Sovereign Weekly

Speakers analyze Nvidia's funding risks, Microsoft's capital intensity, and private credit stress signals in late 2026.

1h 25m Listen to the original
NvidiaMicrosoftOpenAIPrivate CreditADPOil MarketsSemiconductorsCloud Infrastructure

Summary

The episode examines Nvidia's $250bn credit guarantee rejection and rising obligations, noting that despite doubled revenue, stock performance lagged due to extended receivables and external financing dependencies. Microsoft's 60,000 layoffs are framed as a reallocation to AI talent rather than cost savings, while its infrastructure shift raises concerns about declining returns on tangible capital and circular revenue relationships with OpenAI. Private credit markets show signs of strain with CCC spreads widening, and ADP's duration risk is highlighted as a potential vulnerability. The speakers also discuss oil market dynamics, US-China chip export controls, and McDonald's franchise margin pressures.

Episode sections

  1. 0:00 – 9:26

    Nvidia's New Leadership and Funding Chain Risks

    Discussion on Nvidia's interim CEO rejecting OpenAI credit guarantees, rising obligations, and revenue growth vs stock performance.

  2. 9:26 – 13:30

    Tech Layoffs and Cost Reallocation

    Analysis of Microsoft's 60,000 layoffs and ServiceNow's retention costs as talent reallocation rather than margin expansion.

  3. 13:30 – 18:10

    Alibaba Valuation and Governance Discount

    Discussion on Alibaba's 70% discount to base value, governance issues, and cloud growth strength.

  4. 18:10 – 29:45

    The Scarcity of HBM and Packaging

    Identification of HBM and TSMC's COWS packaging as the scarcest AI component, concentrating margin power.

  5. 29:45 – 34:49

    Credit Spreads and Private Credit Strain

    Analysis of CCC spread blowouts, risk migration to private credit, and liquidity rationing signs.

  6. 34:49 – 38:47

    Private Credit and Bank Duration Risk

    Prediction of high yield spread widening and bank vulnerabilities from long-duration bonds and uninsured deposits.

  7. 38:47 – 42:19

    ADP's Hidden Banking Exposure

    Examination of ADP's $46bn client float, unrealized losses, and funding via commercial paper.

  8. 42:19 – 51:46

    Oil Markets and Strait of Hormuz Dynamics

    Argument that oil price declines are due to surplus forecasts, with zero spare capacity behind the Strait of Hormuz.

  9. 51:46 – 55:33

    US-China Chip Export Controls

    Analysis of US export controls as 'permission slips' and prediction of Beijing's rare earth response.

  10. 55:33 – 1:00:23

    US Economic Cooling and Inflation

    Discussion on Philadelphia Fed Index drop, inventory buildup, and persistent cost-push inflation.

  11. 1:00:23 – 1:04:30

    McDonald's Franchise Model Risks

    Analysis of franchisee margin squeezes, high borrowing costs, and weak consumer sentiment.

  12. 1:04:30 – 1:08:39

    Microsoft Financial Analysis and Valuation

    Deep dive into Microsoft's owner earnings yield, accounting changes, and sustainability of returns.

  13. 1:08:39 – 1:10:11

    Azure Growth and Capital Intensity

    Examination of Microsoft's declining return on tangible capital and increasing capital intensity for Azure growth.

  14. 1:10:11 – 1:24:47

    Microsoft Financials and Capital Structure

    Review of Microsoft's commercial backlog, cloud margins, and balance sheet strength.

Assets discussed

AssetKindStanceContext
NVIDIANVDAequityneutralNvidia's new boss has made her biggest move by saying no. Jensen Huang is out, and interim CEO Colette Cress is reportedly fighting the plan he championed a $250 billion credit guarantee for OpenAI.
AlibabaBABAequitymixedOver the last decade, Alibaba turned each retained dollar into about 50 cents. It placed 710 310 million new shares at around 42% of what we think they're worth.
FS KKR Capital CorpFSKequityneutralFSKKR trades at 58 cents on the dollar of its own stated book. That means the market thinks the loans are worth a lot less than the fund says.
AMERICAN EXPRESS COAXPequityneutralAmerican Express runs a deposit funded bank charter. $155 billion of deposits funds its revolving cardbook, and 92% of the deposits at its bank are insured.
Crude oilcommoditymixedSpare capacity. That's oil that producers could switch on quickly if someone else is. else's barrels disappear.
GoldcommoditybearishAnd gold, the traditional war hedge?
Shell plcSHELequityneutralOkay, who does this actually reach? Shell.
MicrosoftMSFTequityneutralMicrosoft. Microsoft. We've had it on the show several times this autumn, and it came up earlier tonight.

“Stance” is how the speakers talked about the asset, not a recommendation.

Claims worth verifying

Nvidia's interim CEO Colette Cress fought a $250 billion credit guarantee plan for OpenAI championed by former CEO Jensen Huang.

Unidentified speaker (speaker_0) ·0:00

How to verify: Check Nvidia press releases or SEC filings (Form 8-K) for leadership changes and any disclosed contingent liabilities or guarantees related to OpenAI. Verify Jensen Huang's resignation date (Nov 12) and Colette Cress's appointment via company announcements. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Nvidia reported revenue of about $116 billion for the quarter, roughly double a year ago, with gross margin near 75% and free cash flow of about $44 billion.

Unidentified speaker (speaker_0) ·1:58

How to verify: Review Nvidia's Q3 FY2026 earnings release and 10-Q filing on SEC EDGAR. Verify revenue, gross margin, and free cash flow figures against the reported numbers. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Nvidia's total obligations (supplier commitments plus guarantees) hit $670 billion, including $118 billion in guarantees and a $32 billion commitment line to Neoclouds.

Unidentified speaker (speaker_0) ·2:46

How to verify: Examine Nvidia's 10-K or 10-Q filings for 'Commitments' and 'Contingencies' sections. Verify the specific breakdown of guarantees and lines of credit. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Nvidia's days sales outstanding (DSO) increased from 60 days to 64 days, and its 10-Q filing was delayed.

Unidentified speaker (speaker_0) ·3:25

How to verify: Calculate DSO from the 10-Q (Accounts Receivable / Revenue * Days). Check SEC EDGAR for the filing date of the 10-Q relative to the earnings release. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Nvidia's guidance for the next fiscal year is $741 billion and explicitly considers customers' access to capital.

Unidentified speaker (speaker_0) ·4:03

How to verify: Review the transcript of Nvidia's Q3 earnings call or the press release for the specific guidance figure and the exact wording regarding 'access to capital'. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Anthropic priced its IPO, raised $81 billion, and was valued above $2 trillion, carrying over $900 billion in outstanding spending commitments.

Unidentified speaker (speaker_0) ·6:06

How to verify: Check the S-1 filing for Anthropic's IPO on SEC EDGAR for the raise amount and valuation. Verify spending commitments in the 'Risk Factors' or 'Management's Discussion' sections. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Anthropic lost about $81 billion against $16 billion of revenue in the first half of the year.

Unidentified speaker (speaker_0) ·6:26

How to verify: Review Anthropic's financial statements (likely in S-1 or 10-K) for H1 revenue and net loss figures. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Microsoft offered voluntary buyouts to 60,000 people, representing about 27% of its workforce of ~223,000.

Unidentified speaker (speaker_0) ·9:50

How to verify: Check Microsoft's press releases or SEC filings (Form 8-K or 10-Q) for workforce reduction announcements and headcount data. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

ServiceNow set aside $211 million for acquired talent, more than the $167 million spent on severance.

Unidentified speaker (speaker_0) ·10:44

How to verify: Review ServiceNow's 10-Q or 10-K for 'Restructuring' and 'Compensation' expenses. Verify the specific amounts for severance and retention/acquired talent awards. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

CCC-rated credit spreads hit 13.61%, up more than 300 basis points in three months.

Unidentified speaker (speaker_0) ·30:24

How to verify: Check ICE BofA High Yield Index data or Bloomberg/Reuters for CCC spread levels on the specified dates (Thursday vs 3 months prior). For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Banks are not the problem today; the credit desk's read is a repricing, not a funding break.

Unidentified speaker (UNKNOWN) ·34:49

How to verify: Review the Federal Reserve's Senior Loan Officer Opinion Survey on Bank Lending Practices for the relevant period. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

The bank survey is dated October 1st, before the CCC added another chunk and before the 10-year spiked.

Unidentified speaker (UNKNOWN) ·34:49

How to verify: Check the release date of the Federal Reserve's Senior Loan Officer Opinion Survey for the October 2026 period. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Private credit funds are paying out 43 cents on the dollar.

Unidentified speaker (speaker_0) ·35:09

How to verify: Search for reports from private credit managers or indices (e.g., Preqin, Bloomberg) on distributions or haircuts in late 2026. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

High yield spreads will go through 4% within the next few months.

Unidentified speaker (UNKNOWN) ·35:30

How to verify: Monitor ICE BofA US High Yield Index spreads against the stated forecast window. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

The January loan officer survey will show tightening clearly higher than 1.7% for the six months out.

Unidentified speaker (UNKNOWN) ·35:30

How to verify: Review the Federal Reserve's Senior Loan Officer Opinion Survey for January 2026. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Banks hold about $4.8 trillion in treasuries and agency bonds, with $327 billion in paper losses.

Unidentified speaker (speaker_0) ·36:30

How to verify: Check the Federal Reserve's Financial Accounts of the United States (Z.1) report for bank holdings of securities. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Uninsured deposits are back to 40% of the total, highest since late 2023.

Unidentified speaker (speaker_0) ·37:15

How to verify: Review the Federal Reserve's Deposit Market Share Report for November 2026. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

American Express runs a deposit-funded bank charter with $155 billion of deposits, 92% insured.

Unidentified speaker (speaker_0) ·37:35

How to verify: Review American Express's Q3 2026 10-Q filing for deposit balances and insurance status. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

ADP owes $46 billion to clients within a year, invested in a bond portfolio.

Unidentified speaker (UNKNOWN) ·38:52

How to verify: Review ADP's Q3 2026 10-Q filing for client funds and investment portfolio disclosures. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

ADP's unrealized loss on client funds portfolio went from $460 million to $1.3 billion in a single quarter.

Unidentified speaker (UNKNOWN) ·39:34

How to verify: Review ADP's Q3 2026 10-Q filing for changes in fair value of investment securities. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Commercial backlog is about $677 billion.

Unidentified speaker (UNKNOWN) ·1:10:11

How to verify: Check Microsoft's latest quarterly earnings release or 10-K filing for 'commercial backlog' or 'remaining performance obligations'. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Return on all capital, goodwill included, is about 27%.

Unidentified speaker (UNKNOWN) ·1:10:11

How to verify: Calculate from Microsoft's balance sheet and income statement (Net Income + Interest Expense * (1-Tax Rate)) / (Total Assets - Goodwill - Intangibles + Debt) or check analyst consensus for 'Return on Invested Capital'. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Cloud gross margin rose a point to 66%.

Unidentified speaker (UNKNOWN) ·1:10:11

How to verify: Check Microsoft's latest quarterly earnings report for 'Cloud Intelligence' segment gross margin. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

GPU cost cut by about 84% by swapping small models into PowerPoint.

Unidentified speaker (speaker_0) ·1:10:31

How to verify: Verify via Microsoft's internal technical reports or blog posts detailing 'PowerPoint' AI implementation and cost savings. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Alphabet's debt went from about $16 billion to about $100 billion.

Unidentified speaker (speaker_0) ·1:10:59

How to verify: Check Alphabet's SEC filings (10-K/10-Q) for 'Total Debt' or 'Long-term Debt' over the relevant period. For recent events, verify 2026-08-21 through 2026-11-21. Use another period only when the speaker explicitly names it.

Primary source

Do your own research

  • Verify Nvidia's 10-Q filing for DSO trends and guarantee disclosures.
  • Check Microsoft's quarterly earnings call transcript for CapEx details and OpenAI revenue recognition.
  • Review SIFMA or Federal Reserve data for private credit liquidity and CCC spread trends.
  • Analyze ADP's 10-K for client float composition and commercial paper maturity schedules.
  • Monitor EIA and OPEC+ reports for Middle East spare capacity and Strait of Hormuz transit data.
  • Assess US Treasury yields and Microsoft's owner earnings yield for valuation context.
  • Review McDonald's franchise disclosure documents for same-store sales and cost inflation trends.
Full transcriptexpand

0:00Unidentified speaker (speaker_0): Nvidia's new boss has made her biggest move by saying no. Jensen Huang is out, and interim CEO Colette Cress is reportedly fighting the plan he championed a $250 billion credit guarantee for OpenAI. Six days later, she reported revenue of about $116 billion, roughly double a year ago. The stock closed at one $155, hundred and fifty five, about a third below where our desk was comfortable. That isn't a personal. personnel story. It's a story about who pays for all of this. It's Saturday, november twenty first, this is Sovereign Weekly. Along the way, I'll show you why the scarcest thing in AI isn't the famous chip, but the memory bolted next to it. Then the weakest borrowers, whose spreads blew out by three hundred basis points in a single quarter. And oil, pricing a piece the ships in Hormuz haven't delivered. Stay for Microsoft, which is offering 60,000 people the door. Is it still a high margin software business, or has it become a capital heavy utility? But first, NVIDIA and why a quarter that doubled left the stock where it is. The most consequential thing NVIDIA's new boss has done so far is say no. Last week we covered NVIDIA as the company the whole AI build-out leans on. Since then, Jensen Huang has resigned, and the interim CEO, Colette Cress,

1:19has reportedly fought a plan he championed: a $250 billion credit guarantee for OpenAI. And that's not a personnel story. That's a story about who pays for all of this. Quick facts first. Huang resigned on November 12th. The board's statements cited personal conduct inconsistent with company standards. Reports also allege ties to Chinese intelligence, and that side of it belongs to the World Desk later. Cress was CFO. Now she's interim CEO. The board hired a search firm, Hydric and Struggles, and has formally ruled out bringing Huang back. Then, six days later, she ran her first earnings call.

1:57Unidentified speaker (UNKNOWN): And

1:58Unidentified speaker (speaker_0): the print was, honestly, monstrous. Revenue of about $100. one hundred sixteen billion dollars for the quarter, roughly double a year ago, gross margin near seventy five percent, free cash flow of about forty four billion, around twice the prior quarter. So the stock closed at one hundred fifty five dollars. That's about a third below the level where our desk was comfortable holding it. And I know exactly what your paying. you're thinking. Revenue doubles, margins go up, cash doubles, and the stock gets a third knocked off it? Fair question. It's the second one. It's looking at the funding chain.

2:45Unidentified speaker (UNKNOWN): The

2:46Unidentified speaker (speaker_0): number I can't stop staring at is this one. Nvidia's total obligations, meaning supplier commitments plus guarantees, hit six hundred and seventy billion dollars. Inside that are one hundred and eighteen billion in guarantees and a thirty two billion dollar commitment line to the Neoclouds, the smaller rented rent a GPU outfits. So Nvidia isn't just selling the shovels. it's cosigning the loans for some of the people buying them. Okay, wait. Look at what that does to the receivables. Days sales outstanding is how long it takes NVIDIA to get paid after it books a sale. That went from sixty days to sixty four. And the quarterly filing that shows the cash details. detail, the ten Q is delayed.

3:56Then the detailed filing that would settle the question shows up late, and the guidance for

4:02Unidentified speaker (UNKNOWN): the first time.

4:03Unidentified speaker (speaker_0): next fiscal year, which is seven hundred and forty one billion dollars, explicitly considers our customers' access to capital. Nvidia wrote that sentence itself. Think about it. Nvidia's forecast now depends on whether its customers can raise money.

4:24Unidentified speaker (UNKNOWN): So where

4:24Unidentified speaker (speaker_0): does Cress's know fit?

4:26Unidentified speaker (UNKNOWN): I

4:26Unidentified speaker (speaker_0): think the market read it slightly wrong in both directions. Let me back up.

4:30Unidentified speaker (UNKNOWN): The

4:31Unidentified speaker (speaker_0): bull read is simple. A CFO turned CEO won't write a quarter trillion dollar blank check. Fewer guarantees, cleaner balance sheet, good for owners.

4:40Unidentified speaker (UNKNOWN): That's

4:40Unidentified speaker (speaker_0): how our own morning brief scored it on the fifteenth. Positive. And I still think that's right for NVIDIA the business. But watch the call itself. CJ Mews asked her straight about vending. vendor financing. She pivoted to investment grade prepayments. Someone else asked about margin compression. Something for the future.

4:59Unidentified speaker (UNKNOWN): Then

4:59Unidentified speaker (speaker_0): she was asked how she ranks the uses of cash. She called buybacks a super top priority, and a minute later said, the top priority is going to be our suppliers. Two top priorities. That's not a ranking, that's a mood.

5:13Unidentified speaker (UNKNOWN): And

5:13Unidentified speaker (speaker_0): it matters because the buyback. authorization just went up by $150 billion, to about $227 billion remaining. Call it $45 billion a quarter. That's a company with an enormous amount of cash competing for it. Shareholders want it back. Suppliers want it up front. And customers want it as a guarantee. Crest picked her side on the guarantee.

5:35Unidentified speaker (UNKNOWN): The

5:35Unidentified speaker (speaker_0): second order bit, the part I don't think people have priced, is who that no hurts. It doesn't hurt NVIDIA. It hurts whoever was counting on NVIDIA's balance sheet standing behind them.

5:46Unidentified speaker (UNKNOWN): And

5:46Unidentified speaker (speaker_0): look at where the rest of the chain went this week. Taiwan Semiconductor, which we've talked about on this show before, has since then reportedly started demanding cash up front from AI chip clients over circular financing worries. The chipmaker is saying, pay me first. Anthropo Anthropic, which we covered last week as it lined up its listing, has since then actually priced the IPO. It jumped 17% on day one, raised $81 billion, and was valued above $2 trillion. It also carries over $900 billion in outstanding spending commitments. In the first half alone, Loan it lost about $81 billion against $16 billion of revenue. Lost $5 for every dollar it took in. And the public market paid up anyway. So read the whole picture together. The supplier wants cash up front. The vendor won't co-sign, and the buyers are going straight to public shareholders for the money. The funding chain isn't snapping, it's being handed to a different set of owners. That's the shift. The risk is migrating out of Nvidia's balance sheet and into equity markets. Which is part of why. tech has lagged the SP 500 by about eight percentage points over three months. The Nasdaq 100 is sitting a hair below its 200-day average, and the volatility index finished Friday above 28, up from about 19 a week ago.

7:06Unidentified speaker (UNKNOWN): Now

7:06Unidentified speaker (speaker_0): the bear case on NVIDIA itself. And it isn't stupid, it goes like this. You don't lose a founder like Huang and keep everything. Elon Musk said critical. deals will be lost without him. Two big holders, Brad Gerstner at Altimeter and Gavin Baker at Atreids, with about $2 billion between them, wrote to the board urging a deliberate process. Huang still keeps the right to nominate two directors. Cress is promising raises and equity before December to calm staff. And then comes concentration. Three direct customers make up 36% of revenue. If the customers access to capital tightens and NVIDIA just refuse to be that access, the forecast has a hole in it. That's a real argument. I'll concede two pieces outright. The governance overhang lasts until a permanent CEO is named, and that's months, not weeks. And the key person risk is live. Our own verdict is formally under review for exactly that reason, plus a cash test we can't run until the 10k. 10Q lands. But, Hopper GPUs got recontracted at up to twice their original price, according to management. Rental rates on the newer B300 parts are up more than 60%, also per management, so treat both with salt. Hyperscale revenue grew 23%, and it beat its own revenue guide again even after you strip out China. That's not a business losing customers,

8:26it's a business who whose customers are fighting to pay it. So the call. I think the $250 billion OpenAI guarantee is dead in anything like its proposed form. I'd put that at about two in three, and I think that's good for anyone who owns NVIDIA. A vendor that stops financing its own demand ends up with smaller but realer numbers. The cost lands on the Frontier Labs and the Neo Clouds. They'll pay more for money. sell more equity, or build slower. For the next two quarters, NVIDIA trades on governance and cash collection, not on ship demand. What proves me wrong? The 10Q. If days to get paid are back at or under 60 and cash conversion clears 70%, then the receivables wobble was timing and the market's discount was overdone. If receivables stretch past 64 or any of those 118 billion in guarantees start getting called, I've been too generous. In that case, the funding chain is already leaning on NVIDIA harder than CRESS is letting on.

9:24Unidentified speaker (speaker_1): Okay,

9:25Unidentified speaker (UNKNOWN): second

9:25Unidentified speaker (speaker_0): story.

9:26Unidentified speaker (speaker_1): Same

9:26Unidentified speaker (speaker_0): theme, different end of the income statement. People,

9:30Unidentified speaker (UNKNOWN): the

9:30Unidentified speaker (speaker_0): world spent the week talking about layoffs. Upwork, Renault, Westpac, Fortis Games, and the giant one, Microsoft. It came up on this show a few weeks ago, and since then, Satya Nadella has put a frame on its voluntary buyout offer to sixty thousand people. Sixty thousand. That's about twenty-seven percent of a workforce of around two hundred and twenty-three thousand. Microsoft gets the full treatment in the deep dive at the end of the hour. So I'll stay disciplined. But one detail belongs here. Nadella pitches it as an agency transformation, with the example of an 11 developer team building AI agents. Small teams, big leverage. And yet, the cost line shows operating expenses rising while headcount falls. Read that again. Fewer people. Higher costs.

10:24Unidentified speaker (UNKNOWN): And

10:24Unidentified speaker (speaker_0): it's not a one-off. ServiceNow. which we discussed back in October, has since then disclosed something that makes the same point even more starkly. It set aside $211 million for acquired talent. That's more than the $167 million it spent on severance for the people it let go in the US, the UK, and India. So it's paying more to bring certain people in than it's saved letting other people in. out. Right. And Uber, which we looked at a couple of weeks ago, cut 10% of its workforce back in September. The stock bounced a couple of percent that day. Since then, its CFO has been redirecting rider incentives into driver incentives in Canada, Brazil, and India. The cut didn't end the spending, it moved it. And flip back to NVIDIA for a second. Its interim CEO is handing out raises and equity to stop people leaving. One company is paying to keep engineers while another is paying to shed them. Same week.

11:26Unidentified speaker (UNKNOWN): So

11:26Unidentified speaker (speaker_0): the listener's question at this point is obvious. If companies are cutting tens of thousands of jobs, isn't that just a margin story? Fewer salaries, fatter profits, and owners win? That's the textbook version. It's what the market rewarded with Uber in September.

11:42Unidentified speaker (UNKNOWN): And

11:42Unidentified speaker (speaker_0): I don't think it's what's happening.

11:44Unidentified speaker (UNKNOWN): What

11:44Unidentified speaker (speaker_0): the numbers show is reallocation, not savings.

11:47Unidentified speaker (UNKNOWN): The

11:47Unidentified speaker (speaker_0): money doesn't fall to the bottom line. It goes to compute, to a smaller number of very expensive people, and to acquisitions. The cut is a funding source. It's how you pay for the capital build when you don't want to borrow more, or don't want another NVIDIA saying no to. you.

12:03Unidentified speaker (UNKNOWN): Now

12:04Unidentified speaker (speaker_0): the Steel Man. Restructuring always costs more up front. Buyouts means severance, acquired talent means retention awards, and the savings only show up a year or two out. Judge Microsoft's OpEx in the quarter the buyouts are paid, and you'll always see it go up. That's accounting timing, not a broken thesis. Fair, and I'll give that point real weight. Part of the cost rise is one. time. If I'm being honest, I can't separate the one-off chunk from the run rate chunk with what's been disclosed. But ServiceNow also granted about $127 million in leadership awards in late October. Those aren't one time. They vest for years. And Microsoft is offering voluntary exits to engineers for the first time. That's not trimming fat, that's changing which people the company thinks it needs. needs. So my call.

12:53Unidentified speaker (UNKNOWN): I

12:53Unidentified speaker (speaker_0): think the companies doing these cuts mostly will not show margin expansion from them in 2027. Better than even, maybe 6040. The savings get eaten by compute and by the price of the people they keep. To expect higher profits. It's a signal about where management thinks the next dollar earns more, and that's a capital allocation bet you have to judge on its own. I'm wrong if two quarters from now, Microsoft and ServiceNow both show operating expenses falling in line with headcount. If the cut actually lands on the bottom line, the textbook was right and I was overthinking

13:30Unidentified speaker (UNKNOWN): it.

13:30Unidentified speaker (speaker_0): Okay, valuation corner. One company, done properly. Alibaba. It came up briefly on the show last week and since then it's reported a quarter and been hit with analyst price cuts.

13:43Unidentified speaker (UNKNOWN): And

13:43Unidentified speaker (speaker_0): it's now the most lopsided gap between price and value anywhere in our coverage. At about $81, the price sits roughly 70% below our base value. That's the margin of safety. Put the other way, the price would have to more than triple to reach our base case. More than triple on one of the largest companies in China.

14:03Unidentified speaker (UNKNOWN): So

14:03Unidentified speaker (speaker_0): what does the price assume? Run it backwards, and today's price implies Alibaba's normal cash earnings shrink about 16% a year, every year for a decade. Even our conservative case leaves a margin of safety above 50%. The price also sits more than 40% below what it'd be worth if it simply never grew again.

14:24Unidentified speaker (UNKNOWN): And

14:24Unidentified speaker (speaker_0): I can feel you reaching for the obvious objection. If it's that cheap, everybody can see it, so what does everybody else know? Good. Let's do that properly, because the bear case on this one is strong. Three parts. One, the commerce engine, the core marketplace money machine, is at its ceiling. Like for like, customer management revenue, basically what merchants pay to get seen, grew 1%. Share is about 40% and falling. Merchants can list on rival platforms at near zero cost. 2. The build may never stop. Trailing capital spending is about $26 billion. Free cash flow is negative, around $14 billion. The cloud profits cover less than a fifth of a quarter's capex.

15:09Unidentified speaker (UNKNOWN): And

15:09Unidentified speaker (speaker_0): the date for positive free cash flow has slipped to the second half of 2028. 3. And this is the one that bites. Owners may never get paid. Our management test asks whether every dollar kept back in the business turned into at least a dollar of market value. Over the last decade, Alibaba turned each retained dollar into about 50 cents. It placed 710 310 million new shares at around 42% of what we think they're worth. Buybacks got

15:38Unidentified speaker (UNKNOWN): a good one.

15:39Unidentified speaker (speaker_0): cut to under 400 million in the half. And control sits with a partnership that owns under 2% of the company. And this quarter didn't help. Revenue beat, but they didn't disclose the like for like merchant number the whole thesis needs and leaned on non operating gains to flatter reported profit.

15:58Unidentified speaker (UNKNOWN): So

15:58Unidentified speaker (speaker_0): why isn't that the end of the story? Because the trough is coming from two spending. programs, not from a broken business, and both are already shrinking. Sales and marketing, excluding stock pay, fell from about 27% of revenue to about 15%. The subsidy war is winding down. E-commerce adjusted profit rose about two and a half times. And cloud is compounding. External cloud revenue up 64%, and an AI product. run rate above $10 billion, and segment margin up from 10% to about 14.5%.

16:33Unidentified speaker (UNKNOWN): The

16:33Unidentified speaker (speaker_0): US chip license numbers that came out this week 75,000 advanced chips, half what Tencent got, matter for the cloud ceiling.

16:42Unidentified speaker (UNKNOWN): But

16:42Unidentified speaker (speaker_0): the why of that belongs to the World Desk.

16:45Unidentified speaker (UNKNOWN): So

16:45Unidentified speaker (speaker_0): what's the market mispricing?

16:47Unidentified speaker (UNKNOWN): I

16:48Unidentified speaker (speaker_0): think it's pricing the governance problem as if it were a business. problem. Those are different discounts. A business problem shrinks the cash. A governance problem means you might not see the cash. The price is charging Alibaba for both, at full weight, forever. My call.

17:05Unidentified speaker (UNKNOWN): I

17:06Unidentified speaker (speaker_0): think the commerce engine holds roughly flat to slightly positive on that like-for-like like for like line. I'd say better than even, maybe 60%. And I think the cloud keeps widening.

17:17Unidentified speaker (UNKNOWN): The

17:17Unidentified speaker (speaker_0): governance discount doesn't go away. That partnership still controls it with under 2%. So the gap to value closes slowly, through earnings recovering rather than the multiple being forgiven.

17:29Unidentified speaker (UNKNOWN): I

17:30Unidentified speaker (speaker_0): lean that the business is fine, and I hold that view with less confidence than I'd like, because the owner problem is real, and the tripwire is explicit. If like for like customer management revenue turns negative for two quarters in a row, the thesis is broken.

17:45Unidentified speaker (UNKNOWN): The

17:45Unidentified speaker (speaker_0): Commerce engine isn't at a ceiling anymore, it's going backwards.

17:50Unidentified speaker (UNKNOWN): So,

17:50Unidentified speaker (speaker_0): pull the week together. Nvidia's interim CEO wouldn't co-sign. The chipmaker wants cash first. Anthropic went straight to public shareholders, and across big tech, the layoffs aren't fattening margins, they're feeding the bills. build. Everybody is scrambling to figure out who actually pays for the AI build-out. That's what moved prices, which leaves the obvious thing we haven't touched: the physical stuff all that money is buying, and why it's so scarce that suppliers can demand to be paid up front. That's next from first principles. Okay, the physical stuff. And I'll start with a slightly odd claim. The scarcest thing in this whole buildup build out isn't really the chip everybody talks about. It's the memory bolted next to that chip and the assembly process that joins the two together. Because that's where the money's actually pooling, not in the famous processor, in the stuff around it. So let's build it from the ground up. What does one of these AI systems physically do all day? When a model answers you, that's inference, meaning the model being used rather than trained. It writes its answer one small moment. chunk at a time, and for every single chunk, it has to read through its parameters. The billions of learned numbers that basically are the model, pulling them out of memory and into the processor. That word-by-word stage has a name, D code. Hang on to it because it comes back at the end. The maths on each chunk is easy

19:10for the processor, embarrassingly easy. The hard part is hauling the numbers in. So how fast the system runs is set by bandwidth, how much data per second you can pour from memory into the processor. It's not set by how clever the processor is. Which is, okay, let me give you the picture because I think it maps really well. Think of the processor as a chef who can chop impossibly fast. Ordinary memory is a pantry down the hall, through one narrow door. So

19:38Unidentified speaker (UNKNOWN): the

19:38Unidentified speaker (speaker_0): world's fastest. chef spends most of the evening standing there waiting for onions. High bandwidth memory, HBM, is what you get if you rebuild that pantry as a tower of shelves bolted right onto the side of the stove, and you make the door as wide as the whole wall. And each piece of that maps onto the real thing. The tower is memory chips stacked on top of each other. When you hear twelve 12 high, that's 12 layers in one stack. The wide door is an enormous number of connections running between the stack and the processor. And bolted onto the stove means they share the same package, which raises the obvious question what's a package? Packaging is how separate chips get physically joined, wired together, and mounted, so they behave like one part. The method that matters here is TSM. TSMCs. It's called COWAS, chip on wafer on substrate, which is a mouthful. All it means is the processor and the memory towers sit side by side on a shared base layer. That keeps the wiring between them incredibly short and dense. Then the whole sandwich sits on a substrate, a A board-like board like foundation that carries power in and signals out. So, three layers of scarcity stacked on each other: the memory towers, the joining line, and the foundation board

20:57underneath.

20:58Unidentified speaker (UNKNOWN): Now

20:58Unidentified speaker (speaker_0): the new part, the part that's one level below the press release. It's called custom HBM. Every memory tower has a bottom floor, the base die. Think of it as the building's ground floor lobby, the bit that directs traffic in and out of all the floors above. For years, that lobby was a fairly plain chip the memory company made for itself. What's changing is that the base die is turning into proper logic, a real processor grade chip. It's customized for each buyer and it's made it a foundry, meaning a contract chipmaker. In practice, that's TSMC. So think about what that does to the supply chain. Memory used to be its own lane. Now memory supply chain. runs through TSMC's packaging line, which means TSMC, which we've covered in recent weeks as the foundry for everybody's processors, is now also the toll booth on everybody's memory. Huh? Yeah, both doors, one gatekeeper. And then the next squeeze.

21:56Unidentified speaker (UNKNOWN): The

21:56Unidentified speaker (speaker_0): material flags substrates, those foundation boards, as the next shortage. You can see the changes. chain reacting in real money. Amcore, a big contract packaging firm, is tripling its investment at its Guangju site in Korea to about $2.3 billion. So, a one-line recap you can carry. The processor is fast, the processor is starved, and the thing that feeds it is now a custom part built through one company's assembly line.

22:25Unidentified speaker (UNKNOWN): Okay,

22:25Unidentified speaker (speaker_0): so why does an investor investing show care? Look at who's actually keeping the money. Gross margin is what a company keeps from each dollar of sales after paying to make the product. Micron came up last week inside the broader chip sell-off. What I want to add since then is the mechanism from underneath its margin. Micron is keeping close to 87 cents of every dollar. As we discussed in the week, Nvidia is the famous one. Its margin is 75. TSMC's is about 66. The memory company is out earning the processor everyone obsesses over? Memory. The bit of a chip story I used to skip, honestly. Commodity stuff, priced like wheat. Not anymore. Same story in Korea. SK Heinex, which we covered a few weeks back, has since reported a roughly seven-fold jump in profit and about $45 billion going back to shareholders. And that's the second order point. Hyperscaler CapEx, meaning what the giant cloud companies

23:35spend on buildings and machines, hit about $200. two hundred and thirty billion dollars in a single quarter. Those dollars don't stay with whoever writes the check. They leak down the chain to whoever owns the narrowest door. Which is exactly why back in the week the chipmaker could say cash first. When you're the only door in the wall, you set the terms. And there's a stranger signal. of how valuable this memory has become. There's an export loophole. Expensive HBM from South Korea, packaged with cheap processors that aren't sanctioned, can legally go to China. So the memory is the controlled substance. The processor is just the delivery vehicle. Which means if Washington closes that gap, a gray stream of demand disappears overnight. That's a real policy risk to the memory sellers. And it isn't in anyone's spreadsheet as a line item. Now, I know exactly what you're thinking. If memory is this scarce and the margins are this fat, why did the stocks get crushed? Micron lost about a third of its value in a month, fair question. So what's the answer?

24:51Unidentified speaker (UNKNOWN): The

24:51Unidentified speaker (speaker_0): answer is the market is betting this is the top of a cycle. It's treating that near 87% margin as a one-year event, a peak, not a new normal. Memory has a long history of boom, then flood, then bust. So investors aren't arguing with the scarcity today, they're arguing with how long it lasts. And I want to give the bear case its full strength, because it isn't stupid. Memory is the most cyclical business in tech, full stop. Fat margins are a flare in the sky, telling every competitor to build. Samsung, which came up a few weeks ago, is working to qualify its own 12 high stacks. Qualify means pass the customer's testing so they'll actually buy it. If Samsung gets there at volume, that's a third big supplier into a three-player market. Then there's a private outfit called TerraFab. claiming a target of roughly 700,000 memory wafers a month. That's about Samsung's entire DRAM capacity. And underneath all that, the deeper threat. Researchers are proposing ways to move decode, that word-by-word stage, off HBM entirely. Fiber attached memory, something called high bandwidth flash. If the pantry gets rebuilt some other way, the tower

26:02Unidentified speaker (UNKNOWN): is a very important thing.

26:02Unidentified speaker (speaker_0): on the stove design loses its monopoly on speed. That's a serious case. And on the long view, I'll concede it. Five years out, the memory moat is genuinely exposed.

26:14Unidentified speaker (UNKNOWN): But

26:14Unidentified speaker (speaker_0): this is where being honest about maturity really matters. Those threats are not all the same kind of thing. What's shipping? Micron's HBM 3E three E is confirmed inside shipping B 300 three hardware. hundred That's real. In machines, earning money. What's underway? Samsung and TSMC are expanding base dye capacity for custom HBM. That's real investment, but it's capacity being built, not capacity delivered. What's a claim? Terafab. Fab. No qualified process? No wafers.

26:49Unidentified speaker (UNKNOWN): A

26:49Unidentified speaker (speaker_0): number in a report, no. What's a paper? Fiber attached memory and high bandwidth flash are archive proposals, meaning research write-ups. No product.

27:00Unidentified speaker (UNKNOWN): So

27:00Unidentified speaker (speaker_0): the bear case leans hardest on the two least mature items on that list. And there's one more piece the bears tend to skip. That custom based eye quietly Slightly changes the economics of memory. Once the bottom floor of the tower is designed for one buyer and built in a a TSMC, the stack isn't a commodity you can swap in from anybody. It's a custom part. Customers can't just flip to whoever's cheapest next month. Wheat doesn't get custom designed for each bakery. That's kind of the whole point. Now, one honest wobble in my own case. Micron paused fab work in Taiwan with no reason given. That supports pricing in the near term, sure. but it could also mean Micron itself thinks the cycle is late. I'd be lying if I said that doesn't nag at me.

27:46Unidentified speaker (UNKNOWN): So

27:46Unidentified speaker (speaker_0): my call. I think the scarcity holds through the next year, and the market's one year event framing is wrong on timing. Not wrong forever, wrong on timing. I'd put it at about two in three that memory margins stay near today's level through the next couple of years. prints. The physical bottleneck, the joining line, and the substrates underneath it is getting tighter, not looser, and the custom based die makes switching suppliers harder. For anyone who owns these businesses, that means the value keeps leaking toward the bottleneck layers, the memory makers, the foundry, the packaging houses, and away from the companies borrowing to buy the finished machine. machines. Over five years though? I lean the other way. That's where Samsung's catch up and the research on moving decode off HBM have time to bite. Scarcity is a one to three year asset here, it's not a permanent one. So what would prove me wrong? If Micron's december twenty 23rd third quarter shows gross margin falling below about eighty 84.5%. four and a half percent. If Samsung qualifies 12 high stacks at real volume, or the big one, if hyperscaler CapEx actually falls quarter on quarter, and memory orders start getting cancelled. Any of those, and the bears had it

29:06right, and I'll say so. Because notice what that last one really is. It isn't a technical risk at all. Every layer I just walk walked you through, the memory, the packaging, the boards, is only scarce because somebody keeps writing $200 billion checks every quarter, and a lot of those checks are borrowed. SpaceX borrowed $40 billion just for processors. Oracle's free cash flow has gone deep into the red. So the real question behind the scarcest part in tech is a very boring one. Who's lending against it? And that takes us straight into the banks and the credit plumbing, which is where we're going next. So, who's lending against it? The honest answer is everybody, at very different prices. And the cheapest money and the most expensive money told two different stories this week. Start with the expensive end because that's where it cracked. Last week we touched credit spreads in passing. Since then, the bottom tier has blown straight through its alarm. Quick gloss. A spread is the extra yield a company pays over the government to borrow, and it pays you for one thing, the chance you don't get your money back. CCC is the weakest rated rung, the company's one bad year from default. On Thursday, their spread hit

30:2413.61%. The data danger line the desk uses is 10. And that's up more than 300 basis points in three months. 300 in a quarter.

30:36Unidentified speaker (UNKNOWN): Now

30:36Unidentified speaker (speaker_0): look at the rest of the junk market. High yield overall? 3.5%. Up five basis points on the week, nine since the start of the month. Investment grade, the blue chips, is 94 basis points. That's fine. flat. It hasn't moved at all.

30:53Unidentified speaker (UNKNOWN): So

30:53Unidentified speaker (speaker_0): the tail is on fire and the body's sitting in a deck chair. And then Friday happened. The SP dropped nearly 4% in a day, and the VIX, the market's fear gauge, closed near 36. That's double where it was on the 11th. Which means the order of events matters here. Credit's weakest borrowers were flashing red weeks before equities. broke.

31:15Unidentified speaker (UNKNOWN): Okay,

31:16Unidentified speaker (speaker_0): let me make that CCC number concrete because a spread on its own is abstract. A CCC company refinancing today pays the 10-year treasury just under 5%, plus that spread. Call it 18.5% all in. 18.5% on a company that was probably borrowing at half that a few years ago. And that isn't a forecast, it's arithmetic. You can't cover interest at 18% unless your business is booming. And these businesses, almost by definition, aren't. So who's actually lending to them? This is where it gets messy, because the bank tape looks fine, better than fine. Business loans at commercial banks are up 11.4% on the year. Over the last three months, they're running at something like 15% annual. annualized. Delinquencies on those loans, people behind on payments, are 1.27%, and that's falling. And I know exactly what you're thinking. If the weakest borrowers are being priced for defaults, why are banks lending faster than they have in ages? And why are their own loans getting healthier? It doesn't fit, somebody's wrong. Fair. Both can be true, and the reason is who the banks are lending to. Banks' loans to non-bank financial firms, the private credit funds, are up just over 20% on the year, past $2 trillion.

32:35So the bank doesn't lend to the shaky company directly, it lends to the fund, and the fund lends to the shaky company. The risk didn't vanish, it moved one step down the pipe, out of sight of the bank's delinquency numbers. And you can see the strain where those funds are listed. Business development companies are the publicly traded private lenders, BDCs for short. FSKKR trades at 58 cents on the dollar of its own stated book. That means the market thinks the loans are worth a lot less than the fund says. Then there's PIK, payment in kind. Plain English, the borrower can't pay the interest in cash, so the lender adds it to the loan. At New Mountain, nearly a fifth of income is that kind of IOU. At Aries Capital, it's about a sixth. Which, I mean, a fifth of your income being we'll pay you later. That's not income. That's a promise to pay with an invoice attached. And the non-traded funds, the ones sold to wealthy individuals with promised liquidity, 16 out of 54 were rationing withdrawals last quarter. That's the highest in 10 quarters. The median investor asking for their money back got 43 cents of each dollar requested.

33:46Unidentified speaker (UNKNOWN): So

33:47Unidentified speaker (speaker_0): that's what has happened. Now, what does it imply?

33:50Unidentified speaker (UNKNOWN): Let

33:50Unidentified speaker (speaker_0): me argue the other side properly first because it's a good argument.

33:54Unidentified speaker (UNKNOWN): The

33:54Unidentified speaker (speaker_0): bull case says CCC is a small, weird corner of the market. It always blows out first and often blows out alone. High yield at 3.5 is middle of the range, not stress. Commercial paper, the short term IOU's big companies roll every few days, is up about $95 billion in three months. Money is still flowing, and the thing that historically leads to faults is bank lending standards, which barely moved. That last point deserves a second, so quick gloss. Every quarter the Fed asks loan offers. officers whether they're tightening. When banks tighten, the marginal borrower gets cut off. And about six months to a year later, that borrower defaults. Standards lead to faults because cutting off credit is what actually causes defaults. Right now, a net 1.7% of banks are tightening on business loans. The danger line is 20. Three years ago it was nearly 51.

34:49Unidentified speaker (UNKNOWN): So

34:49Unidentified speaker (speaker_0): on the bank survey alone, there's no credit crunch coming. And I'll concede that point fully: banks aren't the problem today. The credit desk's own read is a repricing, not a funding break, and I agree with them on that. But I'd push on the survey. It's dated the 1st of October. That's before CCC added another chunk, before the 10-year 10 year spiked, before Friday. A survey of whether doors are closing, taken before. the wind picked up. And the bigger thing is this. If the weakest borrowers now live in private credit, then the bank survey measures the wrong door. The rationing is already happening, it's just happening at funds paying out 43 cents on the dollar.

35:29Unidentified speaker (UNKNOWN): So

35:29Unidentified speaker (speaker_0): here's where I land.

35:30Unidentified speaker (UNKNOWN): I

35:30Unidentified speaker (speaker_0): think the widening spreads from the tail into the broader junk market. High yield goes through 4% within the next few months. I'd put that at better than even, maybe 60%. And I think the January loan officer survey shows tightening clearly higher than 1.7. That's the six months out piece. Once banks start trimming what they lend to the funds, the funds have to trim what they lend to the companies. What proves me wrong? CCC coming back under 12, high yield holding below 4 four through the January survey, and that survey still shows. showing basically nobody tightening. If all three happen, the tail was just the tail, and I'll say so.

36:09Unidentified speaker (UNKNOWN): Now

36:10Unidentified speaker (speaker_0): the other half of the weighing this week, and it's the part nobody talks about. The banks are loading up on duration. In English, they're holding a lot of long-dated bonds, and long bonds lose value when rates rise. Quick primer on what a bank does with your deposit. It either lends it out or buys a bond with it. Deposits are piling in, near 19.9 trillion, close to the record. Loans can't absorb all of that. So banks hold about 4.8 trillion in treasuries and agency bonds. That's the 99th percentile going back to the 70s. And the paper losses on those bonds are $327 billion. About 12% of the bank's combined equity. We've seen this film. 2023, the regional banks. Paper losses only matter when you're forced to sell, and you're forced to sell when depositors leave. So are depositors leaving? Not really. Deposits dipped two tenths of a percent on the week, which is noise. But uninsured deposits are back.

37:15to forty percent of the total, highest since late twenty three. That's the money that runs. And regional bank shares fell about six percent from the eleventh to Friday. One small mercy. The tenure actually came in twelve basis points this week, to four point nine two. So the loss stack got a bit lighter, not heavier. Follow that uninsured deposit number down to one lender and it reads. very differently. American Express runs a deposit funded bank charter. $155 billion of deposits funds its revolving cardbook, and 92% of the deposits at its bank are insured. So against a system where 40% is the money that runs, Amex's funding is mostly the money that stays. That's the point of the charter. It frees the lending book from depending on wholesale. markets, which is exactly where this week's stress lives.

38:07Unidentified speaker (UNKNOWN): What

38:07Unidentified speaker (speaker_0): it doesn't buy is cheap money. Those deposits cost 3.56% the price of rate sensitive online savings, so there's no cost edge over a bank with checking accounts.

38:19Unidentified speaker (UNKNOWN): And

38:19Unidentified speaker (speaker_0): if credit losses climb over the next few quarters, the cushion is capital, CET1 at 10.9% against

38:26Unidentified speaker (UNKNOWN): the cost.

38:27Unidentified speaker (speaker_0): a 7% minimum.

38:28Unidentified speaker (UNKNOWN): A

38:28Unidentified speaker (speaker_0): 2020 style year would pause buybacks. It wouldn't force a capital raise. Which brings me to a second business this lands on, and it isn't a bank. Last week we covered the labor market and payrolls. Since then, the new piece is what's happening inside ADP, the payroll company.

38:47Unidentified speaker (UNKNOWN): You

38:47Unidentified speaker (speaker_0): probably think of it as software. It's actually a quiet participant in exactly this plumbing.

38:52Unidentified speaker (UNKNOWN): When

38:53Unidentified speaker (speaker_0): your business runs payroll through ADP, the money sits with them for a few days before it reaches your employees and the tax authorities. ADP owes $46 billion to clients within a year, and it invests that float in a bond portfolio laddered out five and ten years.

39:09Unidentified speaker (UNKNOWN): That

39:10Unidentified speaker (speaker_0): float earns them about $1.6 billion of revenue next year on management's guide. Sound familiar?

39:16Unidentified speaker (UNKNOWN): It's

39:17Unidentified speaker (speaker_0): a bank's balance sheet wearing a payroll company's clothes. So it has the same two exposures I just walked through. First, duration. The unrealized loss on that client funds portfolio went from about $460 million to $1.3 billion in a single quarter.

39:34Unidentified speaker (UNKNOWN): The

39:34Unidentified speaker (speaker_0): funds now trail what they owe clients by about $1.3 billion. Second, funding.

39:40Unidentified speaker (UNKNOWN): To

39:41Unidentified speaker (speaker_0): avoid selling those bonds at a loss, ADP rolls short-term money, commercial paper on roughly two-day maturities, and repo. Each averaged about $5.4 billion last quarter at roughly 3.7%. The portfolio earned about $3.6, so they were paying slightly more to borrow than the bonds earned. A thin, slightly negative carry, sitting right on the markets where funding stress shows up first.

40:08Unidentified speaker (UNKNOWN): And

40:08Unidentified speaker (speaker_0): that's why the plumbing matters to them. Since we last went through the funding plumbing, the reverse repo facility, the system's overflow tank is still empty, 40 million short of half a billion. Reserves ticked up to just under 3 trillion. The Treasury's account is steady at around 900. 917 billion. Discount window borrowing is sitting at about 9 billion, which is high for a quiet market but stable. None of that's a freeze. So for ADP, this is a slow burn, not a shock. They've got 11.7 billion of committed credit lines, nothing drawn. And through 2008 and 2009, 9, their owner earnings actually rose.

40:51Unidentified speaker (UNKNOWN): The one

40:51Unidentified speaker (speaker_0): date I'd circle is June 2027. That's when a 5.7 billion 364-day facility needs renewing. If private credit is still rationing then and banks have started tightening, that renewal gets priced differently. Which is a strange sentence to say about a payroll Roll company. But that's the point. The plumbing touches businesses that never think of themselves as borrowers.

41:15Unidentified speaker (UNKNOWN): My

41:15Unidentified speaker (speaker_0): call on the plumbing itself? It doesn't bite yet. Maybe two and three it stays quiet through year end, month end bulges aside.

41:24Unidentified speaker (UNKNOWN): The

41:24Unidentified speaker (speaker_0): thing that flips me is reserves dropping below 2.8 trillion. We're a long way above that.

41:31Unidentified speaker (UNKNOWN): So

41:31Unidentified speaker (speaker_0): the short version the cheap money's fine and the expensive money's breaking. And the expensive money is where the AI, neo clouds, and the private credit. funds live. One more detail before we move on. Life insurers sold 18% more funding agreements this year, $302 billion of wholesale money, while their loans grew six. There's a live estimate that up to half the AI build out debt ends up on insurers. Apollo's chief is saying capital for a lot of those neo clouds is going to disappear, and Friday's equity break didn't come out of a credit event, it came out of a world that's getting more expensive to insure. Which is the question for the next segment wars, oil, hormuz. Is any of that actually in the price? Or did the market just find out on Friday? Let's go to the world.

42:19Unidentified speaker (UNKNOWN): And

42:20Unidentified speaker (speaker_0): the honest answer on Friday is no. The fear that hit hits stocks didn't come from the Gulf. Last week we covered the Middle East. meaning Hormuz, Brent, and the Iran blockade. Since then, oil has kept falling, and the Gulf hasn't gotten any more open. So that's the story. Oil is pricing a piece the ships haven't delivered. Start with the version you've been told, because you've definitely heard it. The midterms went against the White House. A Congress that doesn't like this war now constrains it. Qatar says. Washington wants a deal. The president says Tehran wants one. Everybody's tired. So the story goes: the war premium comes out of oil. And the price did exactly that. Brent, the global benchmark, closed Friday at about $92.5. That's down more than 10% since the third of the month. Okay, so what are the ships doing? The strait was moving. somewhere around four ships a day in the second week of November. Normal is roughly six or seven times that. Tankers, which are the bit that matters, were running at about one a day. Normal is closer to 14.

43:28One a day. I'm serious, look at it. That's not a straight reopening. That's a straight with the door held open a crack. Now, I'll be straight with you, the outside trackers don't agree with each other this week. One count had Hormuz averaging closer to seven a day by the 20th, and a weekly tally from the UK Maritime Trade Office logged 263 transits in the week to the 18th, which, if you divide it out, would be above normal. That doesn't fit any other series I have. It may be counting something different.

43:58Unidentified speaker (UNKNOWN): But

43:59Unidentified speaker (speaker_0): I'm not going to hide it, because if that number's real and it shows up in tang, tankers, it's the thing that proves me wrong. We'll come back to that. So, behavior, not statements.

44:11Unidentified speaker (UNKNOWN): What

44:11Unidentified speaker (speaker_0): is Washington actually doing while the market prices its exit?

44:15Unidentified speaker (UNKNOWN): On

44:15Unidentified speaker (speaker_0): the 7th, US Central Command aircraft fired on a commercial ship trying to break the blockade in the Gulf of Oman. The military announced a task force to choke off Iran's land trade by the and its Caspian route. Treasury kept adding sanctions designations. And there were strikes. The president called them minor. That isn't a country winding a blockade down, that's a country widening it. And the diplomacy everyone's pricing?

44:41Unidentified speaker (UNKNOWN): The

44:41Unidentified speaker (speaker_0): Qatari plan, as reported, is UN inspectors in, straight reopens, and no commitment on Iran's enriched stockpile. Which I mean, maybe. But the process. prediction market gives a final nuclear deal by year end under 10%. A normal straight by the 31st, about 8.5%. So the futures market trades like a deal's a coin flip, and the people betting real money on the deal itself say one in eleven. Right. Now why does that gap matter so much? Spare capacity. That's oil that producers could switch on quickly if someone else is. else's barrels disappear. A year ago the Middle East had about two and a half million barrels a day of it. In the October government outlook, Middle East spare capacity was zero. Not low, zero. And the reason is obvious once you say it. The spare barrels sit behind the strait. Saudi, Iraq, Kuwait and Iran between them are pumping five million barrels a day less than a year ago. You can't call on a car cushion you can't ship. So every new outage anywhere is a net loss. No one fills the hole. And now, the strongest case against me, because it isn't stupid. Futures aren't just a political bet. The later EIA figures our Africa desk picked up show unplanned outages down from over seven 7 million barrels a day

45:59to under 5, five, and a projected surplus through the turn of the year. Libya, Nigeria, and Algeria are all pumping near the top of their ranges. Venezuela's output is up. Freight has gone crazy, up about 130% in 13 weeks, and that kills demand at the margin. On that read, oil's falling because the world is quietly working around Hormuz. And I'll concede a real chunk of that. The world is rerouting. The Red Sea and Sue. Suez are carrying the load. Suez is running above its normal pace. That's why we're at 92, not 130. But that workaround is thin. The Red Sea back door runs past a live war. Saudi Arabia has been fighting the Houthis since July. Tankers through the southern mouth of the Red Sea, the Bab El Mendeb, are running somewhere between about sixty and sixty. seventy percent of normal. If that outlet takes a hit, you lose the one channel that's working, with no spare barrels behind it.

46:58Unidentified speaker (UNKNOWN): So

46:58Unidentified speaker (speaker_0): the surplus story is true right up until it isn't.

47:02Unidentified speaker (UNKNOWN): Now

47:02Unidentified speaker (speaker_0): the step that keeps this honest. Is any of it priced? Three numbers. Oil's down. The VIX, which is what traders pay to insure against big stock moves, is near thirty 36, six, up about seventy. 70% since early November. And U.S. high yield spreads, the extra yield junk borrowers pay, are about 3.5%, up less than a tenth of a point in the same window. So equity fear goes vertical, credit barely blinks, and oil falls. All in the same three weeks. Huh.

47:37Unidentified speaker (UNKNOWN): What

47:37Unidentified speaker (speaker_0): that tells me is the very VIX isn't pricing the Gulf. The politics desk ties that spike to AI, the election, and a run of cyber outages, which squares with what we just said in The System. The market's scared of something, but it isn't Hormuz. And gold, the traditional war hedge? Down hard, sitting at the bottom of its yearly range. Nobody's buying the Hormuz insurance policy. There's one more tell. The gap between Brent and WTI, the American crude. Brent's the seaborne world barrel, the one Hormuz actually touches. In early November, that gap had squeezed to about seven and a half dollars, which is what you'd expect if the market believed the strait was healing. On Friday's numbers, it's back above nine. So the cleanest spread in the oil market was started undoing the peace trade, quietly, while the headline price stayed down. And I know exactly what you're thinking. If the physical picture's this tight, why is the physical barrel falling too? Fair. Dated Brent, the price for actual cargoes, is down almost a fifth since late October. That's real, and it's the best point for the other side. My read is that physical traders are also leaning on the surplus forecasts and the rerouting. But they were paying well above futures at the start of the

48:56month, and that premium only disappeared as the political talk picked up. Ships didn't do that, politics did. Okay, who does this actually reach? Shell. We brushed past Shell a few weeks back. What's changed since is that its own filing now ties a piece

49:14Unidentified speaker (UNKNOWN): of the.

49:15Unidentified speaker (speaker_0): business directly to this week's ship count. Shell's big gas to liquids plant in Qatar, Pearl, had a production train damaged this year, and Qatari export volumes were shut in. Management used to talk about a restart in about a year. Now the restart is explicitly Implicitly conditional on passage through Hormuz. So for Shell, a peace bet in the future's market doesn't restart anything. Tankers do. And it cuts both ways, which is the part I find surprising. The same closure gave Shell refining cracks, the margin for turning crude into fuel, of around $42 a barrel, and trading profits that flattered the year. The House dossier puts it bluntly. Refining earns on other people's outages. Meanwhile, freight from the Middle East to China ran ten times year-ago levels, and the CFO conceded that was already pushing down refining, chemicals, and oil realizations. So the strait staying shut is a cost on one line and a windfall on another.

50:20Unidentified speaker (UNKNOWN): If

50:20Unidentified speaker (speaker_0): it reopens, Opens, the windfall fades, and Pearl gets its chance to come back. For a long-term long owner, term the that's a slow, lumpy reset, not a shock. A year or two of earnings that look different line by line, without the business changing underneath. So, the call.

50:38Unidentified speaker (UNKNOWN): I

50:39Unidentified speaker (speaker_0): think the House base case is right. Blockade and bargain, a stalemate. Call it three and five. Talks continue, enforcement widens, ships stay thin. And on that road, I think. oil's next big move is more likely up than down. sixty forty, maybe a touch better. Because the price already banked a piece the strait hasn't delivered, and there's no spare capacity to cushion a single bad day in the Red Sea.

51:06Unidentified speaker (UNKNOWN): For

51:06Unidentified speaker (speaker_0): someone who owns businesses for the long haul, Paul, the takeaway is less dramatic than that sounds. Don't build a plan that assumes cheap, still smooth energy and freight in the next year. Businesses with energy or shipping in their cost line are carrying more risk than this month's oil price suggests. What proves me wrong? Hormuz tankers running sustainably above about seven a day, and the Brent WTI TI gap compressing back below where it was in early November and staying there. If that UK maritime count is real, that's where it'll show. No date on it. Anyone who gives you one is selling something. Second story, and it's the one everybody argued about this week. Chips. The story you're being told. Washington's slamming the door on China. New rules on the 16th require licenses to export or transfer advanced. AI chips to five countries, four in Southeast Asia, plus Brazil. The point is closing the loophole where Chinese firms reach U.S. silicon through third countries. Beijing vows necessary measures. Congress has a bipartisan push to block AI chip sales to China outright. Quick gloss on how that bites. An export control isn't

52:26a tariff. Nobody pays more. It's a permission slip.

52:30Unidentified speaker (UNKNOWN): If

52:30Unidentified speaker (speaker_0): a data center in one of those five countries wants the chips, someone in Washington now has to say yes first. So the cost lands as delay and uncertainty, not price. Now behavior. Same week, a congressional disclosure shows commerce approved licenses for close to 900,000 advanced chips across 21. applications. We touched on Alibaba earlier in the show and last week, and I'm not redoing it. Since then, the numbers come out 75,000 chips. Ten centers. So the door that's slamming shut had nearly 900,000 chips worth of approved traffic going through it. Come on. No, the steel man deserves a fair hearing here. That data's date is six months old, dated May. The main U.S. supplier says it's stopped shipping that chip to China, even where licenses exist. And the same disclosure lists Alibaba among Chinese military companies. So, okay, maybe the approvals are history and the tightening is the present. I'll give that half. The approvals aren't a green light today.

53:36Unidentified speaker (UNKNOWN): But

53:37Unidentified speaker (speaker_0): look at what the executive did in the same way. fortnight. Extended the suspension of the rule that would have caught affiliates of blacklisted companies, extended tariff exclusions on China, paused port fees on Chinese ships for two months, delayed adding a company linked to a Chinese spy to the entity list. That's not a slam door. That's a White House keeping its weapons holstered before a Xi summit, while Congress pushes the other way. Price Priced? The Yuan hasn't moved. 667 to the dollar, flat over the past month, near the strong end of its range. If the market thought a real tech war was breaking out, you'd see it there. And the lever I think is underpriced is Beijing's, not Washington's. Rare Earth Licensing. Washington's trade officials pressed press China on rare earth exports at the top. talks. Yet the equity proxy for strategic metals is down about forty percent in thirteen weeks. The market's pricing those minerals as plentiful. My call?

54:37Unidentified speaker (UNKNOWN): The

54:37Unidentified speaker (speaker_0): headline weapons stay suspended through the summit. And the enforced layer keeps tightening anyway. Licenses, designations, the third country rules, pushed along by Congress.

54:50Unidentified speaker (UNKNOWN): two and

54:50Unidentified speaker (speaker_0): three. And when Beijing answers, I think it answers with permission slips of its own. Rare earth licensing, not tariffs. What would prove me wrong?

55:01Unidentified speaker (UNKNOWN): A

55:01Unidentified speaker (speaker_0): summit deal with reciprocal tariff cuts that also suspends the new data center licensing rules. That would mean the executive actually beat Congress, and I don't think it does.

55:13Unidentified speaker (UNKNOWN): So,

55:13Unidentified speaker (speaker_0): oil's priced for a piece the ships don't show, and chips are priced for a truce, the licenses don't show. Both are bets on politics running ahead of behavior. Which brings us home. Because a $90 barrel and freight that's more than doubled don't stay on a map. They show up in factory prices and in the real economy. Let's go to the economy. six 6.3. point three. That's the number Philadelphia handed us on Thursday, and the four forecasters had penciled in twenty. And I want to start there, because it cuts straight across the story we've been telling about an economy running hot. Quick gloss. The Philly Fed Index is a monthly survey of factory bosses in the Philadelphia Feds patch. It asks one blunt thing. Is business getting better or worse? Above zero means more of them say better. Last month it was just under twenty one. The consensus said no change. It came in at six. Still positive, not a collapse, but two thirds of the momentum gone in a month against a forecast of nothing happening. That gap is the news. So why? I think the answer Cancer was hiding in the growth numbers all along.

56:23Third quarter growth printed 3.4% annually. annualized. Lovely. Except final sales, one actually got bought and used, grew about half that. The rest was companies stocking their warehouses, and the National Factory Survey told the same story. At a strong fifty five and a half while actual output was barely growing. That's orders running ahead of production. Why would you order ahead? Go back to the last segment. Freight more than doubled. Oil's been ripping. If you think your inputs are about to cost more, you buy now. And then you stop. Because the warehouse is full. Now the countercase, and it isn't stupid. Philly's a regional survey. It's jumpy. It was north of twenty a month ago and one. bad print from one district has fooled a lot of people into calling a downturn that never came. Fair, I concede that completely. I'm not calling a recession off one

57:24survey from one corner of Pennsylvania. But I'm not reading it alone either. It lines up with the inventory build. It lines up with the shape of the boom. That's what gives it weight. So my call. This is the restocking hangar. hangover. Factory momentum cools into the winter without going negative. I'd put that at better than even. Call it six and ten. For an owner, that means order books at industrial suppliers soften before the earnings do. You'll hear it in the guidance before you see it in the numbers. What proves me wrong? Monday.

57:58Unidentified speaker (UNKNOWN): The

57:58Unidentified speaker (speaker_0): U.S. Flash PMIs land, which are the early read on this month's purchasing manager. manager's surveys. If those fall as hard as Philly did, this isn't a hangover. It's a turn, and I'll say so.

58:10Unidentified speaker (UNKNOWN): Okay,

58:10Unidentified speaker (speaker_0): what isn't cooling is prices, and the surprise there came from Germany. We covered inflation on last week's show. Since then, the new piece of evidence is German producer prices for October, and it was hot. Producer prices are what factories charge each other and charge at the factory gate, before anything reaches a shop shelf. They run ahead of what shoppers pay. Think of them as the weather radar for your cost base. Forecast was 7.2% a year, actual, 8.1%. The month before, 6.7. That's well over a full point of acceleration in a single month, in the largest economy in Europe. And here's the part I'd underline. Euro area wage settlements came out the same year. day, 2.57%. That beat the forecast too. So someone will tell you, aha, wages are pushing prices up. See, that's the part I don't buy. Wages at 2.5%. Consumer prices across the Euro area at 4. Workers are falling behind, not pulling ahead. That's the signature of cost push. Energy and freight shove prices up from underneath. It isn't paychecks pulling them up from the top. Which matters because cost push doesn't end in a wage spiral, it ends in a squeeze, on margins and on shoppers. And the central bankers know they're late. The European Central Bank's deposit rate is 2.5%

59:30against inflation at 4. That's money that costs less than prices are rising by. Lagarde was out talking on Friday, and the mood music across Europe is about having waited too long. The Bank of England already blinked. It hiked to 4% at the start of the month on an 8-1 vote. Why should you care, sitting here? Because we have the same pipe. Factory gate prices in America are running about 6.5% a year. That's the cost base of every business that buys stuff and sells it on. My call on Europe: the factory gate. wave reaches the shops. Euro area consumer inflation holds at four or higher through the winter. I'd put that at two in three. What breaks it? If the next German producer print drops back below September 6.7, the wave was a blip, and I'm wrong. So where does a factory gate price actually land?

1:00:23Unidentified speaker (UNKNOWN): Let

1:00:23Unidentified speaker (speaker_0): me take it all the way down to one business: McDonald's. On their latest call, management confirmed mid-single digit food and paper inflation for 2027. That's the producer price story arriving at the fryer, but who pays it?

1:00:38Unidentified speaker (UNKNOWN): This

1:00:38Unidentified speaker (speaker_0): is the clever bit of how McDonald's is built. The company owns or controls the corner, signs 20-year agreements, and takes a percentage of sales. That cut comes off the top before the operator pays for labor for food, or for debt. So the beef and the paper cups? That's the franchisee's problem. McDonald's itself runs an operating margin around 47%. The operators in this industry live on 12 to 15. And I know exactly what you're thinking. Then McDonald's is fine, right? Costs go up, the franchisee eats it. McDonald's keeps collecting. Fair. In year one, mostly true. but the cost still leaks back. The operator's only defense is raising menu prices, and the person on the other side of that counter is the most miserable consumer we've measured in a long time. Michigan Sentiment, the University of Michigan's monthly read on how households feel about their finances, came in at fifty-one and a half on Friday, up from forty-eight 48 and in line with the forecast.

1:01:37Unidentified speaker (UNKNOWN): Still

1:01:38Unidentified speaker (speaker_0): in the bottom, percentile of its history, and McDonald's U.S. guess counts were already negative. Comparable sales slightly down. The finance chief wouldn't promise the fourth quarter gets better. So the operator raises prices, fewer people come in. McDonald's takes its percentage of a smaller number.

1:01:55Unidentified speaker (UNKNOWN): Then

1:01:56Unidentified speaker (speaker_0): the rates piece. The company added a brand new risk factor about franchisee financing constraints. It's already cut a 100 openings from this year and pushed its 50,000 store target out to 2028. When borrowing costs stay this high, operators stop building.

1:02:12Unidentified speaker (UNKNOWN): So

1:02:12Unidentified speaker (speaker_0): my read: for McDonald's, this is a slow burn, not a shock. It plays out across 2027, through fewer visits and slower openings. Earnings. It doesn't come through the cost line. For the operators, it's a genuine squeeze, and it's already started. I'd change my mind if guest counts turn positive while those menu prices are going up. That would be pricing power, and I don't see it yet. Now, a fair question from the skeptics chair. Yields fell this week. The 10 year treasury slipped back to about 5%, down from the 5.15 it touched a week ago. Isn't the pressure easing? And I have to be honest here. The desk had a standing call for the 10 year above five and a quarter. This week it went the other way. Not met. But follow it through to a real person. Last week we covered housing. Since then, the 30 year mortgage rate has gone up again, another seventeen hundredths of a point, to seven point two eight percent. Yields down, mortgages up. How? Mortgage rates move after treasuries, and the gap between them widens. when lenders get nervous. Meanwhile, the real yield, the 10-year after stripping out expected inflation, sits near its highest in 20 years. One soft week in the headline rate didn't touch the cost of money people actually borrow at. And households are thin.

1:03:30Unidentified speaker (UNKNOWN): The

1:03:31Unidentified speaker (speaker_0): saving rate is down to 3.8%. That's the cushion, and it's mostly gone. Abroad, the shoppers split. UK retail sales beat in October, up seven-tenths of a percent against a forecast of flat. Canada's early October estimate fell 1.5% when the forecast was a rise. No single consumer story out there, just stretched people, some still spending, some stopping. My call. Recall, Mortgage rates stay above 7% seven percent through the winter, 3 three and 4. four. What turns it is the ten year below four and a half, with producer prices cooling under four. Nothing this week points that way.

1:04:10Unidentified speaker (UNKNOWN): So

1:04:10Unidentified speaker (speaker_0): carry this out of the car. Factories are coming off a stocking binge, costs are still climbing underneath, and the people at the till are stretched thin, which is a hard world for any business whose value sits mostly in profits a decade away. Real yields near a twenty percent. 20 year high make tomorrow's dollar worth less today. And that's exactly the test for this week's deep dive. Microsoft. Microsoft. We've had it on the show several times this autumn, and it came up earlier tonight. But since then, the quarterly filing has been taken apart line by line, and the company has offered voluntary buyouts to 60,000 people. That's roughly a quarter of the workforce. A quarter of the company offered the door. Departures by the thirty first of December. And if you've just joined us, you're listening to Sovereign Weekly, and this is the deep dive. I'll make you one promise up front. By the end of this, I'll give you a single number. One. It tells you whether you're looking at a software company or a

1:05:11utility. Keep that in your pocket. So start with what it actually is. Microsoft is two businesses wearing one coat. The first is the productivity franchise, office, teams, the identity system that decides who in a company can log into what. It's rent. A business pays per seat, per person, every year. And the rent per seat has been compounding at about 11% a year. The commercial cloud version grew 17%. last quarter on flat sales and marketing spend. Flat. They're not even pushing harder to sell it. The second business is Azure. That's renting out computing power, data centers by the hour. It grew 48% to about $33 billion in the quarter. Azure alone delivered roughly three-quarters of the company's entire revenue increase. So one half is an annuity. the other half is a rocket. Simple, right? It's not simple. The rocket burns a lot of fuel, and that's the whole segment. Let's do what it really earns for its owners. We use something called owner earnings. Think of it as the cash you could pull out of the business after paying for everything. And I mean everything. That includes the stock handed to staff. and the spending needed just to keep the machine as good as it is

1:06:30today. Trailing twelve months, that's about one hundred and thirty nine billion dollars. Which sounds enormous. It is enormous. But at today's price, it's a yield of about three and a half percent. A ten-year treasury pays about five. 5%. And that's the hook from the last segment, isn't it? With real yields. near a 20 year high, you're accepting less than the risk free rate today. You're paying entirely for growth that hasn't happened yet.

1:06:59Unidentified speaker (UNKNOWN): Then

1:06:59Unidentified speaker (speaker_0): there's the part I find genuinely uncomfortable. In fiscal 2026, owner earnings jumped about 45% on revenue up 18%. Okay, wait. Earnings up 45% on sales up 18? In a year they spent more on buildings and chips than ever? Right. So how? Capital spending rose by about $51 billion. The depreciation charge, which is the accounting cost of wearing that stuff out, rose by only about 4.5. And this year, Microsoft stretched the assumed Consumed life of its buildings from 15 years to 25. Stretch the life and the yearly cost. shrinks on paper. We add back about $3.7 billion a year to undo it. And that's not the only channel. Money spent to help suppliers buy server components gets booked as other investing, not CapEx. There's a $35 billion receivable sitting there now. There's $26.5 billion of equipment still unpaid in payables. And leave leased data centers skip the CapEx line entirely. Four quiet ways cost moves out of earnings. None of it is fraud. The earnings are cash backed, the accruals look clean. But every one of those choices flatters this year at the expense of later

1:08:19years. Which brings me to the measure that matters most return on tangible capital. In plain terms, it's how much profit. you get for every dollar of physical stuff the business needs to run. The buildings, the servers, the kit. In fiscal 2021, that return was about 197%. Nearly $2 a year back on every dollar of kit. That's what software looks like. Fiscal 2026? twenty twenty six, 38. thirty eight. 38. thirty eight. Still good by. the way. Most companies would kill for it. But the tangible capital base went from about thirty billion to about three hundred twenty billion. Tenfold. More, actually. Over the last five years, it's taken about five dollars thirty of new capital to grow owner earnings by one dollar. Over nine years, about four hundred sixty six. So the growth is getting dearer, not cheaper. And I know exactly what you're thinking. Azure is growing almost fifty percent. Microsoft Cloud cleared sixty three billion in a quarter, up thirty. Copilot daily users are up fivefold on the year.

1:09:28Unidentified speaker (UNKNOWN): If

1:09:28Unidentified speaker (speaker_0): the growth's that good, why am I grumbling about return on capital? Fair, because growth you have to buy is a different animal. Property additions were about forty three billion. percent of revenue last quarter. Free cash flow has now fallen for three straight years, down about twenty three percent most recently. The revenue line is accelerating while the cash line goes backwards. That's not a contradiction once you see it. The cash is being converted into concrete. Now the bull case. It isn't stupid, so let me give it its full weight.

1:10:01Unidentified speaker (UNKNOWN): Amy

1:10:02Unidentified speaker (speaker_0): Hood's Hood, the CFO, says most of this spend is short-lived chips, GPUs, and CPUs, matched to contracted demand.

1:10:11Unidentified speaker (UNKNOWN): And

1:10:11Unidentified speaker (speaker_0): there's a commercial backlog of about $677 billion. On top of that, the return on all capital, goodwill included, is still about 27%. That's way above a 10% hurdle. The cloud gross margin actually. rose a point to 66%. Their own Maya chip claims 30% better cost per token. And swapping small models into PowerPoint cut GPU cost by about 84%.

1:10:39Unidentified speaker (UNKNOWN): So

1:10:39Unidentified speaker (speaker_0): maybe once supply catches up, this fleet earns cloud-like returns. Then that depreciation really is maintenance, and owner earnings compound faster than anybody's modeling. And there's a capital access argument on top. Rivals are borrowing for this build-out. Alphabet's debt went from about 16 billion to about 100. Microsoft is funding a $175 billion annual program out of its own pocket and still producing positive free cash flow. I concede most of that. The incremental returns are well above the hurdle. The balance sheet is a weapon. Over the next two years, they owe roughly $174 billion. And they've got about $209 to cover it. Nobody's in trouble here. See, the part I don't buy is the category. A software franchise earning triple digit returns on a $30 billion asset base is turning into an infrastructure owner, and mature infrastructure earns low teams. teens, not 30s, and nowhere near 197. The bull case needs this one to be the exception. And the constraint isn't even chips anymore, it's electrons. Microsoft walked away as anchor customer on a data center in Lauten, in the UK, because the grid can't connect it until the 2030s.

1:11:59They've pledged to cover the power bills for their American data centers. so local ratepayers don't eat them. And around $25 billion of this calendar year's CapEx is just component prices going up. The filing literally says very few qualified suppliers. So they're financing their own supply chain to stay at the front of the queue. Exactly that. Okay, the moat. And I'll be honest, our view moved. We used to call it Y. wide and strengthening, full stop. Now it's wide and stable in productivity, and narrower and contested in infrastructure. So what is the mechanism specifically?

1:12:36Unidentified speaker (UNKNOWN): It

1:12:37Unidentified speaker (speaker_0): isn't the AI model. Microsoft is deliberately making models swappable. Their security pipeline sends 90% of tasks risks to a small in-house in house model and 10% to a frontier one and matches the big model at half the price. cost. The moat is the harness around the model, identity, memory, workflow. Look at the pieces. GitHub has 225 million users and one in three pull requests now involves an agent. The repository is the toll road. Agent 365 registered 40 million agents in two months, bundled into the top enterprise tier alongside the identity product. And there are 17 exabytes of workplace context, growing about 35% a year. This week fits the pattern. At Ignite, they launched co-pilot workspaces and office agents, though with no pricing yet. They also put a third party's open model inference inside Foundry, billed Build against customers' Azure commitments. Somebody else's models, Microsoft's meter. And internal data reportedly shows a Rival Labs assistant stalling against Copilot inside Office. Context beats cleverness, at least for now.

1:13:48Unidentified speaker (UNKNOWN): So

1:13:49Unidentified speaker (speaker_0): what breaches it? Open protocols. Microsoft is exposing 650,000 of these plug-in actions in Dynamics itself. Those are standard ways for any agent to reach into a company's data. The standard that lets Microsoft demote everyone else's apps to data sources is the same standard that lets someone demote Microsoft. We give the context mote three to five years, not forever. And the first crack in the seat model just showed. Paid commercial seats grew 5% down from 6. It's the first acceleration. deceleration. Windows revenue from PC makers fell 22%. Small numbers. small numbers.

1:14:28Unidentified speaker (UNKNOWN): But

1:14:28Unidentified speaker (speaker_0): the seat is the unit of the whole franchise. On infrastructure, the share picture is uncomfortable. Amazon, which we've covered this autumn, put out an update since then.

1:14:39Unidentified speaker (UNKNOWN): On

1:14:39Unidentified speaker (speaker_0): its October 29th call, AWS growth was 39% on a run rate of about $184 billion. In absolute dollars, it's still adding more than Azure year on year. roughly $13 billion against 11 billion. And Andy Jassy pitched the exact same swapped the model's harness, then added the knife. Bedrock mostly runs on Amazon's own training chips, and anyone on merchant hardware carries a comparable cost structure. That's aimed squarely at Redmond. It is. Our breach odds: Google winning on price at the coating coding and agent layer about 35%. thirty five percent. OpenAI going direct on non Azure compute thirty.

1:15:19Unidentified speaker (UNKNOWN): The

1:15:19Unidentified speaker (speaker_0): seat ceasing to be the unit of value twenty.

1:15:22Unidentified speaker (UNKNOWN): And

1:15:22Unidentified speaker (speaker_0): that backlog, the six hundred and seventy seven billion,

1:15:25Unidentified speaker (UNKNOWN): it

1:15:26Unidentified speaker (speaker_0): came in flat against six hundred seventy eight the quarter before. Oracle, which we covered a couple of weeks back, signed six hundred and sixty-four $64 billion on a tenth of the revenue base. And since then, Microsoft has reportedly stopped renewing. outsource capacity deals with Oracle. A backlog that size isn't a moat, it's a capacity cycle. Which brings us to OpenAI. As we discussed earlier in the show, the relationship is circular, and the numbers this quarter make the circle visible. Microsoft has funded $12 billion of OpenAI's capital, and in the same quarter, it booked $10 billion of revenue from OpenAI. That's about 11% of the company. The amount OpenAI owes Microsoft rose by more than a billion in three months, to about seven. As of January, OpenAI was reportedly about 45% of Azure's backlog. So you lend the customer the money, the customer pays you rent with it, and you book it as growth. That's harsher than I'd put it, but it's not wrong. The defense is real though. Microsoft holds OpenAI's IP rights to 2032. If it goes sour, the worst case is a supplier dispute, not a lost crown jewel. The risk is timing. There are $365 billion of data center leases signed but not yet started, running out to 2034. That's five times the least debt on the balance sheet today. If lab funding tightens, the revenue stops before those commitments do.

1:16:45Management.

1:16:46Unidentified speaker (UNKNOWN): And

1:16:46Unidentified speaker (speaker_0): here the dossier has a real verdict, so let me give it. Nadella Nodella since 2014. twenty fourteen, Hood since twenty thirteen. Over ten years, they turned every retained dollar into nearly eight dollars of market value.

1:16:59Unidentified speaker (UNKNOWN): The

1:17:00Unidentified speaker (speaker_0): buybacks from twenty seventeen to twenty twenty one created about one hundred and forty billion. That's elite. Since twenty twenty four, the buybacks have turned into a treadmill. They're mostly mopping up shares issued to staff at prices above what we're doing. we think the business is worth. Activision earns below a 10% hurdle, and it's the weakest big call of their era. And executive pay is tied to revenue and operating income, not to the return on the hundred and seventy five billion they deploy every year. So they're paid to build, and they're building. Which is why the buyout framing caught my ear. Nadella says it's about agency, not A. AI replacing people. Small teams, 11 developers building agents. For the first time ever, engineers are being offered voluntary exits. But operating expenses are still rising. My read is that this isn't a margin rescue yet, it's the cost structure trying to catch up with the capital structure. And there's no named successor. Judson Altoff, who runs the commercial business, is the visible bench. If Nadella and Hood both left mid cycle, the new team inherits those lease commitments and a pay plan that rewards scale. One more from this week, because it touches the moat in trust. Microsoft traced the November 12th global outage to a

1:18:20bug dating back to July, triggered by a traffic shift from Teams and a 14% loss of server capacity. That replaced their earlier explanation. Then a hot patch broke app installs. Two faults in a week. There's a class action in Canada, and MIT is suing over data center patents.

1:18:39Unidentified speaker (UNKNOWN): A

1:18:39Unidentified speaker (speaker_0): federal review board already called the security culture inadequate back in 2024. four.

1:18:45Unidentified speaker (UNKNOWN): On

1:18:46Unidentified speaker (speaker_0): the other side, it's reportedly on a White House national champions list for federal financing and export licenses. Mixed bag. The outages don't move the valuation. They do chip at the one thing an identity moat can't afford to lose. Right. Valuation. The base case says the price sits about 7% above our value. No margin of safety at all. For a business this predictable, with this little debt, we'd want the price to sit about a fifth below our value. From here, that's a fall of roughly a quarter. And the spread of cases is wide. On our conservative case, the price sits about 64% above value. On the optimistic one, it sits about 29% below. The base case lands near the top of the honest range, not the middle.

1:19:34Unidentified speaker (UNKNOWN): What

1:19:34Unidentified speaker (speaker_0): does the current price assume? Owner earnings compounding about 11.5%. a year for 10 years. That's demanding, but just under the 13.7% revenue growth of the last four years. So it's not crazy, it's simply full. Bought here and held five years, converging to value, the base case gives you about eight and a half percent a year against a ten 10% percent hurdle. And with zero growth, the business is worth only about thirty-seven 37% percent. of today's price. So about 63% of the price is a bet on compounding. The short sellers version, valuing it on reported free cash flow after stock pay, has the price at well over double what it's worth. I think that's too harsh. It treats all the growth CapEx as permanent cost. And I'll rib our own work a bit here. Our earlier edition had the price about 80% above. value. Since then, we changed the method. We stopped treating every Capex dollar as maintenance and started using honest depreciation. The method change explains more than the entire rise in

1:20:39our number. The company's own contribution was real, with operating income up 20%, but most of the gap closing was us getting fairer, not Microsoft getting cheaper. So we were the expensive ones. Fair enough. So, the call.

1:20:55Unidentified speaker (UNKNOWN): I

1:20:55Unidentified speaker (speaker_0): think the productivity franchise holds through this cycle. Seats slow, but the rent per seat and the identity layer keep compounding. I'd put that at about three and four. The infrastructure half is where I'm committing to the less comfortable view. I think return on tangible capital keeps sliding over the next year rather than rebounding. Call it two and three.

1:21:17Unidentified speaker (UNKNOWN): The

1:21:17Unidentified speaker (speaker_0): leases arriving through twenty thirty four and the component inflation mean the asset base keeps outrunning the profit. For someone who owns this business, that means owner earnings growth that's real but slower per dollar invested. A wonderful franchise is now carrying a utility-sized capital budget, and at this price you're paid for none of that. risk. What proves me wrong? One event above all. At the late January call, management guides calendar 2027 CapEx flat while Azure keeps growing above 35%. That would show the short-lived spend can slow and free cash flow turns up. I'd give it maybe one in three. If it happens, I'm wrong on the trajectory and the bull case. gets a lot stronger. And the hard kill criteria, the things that would break the thesis outright. First, cloud gross margin below 64% for two quarters while CapEx is guided up. Slide into the 50s and it's broken. Second, the backlog flat or down for a second straight quarter, while the OpenAI receivable grows faster than OpenAI revenue. Third, commercial seat growth below 3%, or that commercial cloud

1:22:31revenue growth dropping under 12%. Or Azure growth falling below 20% while annual CapEx stays above $150 billion. Quick technical beat, and its positioning context, not a trade. Over the last In two months, the stock has led its sector by more than 5% and tightened into a narrowing range. In the most recent session, it fell about 3% while the industry fell more than 5%. It's below its short-term averages, about 13% under its high, and volume is leaning to the downside. So it's relatively strong in a weak group, with no cushion underneath. And that one number I promised: return on tangible capital. 197% five years ago, 38% now. If it settles in the 30s while CapEx flattens, you've got a software company with a very large landlord arm. If it keeps sliding toward the low teens, you've got a utility with the best software franchise in the world bolted on. And you'd want to

1:23:31pay a utilities price for that.

1:23:33Unidentified speaker (UNKNOWN): So

1:23:33Unidentified speaker (speaker_0): let me pull the whole hour together, because these threads really do rhyme. All week, the money has been leaving the weightless parts of the economy and pouring into heavy things chips, memory, power lines, steel racks. The component makers we talked about earlier are getting paid first.

1:23:51Unidentified speaker (UNKNOWN): The

1:23:51Unidentified speaker (speaker_0): factories are coming off a binge, costs are climbing underneath, and the consumer is stretched. And a five 5% percent treasury means capital is no longer free. In that world, the question for anyone. any company, and Microsoft is just the starkest example, isn't how fast it's growing, it's what the next dollar earns. The best business of the last decade is now answering that question in concrete, and the price assumes the answer comes back good. That's the thing I'd carry out of the car. Growth was the story of the last cycle. The return on the next dollar is the story of this one. Before I let you go. everything we've discussed today is for education and information only. It isn't a recommendation and it isn't personalized advice. And both the host and the platform may analyze and may hold views on the securities discussed today. Thank you for spending the hour with me. Genuinely. Liam signing off. Back next week.

This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights