The Index Hid
the Damage.

Three answers to the same question

For two issues I've been asking who pays for the AI buildout. I had been looking at it mostly from the suppliers' side: the chipmakers, the optical equipment, the power systems, the companies selling the physical pieces of a new computing economy. Then the biggest buyers started reporting, and the question got more specific. How are they paying, and what happens when the next round costs more than the last?

SpaceX reported $14.58 billion in third-quarter revenue, including $7.56 billion in its Super Intelligence segment, which includes AI services. That segment reported $431 million in operating profit, its first positive quarterly operating result. The growth is striking, but so is the spending. Capital expenditure was $19.59 billion for the quarter, and the company guided to another $35 billion to $40 billion in the fourth quarter. It carries $36.1 billion of debt. Reports say it is discussing a raise of roughly $40 billion, though discussions are not financing secured. Bloomberg also reported talks over access to about 80,000 U.S. cell sites. That is a potential network strategy, not a completed agreement.

There's a real operating business underneath those ambitions. There's also a very large requirement for outside capital if the reported spending plans proceed. The lender and the next investor matter nearly as much as the next customer.

Meta provided a different answer. It generated $64.6 billion in quarterly revenue, up 26%, and $32.79 billion in operating cash flow. Capital spending, including finance-lease payments, came to $34.91 billion. By that measure, free cash flow was negative $2.12 billion, against a positive $10.63 billion a year earlier. Meta also approved the creation of non-voting Class C shares, effective January 19, 2027, with a stock dividend to follow. That gives it a potential way to issue shares without the same effect on voting control, although the announcement did not say it was preparing an AI equity raise. It's an option, and I want to understand why the option matters now.

Tesla went in the other direction. It lowered its 2026 capital-spending outlook toward $20 billion, while its SEC filing still described expected spending as more than $20 billion, and it reported positive quarterly free cash flow of $1.53 billion. Rather than fund every previously planned project on the original timetable, it pulled back its budget.

Those responses aren't interchangeable. SpaceX's reported financing discussions do not establish that it has raised money. A new share class doesn't establish that Meta needs to sell shares. And a lower capital budget may reflect discipline, weaker opportunity, or some of each. But the distinction is useful: one company is looking outward for capital, another has enough internal resources to fund an extraordinarily expensive choice, and a third has chosen to spend less.

Nvidia's Jensen Huang spoke this week about a growing scarcity of capital in the industry. I don't need to treat that as a forecast to recognize the issue. Customers can want the equipment and still be constrained by what it costs to deploy.

A quiet index, a painful book

The market's weekly summary looked ordinary enough. The Nasdaq Composite fell 0.71%, the Nasdaq-100 lost 0.79%, and the S&P 500 declined 0.47%. The Dow gained 0.41% and the Russell 2000 rose 2%.

The Frontier File portfolio did considerably worse. It closed October 23 at $7,692.86, down $277.30, or 3.48%, from the prior Friday.

The damage was concentrated. BWXT fell 7.17%, costing the book about $136. Celestica fell 14.42%, or another $57. Tower Semiconductor declined 14.33% and GE Vernova 11.53%, together contributing roughly $67 in losses. Those four names accounted for about $260 of the $277 net decline. The nineteen stocks in the book lost approximately 5.47% together. PSQ contributed only about $20 in gains.

Rocket Lab, incidentally, rose 3.46% for the week. It was a useful reminder that a difficult week for our infrastructure holdings wasn't a universal collapse in everything speculative.

Against the $5,270 in recorded contributions, the portfolio's running tally remains ahead by about 46%. But that backward-looking result does not solve the next decision: PSQ and BWXT together now represent almost 54% of the book.

I built exposure across different businesses. The week showed how little that means when several of those businesses are all being repriced for expensive growth, and when my protection tracks an index that barely moved.

BWXT: a conversation still needs a customer

BWXT was the biggest source of damage. On October 22 it issued a statement addressing reports of possible nuclear investment in Poland. Management confirmed discussions involving microreactors and small modular reactors, while making clear that no investment decision had been made and any commitment would depend on firm orders and the extent of the opportunity. Poland's defense ministry also described a meeting about potential cooperation.

I liked the discipline in BWXT's statement. Committing money because a market sounds exciting is not the same as building against a paying customer. Its existing naval nuclear and government work did not disappear this week, and Poland has not become a signed order simply because discussions are happening at a high level.

The share price fell 7.17% over the week and finished Friday at $127.19. I cannot attribute that entire decline to Poland. Thursday's selloff also hit the wider technology and semiconductor complex. But I have to distinguish what BWXT already earns from the opportunities I want it to win.

The price action itself tells the story: BWXT fell more sharply than the Nasdaq-100, and our inverse-index hedge did little to protect a large nuclear position. A broad technology hedge cannot substitute for an underwriting decision about one nuclear company.

BWXT reports November 2. Before the release, I want a written assessment of backlog conversion, margins, guidance, and any capital commitment tied to new nuclear projects. If those weaken, the thesis needs to be rewritten. If the business reports well but the stock remains volatile, the question becomes whether 22.8% is an appropriate position size. A good company can still be too much of a portfolio.

Celestica: a $310 financing price isn't a floor

Celestica finished Friday at $260.01. In August it sold approximately 11.1 million newly issued shares at $310, raising about $3.45 billion in gross proceeds. The shares are now roughly 16% below that offering price.

I don't think the offering price establishes fair value. Investors who participated at $310 can be wrong, and issuing shares brings dilution alongside the cash. What it does provide is a useful question for the next earnings call: what is that money going to produce?

Celestica is in an important part of the physical AI supply chain. Its DS6005 switch is designed for 102.4 terabits per second using co-packaged optics, which puts optical connections close to the processing hardware. That can help the power and bandwidth problem, but specifications are only the starting point. I need customer adoption, margin evidence, and a return on the capital being raised and spent.

The stock fell 14.42% in a week when the broad Nasdaq-100 fell less than one percent. It could be a violent overshoot ahead of earnings. It could also be the market lowering what it will pay for years of anticipated growth.

Celestica reports after Monday's close and holds its Investor Day on Tuesday morning. I'm not adding before those events. I'd rather be late to a real improvement than early to an assumption.

The argument against the financing scare

I can make a strong case against my own concern.

TSMC and ASML have already delivered hard evidence of semiconductor demand. Lam and other equipment companies have described capacity constraints. Meta's underlying operating business remains healthy: excluding a $9.89 billion legal charge, operating income was around $22.9 billion, roughly a 35% margin. Meta also had $82.45 billion in cash and securities against approximately $83.7 billion in debt.

One quarter of negative free cash flow doesn't put Meta in the same category as CoreWeave. Meta can choose to spend more than it generates in a particular quarter because it has a huge, profitable business and substantial liquidity. A smaller infrastructure operator dependent on bond and equity markets has much less room to make that choice.

Even the macro signals were mixed. Brent crude settled at $116.72 on Friday, a warning for energy-intensive infrastructure, but the ten-year Treasury yield was around 5.00%, lower than the previous week's 5.09%. Rates did not simply rise every time inflation or oil made news.

The most important counterevidence is still ahead. Microsoft, Alphabet, and Amazon have not yet provided the reports I want to evaluate. They are large enough that their cash-generation capacity could confine the financing stress to weaker borrowers instead of the entire AI buildout. If that's what their statements show, I want to reduce protection rather than defend an increasingly stale bearish position.

What PSQ is protecting, and what it isn't

PSQ is a fund designed to deliver the inverse of the Nasdaq-100's daily return, before fees and expenses. The daily part is important: over time, compounding can make its multi-day result differ from simply taking the negative of the index's cumulative return.

At Friday's close it represented 31% of the portfolio. The immediate reaction is to call that too large. The underlying stocks, however, have also been more volatile than the index. A 60-session sensitivity estimate from the research desk puts the stock sleeve's historical response around 1.46 times daily Nasdaq-100 moves. On that estimate, roughly $5,140 in stocks carries about $7,500 of index-equivalent exposure, against $2,383 in PSQ. It is a rough sensitivity calculation, not a stable hedge ratio or a promise about future losses.

The index was down less than one percent this week; our stocks were down 5.47%. That's the risk the estimate cannot solve. PSQ is an imperfect answer to sudden repricing in a nuclear company, an equipment manufacturer, or a specialty foundry. It can gain on a bad Nasdaq day while those holdings lose much more for reasons outside the index.

Issue 023 gave me three reasons to review reducing the hedge toward 15%: a Nasdaq-100 close above about 30,035; a meaningful retreat in CoreWeave's bond yields from roughly 15%; or evidence that the major AI customers can finance their spending comfortably.

As of Friday, I don't consider any of those tests satisfied. The Nasdaq-100 closed at 28,957.53; CoreWeave's bond yields had briefly passed 16% and returned near 15%; and Meta's results raised new questions rather than settling the funding issue.

I am not rewriting that rule just because a smaller PSQ position would feel easier to own. I am, however, making the third condition measurable before the next earnings arrive.

For Microsoft, Alphabet, and Amazon, I'll apply the same three-part screen. First, did quarterly operating cash flow cover capital expenditure, including relevant finance-lease cash payments? Second, did the company avoid announcing a new financing structure specifically needed to support AI expansion? Ordinary refinancing doesn't automatically fail that test, and an authorized share class isn't the same thing as an issuance. Third, did management maintain or raise its spending plans rather than retreat from its investment thesis?

If at least two of the three companies satisfy all three conditions, and the Federal Reserve does not deliver a materially more restrictive surprise on Wednesday, I'll review reducing PSQ toward 20% of the book on Friday, October 30. If the Nasdaq-100 also closes above 30,035, the existing review toward 15% remains on the table. If one or none pass, the hedge stays in place for now, without an additional allocation.

That is a decision framework, not an automatic order or an assertion that three quarterly cash-flow statements tell the whole story. Timing differences, lease accounting, and the companies' existing cash reserves still matter. But setting the conditions before the results arrive will make it harder to invent an explanation for whatever the market does next.

BWXT requires its own risk review. Reducing PSQ won't repair a concentrated nuclear position, and adding another hedge to cover BWXT would make the portfolio harder to understand. I'd rather reassess the underlying position size and the business evidence directly.

What I'm not doing

• The hedge doesn't get resized simply because the week felt uncomfortable.
• Celestica and Tower don't get new capital before their reports merely because their stocks fell 14%.
• I'm not rushing Ciena, Coherent, or Micron back into the portfolio simply because their prices have declined; each needs a fresh investment case.
• I'm also withholding judgment on Micron's Taoyuan labor negotiations: 1,633 of 1,637 voting union members rejected the offer, and talks resume Tuesday, but current reporting does not establish a production stoppage.
• Finally, I have no interest in stacking another inverse fund on top of PSQ to disguise a mismatch that would be better solved with position sizing.

The next receipts

Monday, October 26, Celestica reports after the close. Its Investor and Analyst Day follows Tuesday at 8 a.m. Eastern. I want to know how the capital raised in August is being deployed and what return management expects from it.

Tuesday, October 27, also brings Micron's Taoyuan labor talks, plus earnings from Seagate and Baker Hughes after the close. For Micron, the difference between a rejected offer and an actual production disruption is central to any re-entry research. For Baker Hughes, I'm watching power equipment orders and margins.

Wednesday, October 28, GE Vernova reports before the opening bell, the Federal Reserve decides policy at 2 p.m., and Microsoft and Alphabet report after the close. That is the first real test of the financing screen and a separate check on power-infrastructure demand.

Thursday, October 29, brings GDP and September PCE inflation at 8:30 a.m. Eastern, followed by Amazon and Reddit after the close. Amazon completes the three-company financing test. Friday's employment cost index adds one more read on inflation before the hedge review.

On Monday, November 2, Tower Semiconductor reports at 10 a.m. Eastern and BWXT follows after the close. For Tower, I want silicon-photonics demand and utilization. For BWXT, I want backlog turning into revenue at margins that justify its size in the book, plus a straight answer on how it is approaching opportunities such as Poland.

The next dollar needs a better map

Issue 022 asked who pays. Issue 023 watched the bond market start charging a higher price for the answer. This week, the companies themselves put different responses on the table: borrowing, preserving financing options, and spending less.

The demand for physical AI infrastructure still looks real. I also see a growing difference between companies that can fund the next stage from their own operations and companies that need someone else to provide the money first. The market may have moved too far in punishing some of those suppliers. It may also be correcting expectations that were too generous to begin with.

My portfolio didn't experience the quiet week the index reported. It had a concentrated loss in businesses that matter to the long-term thesis, while a large index hedge barely helped. That's information about the way I've assembled the book, not just about how stocks traded for five days.

Next week the largest buyers give us better evidence about their ability to fund the buildout. I've written down how that evidence changes my view of PSQ, and BWXT has its own test the following Monday.

The index hid the damage. The next set of receipts should help show whether it also hid an opportunity.

See you next Sunday.

— Alex

Research and commentary only. Concentrated stock positions and daily-reset inverse ETFs involve substantial risk. The 60-session sensitivity estimates depend on the lookback period and are not forecasts.

Source desk.

• SpaceXQ3 2026 results and SEC exhibit
• Meta PlatformsQ3 2026 earnings and Class C announcement
• TeslaQ3 2026 Form 10-Q and spending guidance
• BWXTPoland investment clarification, Oct 22
• CelesticaAugust 2026 $3.45B equity offering
• CelesticaDS6005 co-packaged optics introduction
• ReutersSpaceX cell-site negotiations reporting
• ReutersCoreWeave private-credit financing pressure
• NasdaqNasdaq-100 official historical close
• ProSharesPSQ fund objective and daily inverse exposure
• CelesticaQ3 2026 results / investor day schedule
• Federal ReserveOctober 27–28 FOMC calendar
• U.S. BEAOctober 29 GDP and personal income/PCE release schedule
• U.S. BLSOctober 30 employment-cost report schedule
• Tower SemiconductorNovember 2 earnings notice
• BWXTNovember 2 earnings notice

Research and commentary only. Concentrated equities and daily-reset inverse ETFs involve substantial risk. Historical correlations and sensitivity estimates can change and do not predict future returns.