The Buyers Showed
Their Hand.

The test was worth writing down

Last Sunday I gave myself a test before Microsoft, Alphabet and Amazon reported. I wanted to know whether their operating cash covered capital expenditure, including relevant lease payments; whether they avoided a new financing structure specifically necessary to expand AI; and whether they kept their spending plans intact. Two clean passes out of three, without a hawkish surprise from the Fed, would put a reduction in my PSQ hedge toward 20% on the table.

I got one pass. I didn't get the permission to trim.

Alphabet generated $42.3 billion of operating cash in the quarter and spent $51.4 billion on capital expenditures. Google Cloud revenue rose 103% to $30.7 billion, yet the cash investment ran about $9.1 billion ahead of operating cash. Its filings also show how aggressively it has been raising money this year: roughly $30.5 billion of common equity, $19 billion of mandatory convertible preferred stock, and $80.8 billion of senior notes. Some of those proceeds are explicitly available for expanding AI infrastructure. This is a company with an immensely valuable core business, not a distressed borrower. But it failed the cash-coverage part of my test.

Amazon generated $42.4 billion of operating cash and reported $75.9 billion of cash capital expenditures, net of certain proceeds and incentives. The quarterly cash gap was about $33.5 billion. Its trailing-twelve-month free cash flow was negative $42.4 billion, compared with positive $14.8 billion a year earlier, and its cash-flow statement shows $48.1 billion in long-term debt proceeds during the quarter. AWS revenue grew 39% to approximately $46 billion. Amazon still plans around $220 billion in cash capital spending this year and expects the bill to rise meaningfully in 2027. Demand passed. Quarterly self-funding didn't.

Microsoft deserves a different reading. Operating cash flow reached $58.0 billion, against $39.4 billion of cash purchases of property and equipment; even its broader $45.2 billion capital-spending measure, which includes new finance leases, came in below operating cash. Its cloud growth was strong, and management still expects spending to accelerate, although some third-party capacity deliveries shifted. The cash and spending tests pass.

Then, the day after its call, Microsoft priced $25 billion of senior unsecured bonds. That's important, but I won't move the goalposts to turn it into a failure. The rule I published asked about a new financing structure specifically needed to support AI. Microsoft's offering was ordinary corporate debt, for general purposes that may include capital expenditures. A financially strong company choosing long-term borrowing is different from a borrower that cannot proceed without an inventive new vehicle. It makes the cost of the buildout more visible; it doesn't erase Microsoft's cash coverage. I count Microsoft as a qualified pass.

So the score is one of three, against the two I required. Had I applied an even stricter test barring every new bond issue, the score would have been zero. The portfolio decision would be the same, but the distinction matters: my test was about funding capacity, not about whether strong companies ever use debt.

The Federal Reserve held its policy rate at 3.75% to 4.00% on Wednesday. Two policymakers wanted a quarter-point increase. With the company test already short of the threshold, there is no reason to retrofit that vote into a separate excuse to trade.

The seller is joining the financing table

The larger discovery this week wasn't another hyperscaler cash shortfall. It was what some suppliers are doing to keep the orders moving.

Nvidia announced it would invest $3 billion in Meta's equity offering while expanding a partnership under which Meta plans to deploy more Nvidia infrastructure. Meta also outlined a much larger capital-raising program, including common stock, convertible preferred shares and a proposed future at-the-market facility. The latter is capacity to sell shares later, not money already raised. And Nvidia's investment doesn't mean every dollar of equipment revenue is being recycled. It does mean a supplier is also becoming an investor in a major customer.

Broadcom offered an even clearer example. On October 26, a group of banks and investment firms announced a $63 billion financing vehicle to purchase Broadcom accelerators and networking equipment and lease it to Anthropic, with deliveries expected in 2027. Broadcom is providing residual-value support on part of the financing. That's a contingent obligation, not a $63 billion cash check from Broadcom. But if the hardware has to be recovered and resold under adverse conditions, the supplier has agreed to carry some of the risk.

These structures might be perfectly sensible. A reliable customer, scarce hardware and financing matched to long-lived assets can produce excellent returns. The question is how the arrangement behaves when prices fall, demand changes, or the useful life of hardware turns out to be shorter than the financing. A supplier that makes money on an order and also underwrites part of the buyer's ability to pay deserves a different kind of research.

This gives me a new question for the companies I own: how much of the growth is customer-funded, and how much requires the supplier's capital, inventory, guarantees or equity? I still care whether the orders are real. Now I want to know how the cash travels between everyone involved.

The book barely moved. The risk didn't disappear.

The Nasdaq-100 lost 1.98% last week and the Nasdaq Composite fell 2.00%. The S&P 500 was off 0.57%, while the Russell 2000 gained 1.29%, although small caps fell on Friday too. The technology damage was concentrated late in the week; Friday's Nasdaq-100 decline alone was 3.46%.

The Frontier File research portfolio finished October 30 at $7,683.80, down $9.06, or 0.12%, from the previous Friday.

The stock sleeve lost approximately $55. PSQ returned about $46 of that; BWXT gained roughly $48, Rocket Lab $15, GE Vernova $11 and Celestica $7. Reddit alone cost about $60, while Alphabet and SpaceX each took roughly $20 from the book.

Those offsets were useful, but they didn't come from one coordinated defense. BWXT rallied while much of technology weakened. PSQ benefits from daily Nasdaq-100 declines, not from every risk facing BWXT, Celestica or Reddit. Together BWXT and PSQ now make up about 55% of the book. A nearly flat weekly result is not evidence that the concentration has become safe.

The hedge passed its rule. Its design still needs work.

At the Friday close, PSQ represented 31.6% of the book. The review toward 20% did not trigger. The earlier Nasdaq-100 level for reviewing a reduction toward 15%, around 30,035, also remained out of reach; the index finished at 28,384.85. I haven't established a reliable enough bond-yield mark to declare the separate CoreWeave credit trigger satisfied either. So I am leaving the PSQ allocation unchanged for now and not adding to it.

That is consistency, not a fresh endorsement of the size.

PSQ seeks the inverse of the Nasdaq-100's daily performance. The financing risk I'm worried about is slower and more uneven: refinancing costs, lease commitments, customer prepayments, equipment residual values and the amount of cash suppliers must put up before they get paid. Those pressures can hit individual holdings while the index rises, and PSQ can lose value during a recovery even if the credit questions are unresolved. Its daily reset also makes longer holding-period returns path-dependent.

The argument against my defensive position is straightforward. Cloud revenue is growing quickly across all three big buyers. None of them abandoned its buildout. They have reserves, profitable businesses and access to capital on terms that less-established builders can only envy. One quarter in which capital spending exceeds operating cash does not establish an insolvency problem or a bad investment. If the market starts treating the cash gap as a deliberate, affordable investment cycle, the hedge could become expensive very quickly.

I want a better framework before the next change. In addition to the existing Nasdaq-100 review level, the next hedge review will separate two questions: whether the largest buyers can keep funding their plans on reasonable terms, and whether the long positions actually behave like the index I'm shorting. I won't quietly replace a published rule after a volatile Friday; I also won't keep a permanent one-third inverse position simply because its original thesis once made sense.

BWXT needs its own treatment. At 23.5% of the book, it's a company-specific concentration with an earnings report due Monday after the close. Strong backlog and margins would reinforce the business case, but they would not automatically justify letting the position grow unchecked. Disappointing guidance or weakening conversion of backlog into revenue would make a reduction toward a 20% weight more compelling. If the weight rises past 25% through price alone, I want an explicit concentration review even without bad news.

Celestica's numbers improved. The cash math got harder.

Celestica's quarter made the demand case better. Revenue grew 83% to $5.83 billion, the adjusted operating margin reached 8.5%, and management lifted its 2026 revenue outlook to $21 billion. Its 2027 forecast is $40 billion of revenue, a 9% adjusted margin and $21.50 in adjusted earnings per share. At Friday's $265.53 close, that is roughly 12 times management's 2027 adjusted EPS estimate.

For a business forecasting that kind of growth, 12 times earnings looks inviting. But it's a forecast on a non-GAAP basis, two years ahead, from a manufacturer that needs to keep buying components and expanding capacity ahead of delivery. It is not a promise of cash that shareholders will receive.

Alongside the higher revenue guidance, Celestica lowered its 2026 free-cash-flow outlook from $600 million to $460 million. Management attributes the change primarily to working-capital investment for next year's expansion, and it expects $1.5 billion to $2 billion of capital expenditure in 2027. That could be sensible preparation for contracted demand. It could also mean more of Celestica's money is tied up in inventory and capacity while customers collect the benefit of faster deployment.

The stock reached $299.80 on Wednesday before closing Friday at $265.53, still roughly 14% below the $310 price of its August equity offering. Neither the offering price nor the low valuation multiple is a floor. I'm keeping Celestica as a Growth position at roughly 4.5% and watching cash conversion. If working capital begins coming back as cash and management maintains its 2027 margin target without another financing round, that is a stronger reason to consider adding than the stock's recent fall.

Reddit's front door is changing

Reddit reported $895 million in revenue, up 53%, with $283 million in net income and $363 million in free cash flow. It spent about $100 million repurchasing shares during the quarter at an average price near $150. By Friday, the stock had closed at $91.73 after falling about 29% in one day and 36% for the week.

The operating report doesn't look like a broken business. The distribution problem is real enough to change its outlook, though. CEO Steve Huffman said search referrals had been falling by more than a million per week on average in October, and that continued deterioration could push total daily unique users down by a low double-digit percentage in the fourth quarter. He suspects Google's AI Overviews and AI Mode are a major reason, alongside poorer rankings for machine-translated content outside the U.S. He cannot inspect Google's ranking decisions, so neither can I call that a proven cause.

The user mix matters. U.S. daily unique users were down 2% year over year, but logged-in users grew while logged-out users declined. Reddit says logged-in and app users account for more than 80% of monetization. So the traffic being lost may be less valuable per user than the core audience it retains. The company is preparing login-required tests in some international markets, which may help direct relationships but could also make casual discovery harder.

At about $19 billion in estimated fully diluted equity value, with roughly $3 billion of cash and over $3 billion in trailing revenue, Reddit is considerably less expensive than it was a week earlier. I still don't know whether it's cheap. The fourth-quarter revenue guide of $1.00 billion to $1.03 billion gives me the next test: can revenue and logged-in engagement hold up even if search traffic continues to erode? Until I see that evidence, this remains a small, roughly 1.4% Growth position under review, not a dip-buy.

The cleaner receipts

GE Vernova offered a useful counterexample to the financing-heavy stories. It reported $29.3 billion in orders, a $197 billion backlog and $5.4 billion in quarterly free cash flow, while raising guidance and expanding its repurchase authorization. That is stronger evidence of cash generation from infrastructure demand. It doesn't mean every order has the same economics, but it moves the discussion beyond a large pipeline of potential contracts.

Honeywell Aerospace also reported a solid quarter as a standalone company, including $2.21 of adjusted earnings per share and a higher full-year outlook, while declaring its first dividend. The shares still declined over the week. Sometimes the business receipt and the price reaction disagree without giving me an obvious trade.

And Intel provided a reminder that AI scarcity has consequences outside the data center. Management described memory shortages constraining the PC market and reduced its view of future PC units. Higher memory pricing might benefit suppliers such as Micron, but it can also raise costs and suppress volumes elsewhere. That's a reason to keep studying the memory opportunity, not a reason to buy it without a fresh valuation.

What I'm not doing

• I'm not trimming PSQ just to make the position look smaller. The review conditions did not clear, although the hedge-design problem deserves a new, explicit rule.
• I'm not adding to PSQ after a rough technology Friday. More Nasdaq protection wouldn't solve the concentration in BWXT or the funding questions at individual suppliers.
• I'm not buying Reddit because it fell 36%. I want evidence that its most valuable users and revenue can outgrow the decline in search referrals.
• I'm not chasing Celestica's $40 billion forecast. The free-cash-flow cut and next year's capital requirements are part of the same investment story.
• I'm not filling the book with new names. Micron, Ciena and Coherent remain research candidates; each needs an entry price and an updated operating case, not merely nostalgia for a former position.

The next receipts

Monday, November 2: Tower Semiconductor's third-quarter results and morning call will be the next test of silicon-photonics demand, utilization and cash funding for new capacity. BWXT reports after the close. With nearly a quarter of the book in BWXT, backlog conversion, margins, guidance and its approach to opportunities such as Poland all matter.

Tuesday, November 3: AMD and Arista report. I'll look for evidence that spending on AI accelerators and the networks connecting them remains broad enough to support supplier expectations, and for clues about customer concentration.

Wednesday, November 4: Microsoft's $25 billion note offering is scheduled to settle. The financing terms are another useful cost-of-capital marker, not by themselves a verdict on Microsoft.

Friday, November 6: The October employment report arrives, and Oklo holds its third-quarter business update. For a speculative nuclear developer, the milestones, fuel path and financing requirements matter far more than a price reaction.

Tuesday, November 10: October CPI adds a fresh inflation reading before the next round of rate expectations settles in.

Who keeps the money?

I've spent several issues asking who will pay for the infrastructure the AI industry wants. This week the buyers answered a large part of that question. They intend to build, and they have access to an extraordinary variety of funding. Some of it comes from the cash their businesses generate. Some comes from shareholders and bondholders. Increasingly, some of the risk is being shared with the people selling them the equipment.

That changes what I want to know about the suppliers. A full order book is good. A full order book backed by a customer that can pay without leaning on the supplier's balance sheet may be better. Before I give any company credit for the next wave of demand, I want to know where the working capital sits, who guarantees the hardware, and how much cash makes it all the way back to shareholders.

The Frontier File book ended a rough technology week almost unchanged. I'll take the stability, but I'm not mistaking it for proof that the portfolio is finished being tested. The hedge did some work. So did a nuclear holding. The uncomfortable part is that they were protecting me from different things.

Next week I want the companies, not the index, to tell me what deserves the next dollar.

See you next Sunday.

— Alex

Source desk

1. SOURCE 01Microsoft FY27 Q1 results and October 29 debt offering.
2. SOURCE 02Alphabet third-quarter Form 10-Q.
3. SOURCE 03Amazon third-quarter Form 10-Q and earnings release.
4. SOURCE 04October 28 FOMC statement.
5. SOURCE 05Nvidia and Meta collaboration, including equity investment and Reuters on Meta's financing plans.
6. SOURCE 06Broadcom/Anthropic financing announcement.
7. SOURCE 07Celestica third-quarter results and outlook.
8. SOURCE 08Reddit third-quarter results and earnings-call transcript.
9. SOURCE 09GE Vernova third-quarter results.
10. SOURCE 10Honeywell Aerospace third-quarter results.
11. SOURCE 11Intel third-quarter results.
12. SOURCE 12Associated Press market recap and Nasdaq-100 historical data.
13. SOURCE 13ProShares PSQ fund objective.
14. SOURCE 14November 2026 BLS release schedule.
15. SOURCE 15Company investor-relations notices: Tower, BWXT, AMD, Arista, and Oklo.

Research commentary; figures refer to October 30 closing marks. Portfolio concentration and daily-reset inverse funds carry material risk.