The Demand Held.
The Prices Didn't.

The factory didn't get the memo.

There are easier ways to explain a bad week in technology stocks than the one we just had.

If AI spending had collapsed, the story would be straightforward. Customers stopped ordering. Suppliers got nervous. Stocks fell.

Except the customers haven't stopped ordering, and the suppliers closest to the physical infrastructure just delivered some remarkable results.

Taiwan Semiconductor Manufacturing Company reported third-quarter revenue growth of 50.9% in New Taiwan dollars. Net income increased 68.7%, with gross margin reaching 66.4%.

In U.S. dollars, the company generated $46.72 billion in quarterly revenue, up 41.2% from a year earlier. Its fourth-quarter revenue guidance landed between $52.3 billion and $54.3 billion.

Those aren't projections about what artificial intelligence might become. That's money already collected from manufacturing chips.

ASML followed with €11.7 billion in quarterly sales, nearly €5 billion in net income, and an unusually revealing statement about future demand.

Its planned production capacity for 2027 and 2028 won't fully satisfy what customers are asking for. Some customers are expressing interest in placing orders for 2029 and even 2030.

Interest is not an order. A capacity shortage is not automatically a higher share price. We've established those distinctions enough times that I don't want to forget them when the numbers are exciting.

But this is significant. The company making advanced chips reported extraordinary business. The company making the machines required to produce those chips says its customers want more.

And the Nasdaq Composite fell 4.5% for the week.

The S&P 500 lost 1.9%. The Russell 2000 declined 3%. The Dow managed to gain 0.7%.

That last number matters. This wasn't everyone suddenly deciding that owning a business was a terrible idea. It was a particularly ugly week for companies attached to the expensive end of the AI growth trade.

A business can become more valuable while its stock becomes less valuable. The difference is what investors were already willing to pay.

The invoice got another line item.

September's inflation reports weren't friendly to anyone hoping the Federal Reserve would deliver cheaper money.

Consumer prices increased 0.6% for the month and 3.7% over the year. Core inflation, excluding food and energy, increased 0.3% monthly and 2.6% annually.

Producer prices were more alarming: up 1.3% for the month and 6.3% year over year. The monthly increase was the largest since March 2022.

Energy drove a substantial part of it. Final-demand energy prices jumped 10.8%, and producer gasoline prices climbed 20.3%.

That doesn't mean every manufacturer suddenly experienced a 20% cost increase. These are different measures of prices at different points in the economy. But even excluding food, energy and trade services, producer prices increased 0.5% in September and 5.0% from a year ago. There was inflation pressure underneath the headline.

Here's what I find more interesting.

The 10-year Treasury yield didn't respond by marching higher all week. It declined from approximately 5.24% on October 9 to 5.09% on Friday.

Investors were willing to accept lower yields on government debt even as they became less comfortable financing some AI companies. One market was repricing the safety of government bonds; another was asking how profitable a capital-intensive buildout would really become.

Which brings me to the number that bothered me more than the Nasdaq's decline.

Fifteen percent.

That's approximately where yields on some CoreWeave bonds traded this week.

CoreWeave rents AI computing capacity to customers that need processing power without building all the infrastructure themselves. Its business requires extraordinary capital: servers, facilities, electricity and financing often need to be secured before much of the associated revenue arrives.

In June, CoreWeave issued bonds at a yield of roughly 9.675%. By this week, those bonds were trading at yields approaching 15%.

When a bond's market price falls, its yield rises. The new yield reflects what a buyer could earn at today's lower price and the compensation investors demand for taking the credit risk. It does not mean every new dollar CoreWeave borrows will necessarily carry a 15% rate.

But the repricing is hard to ignore.

A company can have substantial customer commitments and still face financial pressure if the cost of fulfilling them keeps increasing. CoreWeave isn't every AI company; Microsoft and Alphabet finance their operations under very different conditions. But CoreWeave is a useful place to watch the strain because it sits between the customers and the expensive physical assets required to serve them.

Last week I asked who would finance the buildout. This week, the bond market offered a partial answer: some borrowers are going to pay considerably more than they expected.

Four orders and a smaller hole.

The revised October 12 ledger is a retrospective research reconstruction adopted after the week ended. It applies historical Monday opening prices to an October 2 target-weight book. These aren't independently confirmed live brokerage executions, and I won't mistake the outcome for a trading call made with foresight.

With that distinction established, this is how the defensive version of the book reads.

At Monday's open, the ledger closed Ciena for $671.42, Coherent for $461.02 and Micron for $1,104.97. The combined proceeds were approximately $2,237.40. The fourth order placed those proceeds into ProShares Short QQQ (PSQ) at $24.53 per share.

PSQ is designed to deliver the inverse of the Nasdaq-100's daily performance, before fees and expenses. It's a hedge against a particular kind of market decline, not a substitute for cash or guaranteed insurance.

Why those three positions?

As of October 9, they represented approximately 27% of the reconstructed portfolio: three different businesses exposed to one big question. Can the companies building the AI economy finance expansion at the pace investors expect?

That concern was already in Issue 022. I wrote about expensive capital, enormous spending commitments and the risk of owning several businesses exposed to the same investment cycle. Then I said the October rebalance should hold.

That's a contradiction worth examining. The defensive scenario shows what acting on the earlier warning might have looked like.

But Monday's decision couldn't have cited Wednesday's inflation numbers, Thursday's TSMC results, CoreWeave's credit repricing or Rocket Lab's Neutron update. Those facts hadn't arrived. And the stock selection benefited from hindsight: Rocket Lab fell about as much as Coherent and stayed in the book.

So I'm not going to turn a useful exercise into a story about predicting the week perfectly.

The reconstructed closing book

An unchanged version of the October portfolio would have finished near $7,430.17, down about 12.35%. The reconstructed defensive version finished roughly $540 higher.

That $540 needs a breakdown. Approximately $414 came from selling the three stocks before their subsequent declines. Another $126 was the unrealized PSQ gain between Monday's opening purchase and Friday's close.

If I'd sold the three positions and simply held the proceeds in cash, the portfolio would have finished near $7,844. Selling accounted for most of the relative improvement. The hedge added the rest—and added a risk that cash wouldn't have.

Even in this reconstruction, the timing wasn't perfect. The three positions had already lost approximately $49 between the preceding Friday's close and Monday's open. And the other 19 stocks still lost roughly $584 over the week.

A book that fell almost 6% isn't a victory lap. It is a useful study in how concentrated exposure can be managed—and how hindsight can make a trading rule look better than it was.

The scenario remains above the $5,270 recorded as contributed capital, but that comparison inherits the assumptions of the reconstructed book. It isn't a verified brokerage return.

The more urgent question is what the portfolio looks like now.

Because the protective trade created its own problem.

The hedge became the largest position.

At Friday's close, PSQ represented 29.65% of the reconstructed book. BWXT represented 23.73%. Together, they were approximately 53.4% of the portfolio.

That's a strange place to arrive after spending several weeks talking about concentration risk.

I reduced exposure to three technology companies with related economic sensitivities and replaced them with one position making a directional bet against the Nasdaq-100.

That can help during a technology selloff. It can also become expensive quickly if the market recovers.

The book is still about 68% long individual stocks, so it hasn't become an outright bearish portfolio. But PSQ isn't an exact hedge for those stocks. A Nasdaq-100 decline could lift PSQ while BWXT falls for an unrelated nuclear-industry reason. A technology rebound could punish PSQ while some of my smaller industrial positions barely participate.

Then there's the daily reset. PSQ seeks negative one times the index's daily, not cumulative, return. Over multiple days, compounding can make its performance drift from a simple inverse of the Nasdaq-100's total move.

This is a tactical tool, not a position I want to forget about for six months.

A 5% Nasdaq-100 rebound would imply roughly a $118 loss on the current PSQ position, around 1.5% of the whole book, before considering compounding or any gains in our remaining stocks. It's a simplified sensitivity, not a price forecast.

The exit plan on my desk

The Nasdaq-100 closed around 29,188 Friday, down from approximately 30,883 on October 9. Recovering half that decline would put the index near 30,035.

If the Nasdaq-100 closes above approximately 30,035, I will review cutting PSQ toward 15% of the book. That's a review trigger, not an automatic market order.

I'd also review a reduction if CoreWeave's bond yields retreat meaningfully from 15%, or if upcoming earnings from major AI customers show they can keep funding infrastructure spending without impairing their financial position.

The opposite case matters too. If credit problems spread beyond one borrower while equity prices keep breaking down, retaining some protection could make sense. But that isn't permission to increase the hedge every time another technology stock falls.

A hedge is supposed to manage risk. If it becomes the main source of risk, the design needs changing.

And if PSQ is reduced, the proceeds don't automatically have to go back into Ciena, Coherent or Micron. Cash is an available decision.

I think we sometimes forget that.

Rocket Lab earned a harder question.

Rocket Lab declined approximately 25.5% for the week, closing near $50.84. It wasn't solely a sympathy move with technology stocks.

On October 14, the company announced that the first flight of its Neutron rocket is now expected in 2027. The SEC-filed release establishes the change in timing; subsequent reporting on the development update describes manufacturing difficulties involving carbon-composite propellant tank components.

Neutron is central to Rocket Lab's ambitions beyond its existing Electron launch business. Electron still has customers, and the space-systems operation still has a business. Its previously announced 20-launch agreement hasn't vanished because another rocket's timetable moved.

But the timing of a new launch vehicle affects when opportunities open and how long development capital stays committed. This program has already experienced delays.

In Issue 022 I wrote that Rocket Lab's earlier decline didn't undo the contract behind its October allocation. True, but incomplete. The contract is intact. The development schedule has moved. Both belong in the valuation.

At approximately 4.4% of the revised research book, Rocket Lab isn't the largest position. It is one of the positions most urgently in need of fresh underwriting.

I want the updated technical milestones, manufacturing fixes, launch schedule and financial implications before deciding whether the share price compensates for the execution risk.

A 25% decline isn't a thesis. Neither is a 25% discount.

One new research candidate.

ASML moves onto the active research radar.

Not because its shares fell this week. Because its operating results suggest a genuine industry constraint, and I want to know whether the stock's price properly reflects the economics of controlling that constraint.

Advanced chip manufacturing requires extraordinary equipment. ASML's lithography systems occupy one of the hardest-to-replicate parts of that production chain.

An attractive competitive position isn't a complete investment case.

I need to know how much future revenue is backed by firm orders versus management expectations, how much additional capacity ASML can deliver without surrendering margins, and how much of the future growth is already reflected in the valuation.

I'm especially interested in whether stated interest in equipment for 2029 and 2030 becomes signed commitments.

Until those questions have answers, ASML is research, not a purchase.

There's no prize for finding a wonderful business and paying the wrong price.

What I'm not doing.

• No foresight story. The reconstructed book still lost 5.98%. Its result teaches risk management, not that I called the selloff in real time.
• No automatic return to Ciena, Coherent or Micron. Their businesses merit fresh research, but price, financing and company evidence have to justify re-entry.
• No permanent PSQ allocation. A nearly 30% inverse position needs clear reduction conditions. It's not a new Core holding.
• No Rocket Lab shrug. Neutron's move into 2027 requires a revised investment case, not another reassuring paragraph about the future of space.
• No shopping spree because prices fell. ASML gets researched. Quantum stays small. A low share price, a sharp decline and an undervalued business are different things.

The next receipts.

There are enough upcoming events to challenge the portfolio without manufacturing more reasons to trade.

No need to invent catalysts. There are enough real ones on the calendar.

The next dollar has a different assignment.

Two weeks ago, the question was which companies deserved more capital. Last week, it was who would pay for the infrastructure being built.

This week, the question became more uncomfortable.

What happens when the people making the equipment are doing exceptionally well, but the companies financing its deployment begin to look less secure?

TSMC has its revenue. ASML has customers asking for more capacity. CoreWeave has debt trading at yields that suggest investors want significantly more compensation for the risk.

And my reconstructed portfolio is left with a hedge that worked for a few days but now requires a decision of its own.

I still believe the physical AI buildout has years of work ahead. But what I believe about the industry and what I'm willing to pay for its stocks are separate questions. And what I'm willing to own through a major drawdown is a third.

The October rebalance put meaningful capital behind the research. The defensive exercise showed how much of that capital was exposed to the same market concern, even across different companies.

The next improvement isn't making four trades every Monday. It's having a system that identifies concentrated risk before the market decides for me—and tells me when the defense has done enough.

The demand held. The prices didn't.

Now I have to decide how long protection is worth paying for.

See you next Sunday.

— Alex

Source desk

1. TSMC — Third-quarter 2026 earnings
2. ASML — Third-quarter 2026 results
3. Associated Press — October 16 market close and weekly results
4. BLS — September consumer price index
5. BLS — September producer price index
6. U.S. Treasury — Daily par yield curve
7. Reuters — CoreWeave bond rout
8. Rocket Lab — October 14 Neutron SEC filing
9. Investor's Business Daily — Neutron manufacturing update
10. ProShares — PSQ fund objective and daily inverse exposure
11. ChartExchange — Nasdaq-100 historical closes
12. SpaceX — October 20 earnings webcast notice
13. Celestica — October 27 investor event
14. Baker Hughes — October 27–28 Q3 earnings
15. BWXT — November 2 Q3 announcement

Accounting methodology: The October 12 orders and all October 16 position values were reconstructed retrospectively from October 2 target allocations and historical prices; they are not contemporaneously verified brokerage executions. The $7,970.16 portfolio mark is a scenario calculation, not verified NAV. Several underlying price inputs remain independently unverified. The figures assume fractional shares, no fees, slippage, taxes, dividends or unrecorded cash flows. The PSQ gain is unrealized. Research and commentary only; concentrated equity holdings and inverse ETFs can lose substantial value.