Growth size. Do not promote midweek sentiment to an operating event.
Starter sizing was the whole point.
Monday down. Tuesday down. Wednesday +9.3%. Then it gave most of it back and finished the week about +1%.
That is how you learn a multiple without letting the multiple run the book.
Can Reddit convert human provenance into durable economics? Still the question. Still not a second tranche.
No new proof. Trim stays a trim.
No structural change this week. Greater capability still carries greater complexity. The trim does not need to be reversed because the tape got quiet.
Green is not a promotion.
Intel and SPCX were green. That is not a promotion. Oklo green is not BWXT. Serve near $5 is not cheap. QBTS stays tiny.
No averaging. No new sub-$5 name because it “could run.”
Three threads into CPI week.
The portfolio-wide exam. Jobs were too strong to retire the hike. Oil is still an inflation problem. A hot print reprices every long-duration name you already own. A cool print does not make ASTS or OKLO Core.
Not a holding. A read-through. Nvidia confirmed GPUs. Broadcom confirmed custom silicon and networking. Oracle can tell us whether AI cloud demand is becoming economically attractive growth — or just an enormous capex bill. That distinction hits MSFT, AMZN, GOOGL, and the whole financing argument.
BWXT already gives me nuclear manufacturing. OKLO gives me nuclear optionality. The adjacent bottleneck I do not own is the grid — turbines, transformers, switchgear, interconnection. GE Vernova has a $176 billion backlog and more than $5 billion of data-center orders year-to-date. It is not cheap. With cash effectively zero, GEV is a watchlist entry, not a Tuesday trade.
No chase. No rescue. No fifth idea.
I am not buying Friday’s memory candle, adding BWXT because it worked, or averaging Planet because the business got better. ASTS does not get promoted because it bounced. Broadcom does not force a new position.
The book already owns the layers that matter. With cash effectively zero, any new idea is a funded trade: something else has to leave first.
If I want a 2–3% tactical reserve, it comes from the weakest seasoning — SERV / QBTS-class — not from Core, not from RKLB, and not from PL.
The next dollar — if I free one — is reserved for a CPI dislocation in a name that already produced a receipt: Planet or Rocket Lab. Or for a first look at grid equipment. It is not reserved for Friday’s memory candle.
The calendar is the strategy.
| Date | Catalyst | Why it matters |
|---|---|---|
| Mon · Sep 7 | Labor Day · market closed | No invented trade. |
| Wed · Sep 9 | Employer Costs · 10:00 ET + larger Treasury long-end buybacks begin | Wages plus a bond-market technical. If Growth catches a bid on a calmer long bond, do not confuse duration relief with company proof. |
| Thu · Sep 10 · 8:30 ET | August PPI | First direct inflation check of the week. |
| Thu · Sep 10 · after close | Oracle earnings | AI cloud growth versus capex economics. |
| Fri · Sep 11 · 8:30 ET | August CPI | The week’s real portfolio-wide decision point. |
| Sep 15–16 | FOMC | Rate decision and updated projections. |
| Sep 29 | BWXT Investor Day | Core thesis, capital allocation, long-range economics. |
| Sep 30 | Micron earnings | Direct memory-cycle receipt. |
A year ago a beat was a multiple.
This market is starting to ask what kind of beat.
Nvidia and Broadcom: demand now, cash now. Paid.
Rocket Lab: hardware now, economics still arriving. Respected, not rerated.
Planet: better business, worse near-term guide. Punished.
Micron: scarcity the tape can touch this quarter. Chased — which is why you don’t.
That is also why “the future is real” stopped being an investment process.
Reality is the easy part. Timing, conversion, and the hurdle rate are the job.
This week wanted to make us feel smart.
Memory worked. Nvidia worked. BWXT worked. The launch worked. Planet’s quarter worked on paper.
A little voice says: see?
Then Friday’s jobs number walked in and put the hurdle rate back on the table.
The dangerous trade is not missing Micron’s next three points. The dangerous trade is treating a confirmed build as permission to stop asking what things cost.
You already used capital where it belonged. BWXT is full. RDDT is a starter. Archer is trimmed. Cash is tight.
That structure is the strategy. Tuesday does not need a hero.
Proof still is not a multiple.
Issue 15: the market raised the rent.
Issue 16: the future paid, then the landlord raised it again.
Issue 17: the receipts arrived — and proof still is not a multiple.
Own the bottlenecks that already get paid.
Let Growth convert cadence and backlog into economics.
Keep Planet on the research board, not the buy board.
Leave Rocket Lab upgraded and the same size.
Leave Reddit small.
Leave the moonshots alone.
And remember: with no dry powder, restraint is a position.
— Alex