The Orders Survived.
The Prices Didn't.

The week that didn't fit into a week

I fell behind on the usual Sunday report. By the time I caught up, the story had changed three times. Thursday was a rout, Monday was a relief rally, and Tuesday's inflation report took some of that relief away. A weekly index return would have told me that technology had a bad week. It wouldn't have told me why some of the businesses I follow were reporting record sales at the same time their shares were getting crushed.

On November 5, the Nasdaq Composite fell 5.5% and the Nasdaq-100 lost 6.4%. Election results fed fears that data-center development might face a much less cooperative political environment. The most concrete evidence was local rather than federal. Of 41 data-center-related ballot measures decided November 3, voters favored limiting development in 40: they approved 27 restrictive or opposition measures and rejected 13 measures that would have supported development. Some were binding bans or zoning restrictions; others were advisory votes. Forty results in favor of limits are not forty newly enacted bans, and they certainly aren't one national prohibition.

Still, that is a real change in the buildout's operating environment. A permit is an input. So are land, water and grid access. A company can have a signed equipment order and still wait longer than expected for the building that will house it. That's a cost, and investors started repricing it before anyone could tell precisely which projects might be delayed.

After a partial Friday recovery, the Nasdaq Composite ended the week down about 3.4%. Monday brought a strong rebound, with the S&P 500 up 2.35% and the Composite up 2.74%. Tuesday brought October's CPI: headline inflation rose 0.4% for the month and core inflation rose 0.3%, both above expectations. The ten-year Treasury yield finished around 4.98%, up from 4.88% Monday, and the Composite surrendered 0.74%. The Fed had held rates at 3.75%–4.00% on October 28, but the new inflation reading increased expectations of another increase in December. That remains a market expectation, not a decision.

What I see is three different risks arriving together. Customers still want computing power. Communities may make new capacity harder to build. Lenders and shareholders are asking what it costs to finance that capacity in the meantime. The businesses can be healthy while their stocks become cheaper, and that is precisely where I want to do the homework.

What the portfolio actually did

The Frontier File portfolio stood at $7,683.80 on October 30. By Friday, November 6, it was $7,377.68, a decline of $306.12, or 3.98%. The Monday rebound brought it to $7,593.11. Tuesday's closing mark was $7,556.04, down $37.07 for the day and $127.76, or 1.66%, from the October 30 starting point.

Baker Hughes, BWXT, Celestica, Seagate and Nvidia together cost the book about $249 since October 30. Rocket Lab contributed roughly $55, the largest positive contribution, while PSQ returned about $48. PSQ did what it is built to do as the Nasdaq-100 declined. That doesn't mean it hedged the things that hurt most. A daily inverse index fund is an imperfect defense against margin pressure at an equipment supplier, a delayed industrial project or a company-specific surprise.

PSQ and BWXT now represent almost 56% of the portfolio together. I am glad the hedge absorbed some damage, but I am not calling that a completed risk-management system. The review level previously set for PSQ remains a Nasdaq-100 close above approximately 30,035. At Tuesday's 27,740.79 close, the index would have to rise about 8.3% to reach it. I am not casually reducing protection to fund a new idea, and I am also not treating a one-third inverse position as a permanent home for capital.

There is another company-specific item worth separating from the general selloff. Nvidia's board announced an independent inquiry following reports of a foreign intelligence effort targeting its CEO. It expressly said it had no reason to believe company or national-security secrets had been compromised. The shares closed Tuesday at $168.89, down about 5.5% that day. There is an open governance and information-risk question here; it is not evidence that AI demand suddenly failed. I am not adding to Nvidia on this particular dip while the board's review is unresolved.

Tower: the capacity is getting paid for, partly

Tower Semiconductor gave me the cleanest operating reason to reopen a buy decision. Third-quarter revenue was $526 million, up 33%, and gross profit reached $179 million. Management guided to $600 million in fourth-quarter revenue, plus or minus 5%, and is targeting an annualized silicon-photonics revenue run rate of $1.1 billion in the fourth quarter. Annualized matters here: it describes the pace at which sales could be running, not another $1.1 billion of revenue booked in one quarter. Management also says customer contracts are being extended through 2029.

The financing side is more interesting than a growth percentage. Tower generated $185 million of operating cash and spent approximately $191 million net on property and equipment in the quarter. That looks almost self-funded, and in one sense it was. But the $185 million includes a roughly $23 million increase in customer advances. Without that increase, operating cash was closer to $162 million, leaving a roughly $29 million gap against capital spending. The customers are helping fund the buildout. That's encouraging evidence of commitment, although advances are not free money and don't guarantee future utilization or margins.

Tower closed October at $166.48, fell to $137.34 on November 6 and ended Tuesday at $147.60. The share price has come down while the silicon-photonics evidence has improved. That is the sort of disconnect I will investigate rather than automatically celebrate. Its factories still need to be equipped, depreciation will follow, and a customer extension is not the same thing as cash collected at attractive margins.

Tower is already in my Growth sleeve. I favor a staged addition capped at about 1.5% of the book, roughly $113 at this closing mark, with the first part contingent on the stock stabilizing and no deterioration in the capacity-funding picture. That would leave approximately $56 of the currently tracked cash. The second part should depend on continued evidence of customer-backed demand. I prefer that to treating one ugly week as permission for an oversized purchase.

Coherent: more revenue, a much larger cash bill

Coherent's latest quarter is why I haven't abandoned the optics bottleneck. Fiscal first-quarter revenue reached $2.37 billion, up about 50% from a year earlier, and non-GAAP gross margin was 40.6%. Management expects more than 60% pro forma revenue growth for fiscal 2027, helped by the migration from copper to optical connections in and around AI data centers. That's a meaningful operating forecast, though the pro forma basis matters when comparing growth rates.

Then I read the cash-flow statement. Operating cash flow was negative $92.7 million. Capital expenditures were $775.5 million, and the company raised about $773 million through credit-facility borrowings during the quarter. Strong revenue and negative operating cash can coexist during a capacity ramp, especially when inventory, receivables and equipment spending move ahead of deliveries. It is still a financing obligation, not a footnote I can wave away because photonics is the right technology.

Coherent fell to $169.76 Thursday, rebounded to $225.19 Monday and closed Tuesday at $214.66, close to its October 30 level. The stock has already demonstrated how quickly a bargain-looking entry can disappear. I like the operating story, and I like that it competes in the part of the physical AI chain where bandwidth and power are becoming constraints. I don't yet like the idea of calling it cheap before cash conversion improves.

A 1%–1.5% Growth re-entry remains a live research idea, but not an immediate order. I'd want evidence that working capital stops consuming cash at this pace, or that customer commitments more clearly cover the capacity investment. If another quarter brings a large cash deficit and additional borrowing without a comparable improvement in contracted economics, the apparent upside would be less attractive than the stock chart suggests.

Tower and Coherent are two ways of expressing much the same optical demand. Owning both doesn't automatically diversify that risk.

BWXT: good results, an oversized answer

BWXT reported $953 million in quarterly revenue, $179.7 million of adjusted EBITDA and $1.26 of adjusted earnings per share. Management narrowed its 2026 earnings and free-cash-flow guidance and offered preliminary 2027 expectations for continued growth. Its reactor technology was selected for the Army's Janus program at Fort Campbell and for a transportable nuclear-power project in Canada. Those are encouraging milestones, although selection is not the same thing as full lifetime project revenue already in hand.

Shares finished Tuesday at $126.64, compared with $130.70 on October 30. The company-specific evidence has not given me a reason to discard the core naval and nuclear business thesis. The 23.1% weight gives me a reason to revisit position sizing anyway. If BWXT crosses 25% of the book through price appreciation, I want a formal review. If guidance, backlog conversion or margins deteriorate, the case for reducing toward 20% becomes stronger. I am not adding simply because the price has been punished.

A sound company can be too large a bet. I don't want PSQ pretending to solve that problem.

The money is starting to travel in circles

Issue 025 asked how much of AI's next spending cycle would be financed by the suppliers themselves. Anthropic's amended SEC filing on November 10 gives us unusually specific evidence. It proposes an IPO of 825 million shares at $70–$78 each. Of those, approximately 801.4 million are company-issued shares and 23.6 million are being offered by existing shareholders, so the full headline offering value is not all money going to Anthropic. It remains a preliminary offering, not a completed raise.

More revealing is the proposed $26 billion concurrent private placement, conditional on the IPO and other closing requirements. The seven named investors are Google, Nvidia, AMD, Broadcom, Micron, Microsoft and TSMC. Three names in my book—Alphabet, Nvidia and Microsoft—are on that list. The filing says their purchases are covered by agreements, but the placements have not closed. Some of those companies also supply computing equipment, services or other infrastructure to Anthropic.

I can see the attraction. If one customer raises substantial capital, more hardware orders can become financeable. I can also see the danger of treating the same dollars as independent confirmation of demand every time they move from a supplier to a customer and back to the supplier's order book. The question is not whether such arrangements are inherently bad. It's whether the end customer eventually generates enough cash from the computing service to pay everyone without needing a new investor each time.

A Tuesday after-close receipt added another layer. CoreWeave reported quarterly revenue of $3.66 billion and said it expects approximately $9 billion in customer prepayments and deposits over the next three months. At the same time, it reported about $846 million of net interest expense and a $709 million net loss for the quarter. Customer financing can be an asset to an infrastructure builder; it doesn't make the debt expense disappear. I am keeping that disclosure separate from Tuesday's 4 p.m. market marks.

This is becoming the standing research question for the portfolio: who pays before the factory ships, who carries the equipment risk after delivery, and who ultimately converts the order into cash?

Three aggressive setups, not three purchases

First: Tower Semiconductor, Growth. Incremental allocation of up to about 1.5%, staged, rather than a full-size bet at the first bounce. The advantage is record reported revenue, a tangible silicon-photonics target and some capacity funded by customer advances. The disqualifier is a breakdown in demand visibility or cash generation as capital spending continues. Tower earns first consideration for the currently available cash, not a blank check.

Second: Coherent, aggressive Growth. A potential 1%–1.5% re-entry only after more convincing working-capital and financing evidence. The technology and sales momentum are strong, but cash from operations was negative while the capacity bill was enormous. If customers delay or the funding burden rises without better cash conversion, I stay out. This is a correlated extension of the Tower thesis, not a separate safety net.

Third: Micron, cyclical Growth. At $725.11 Tuesday, down about 18% from October 30, the reward could be meaningful if memory scarcity persists and profitability improves. Reports of halted construction work at its Taoyuan project are enough to keep me cautious; Micron says existing Taiwan operations continue normally, which is different from saying the expansion risk is resolved. I want credible de-escalation in the labor dispute and updated memory-price evidence before considering even a 1% starter. A large decline alone doesn't meet either condition.

There is $169.26 in tracked cash. All three ideas cannot be funded from it, and buying all three would compound exposure to the same AI capital cycle. Any reallocation beyond a small Tower starter would require an explicit capital decision, including a separate review of PSQ. I won't retroactively change the hedge rule to make an exciting purchase look affordable.

The case against my optimism

The bearish case has improved too. Local opposition can delay the sites that were supposed to receive new equipment. A ten-year yield near 5% raises the price of financing assets that may depreciate quickly. Suppliers can report record demand while absorbing inventory, borrowing to expand and guaranteeing residual values for customers. That is a plausible recipe for several quarters of great revenue and disappointing shareholder returns.

The bullish answer is equally grounded. Tower, Coherent and AMD reported actual revenue growth, BWXT reported operating earnings, and customers are putting money behind certain capacity plans. An election result isn't a blanket moratorium, and a difficult cash-flow quarter doesn't by itself make a growing company distressed. If the most durable suppliers can keep margins while collecting advances and reducing their own funding gap, today's lower prices may prove too pessimistic.

I'm not trying to call the market bottom. I'm trying to separate a genuine improvement in business evidence from a temporary change in what investors are willing to pay for it.

What I'm not doing

• I'm not treating 40 local ballot results as a nationwide construction ban. The distinction between a binding permit restriction and a nonbinding vote matters.
• I'm not buying Tower, Coherent and Micron together. They overlap in the same investment cycle, and the cash isn't there.
• I'm not calling negative operating cash flow irrelevant because orders are growing. The financing is part of the business model I'm underwriting.
• I'm not averaging down Nvidia while its board inquiry is unresolved, or adding to BWXT without revisiting its concentration.
• I'm not making PSQ a permanent short. The published review conditions still exist, but protecting against the Nasdaq isn't the same as protecting against company-specific financing risk.

The next receipts

Thursday, November 12 brings Cisco and Applied Materials results after the close. Together they'll provide another read on network orders, semiconductor-equipment spending and whether customers are moving forward with planned capacity. Friday, November 13 brings October producer-price inflation, the next check on whether Tuesday's CPI surprise is broadening. I'll follow Anthropic's amended offering terms through any eventual pricing rather than treating the prospectus as cash already received.

Nvidia reports November 17. That is both an operating-demand test and an opportunity for management to address the board's open inquiry. The Fed's next scheduled meeting is December 8–9. Between now and then, I'll be watching Tower's silicon-photonics run-rate follow-through, Coherent's cash conversion, BWXT's weight and Micron's construction dispute. Those are the receipts that change allocations, not the next loud trading day.

The next dollar

Across the last several issues, my question has moved from whether the AI buildout is real, to who pays for it, to whether the supplier ends up financing the buyer. This week added something less glamorous: the town that grants permission, the lender who sets the price, and the months that pass before equipment becomes a paying installation.

The orders are still arriving. Some of the businesses filling them look better than they did a month ago, and their stocks are cheaper. That combination deserves attention. It doesn't deserve unlimited capital.

Tower has the clearest improvement in evidence today. Coherent has the harder cash-flow homework. Micron still has a condition to clear. My next dollar belongs to the business that can turn an order into cash without needing the whole financial chain to stay generous forever.

The orders survived. Now I want to know who keeps the money.

See you next Sunday.

— Alex

Source desk

1. Ballotpedia, November 3 local data-center ballot outcomes (Nov. 6, 2026).
2. Associated Press, November 6 market close, November 9 and November 10.
3. Bureau of Labor Statistics, October CPI; U.S. Treasury, daily par yields; Federal Reserve, October 28 statement.
4. Tower Semiconductor, third-quarter results, outlook and cash-flow detail (Nov. 2).
5. Coherent, fiscal Q1 2027 results and cash-flow statement (Nov. 4).
6. BWXT, third-quarter results and outlook (Nov. 2).
7. Nvidia board statement (Nov. 8); independent directors' update (Nov. 10).
8. AMD, third-quarter results (Nov. 3).
9. Micron Taoyuan construction dispute, Tom's Hardware.
10. Anthropic, amended Form S-1 registration statement (filed Nov. 10).
11. CoreWeave, third-quarter 2026 earnings release (Nov. 10, after close).
12. ProShares, PSQ investment objective; Nasdaq-100 index history.
13. Cisco, November 12 earnings notice; Applied Materials, November 12 earnings notice.
14. BLS, November 2026 economic-release calendar; Nvidia, November 17 earnings announcement; Federal Reserve, December meeting schedule.

Research commentary. Portfolio figures are marked to the November 10 close; later corporate disclosures are identified separately. Concentrated positions and inverse funds carry significant risk.