My notes got stuck in the pipe for about a week. Rather than dump a week of issues on you at once, here are the six reads from that stretch that still matter. The common thread: the AI trade finally got a real stress test, and the interesting damage showed up in credit, not in the index.
The Comprehensive Guide to the Bank of Nvidia
Hunterbrook went through Nvidia’s latest 10-Q and added up every disclosed commitment and guarantee: $530.5 billion. That includes $279B in supplier purchase commitments (up 134% in a quarter), $108.5B in rent guarantees (mostly a $105B backstop tied to SB Energy and OpenAI), and $99B in equity stakes across 242 companies.
How I read it: No single loop scares me. The correlation does. Chip revenue, equity marks, and rent guarantees all depend on the same thing: continued access to cheap risk capital. Steelman: a company expected to generate roughly $500B of free cash flow through FY2028 can carry a lot of guarantees. My guess (low-to-moderate confidence): the first crack shows up in the IPO pipeline, not in Nvidia’s earnings. SB Energy delayed, Nscale delayed, Firmus pulled.
Nvidia: Now What?
After Jensen Huang resigned amid a scandal, Nvidia fell nearly 14% in a week and more than 30% from its highs, close to $2 trillion in market cap. The S&P 500 barely moved. Nvidia now trades at nearly a 40% discount to the market while earnings estimates keep rising.
How I read it: That’s awkward for “if NVDA sneezes, the index catches pneumonia.” Either breadth underneath was real, or the market had already demoted Nvidia from load-bearing beam to large tenant. It doesn’t disprove the bigger point that AI capex drives the economy. A stock repricing is not the buildout slowing. No earnings model priced in a CEO scandal, and that governance tail was sitting inside what everyone treated as a momentum story. Ben Carlson’s closing checklist is the useful part: what’s your edge, when do you sell, and when do you admit you were wrong?
Treasuries lose their haven status
Joachim Klement summarizes a Banque de France study ranking assets by how they behave during spikes in risk aversion. The dollar still ranks first. Treasuries have slipped below Swedish and Norwegian government bonds, and haven status broadly tracks low national debt. Gold qualifies but ranks below most sovereigns. Bitcoin, oil, and silver show no haven traits at all.
How I read it: This separates monetary credibility from fiscal credibility. If it holds, haven demand is moving from “deepest market” toward “cleanest issuer.” Steelman: in a real crisis only Treasuries can absorb trillions of dollars of flight. True, but the study measures behavior, not capacity, and behavior usually changes first.
A Layman Thinks About AI
Howard Marks’s new Oaktree memo works through eleven questions on AI. The line I’d keep:
The direction is fairly predictable, maybe a six or seven… The things investors get paid for are much less predictable, around a 2: how fast it happens, how big the profits are, and who keeps them.
How I read it: Most AI bull cases are arguments about direction, and they’re usually right. Direction isn’t what you’re paid for. The credit section was the newest material for me: by my notes, roughly $600B of AI-related debt year-to-date, about 12% of the USD investment-grade index, with hyperscalers acting as rate-insensitive borrowers. That moves the AI question into the bond market, where most equity holders aren’t looking.
The AI Trade Deconstructed
Aswath Damodaran splits “the AI trade” into suppliers, builders (hyperscalers), product companies, and customers. AI stocks were 37% of US market cap going into November, and on November 5 they lost almost $1.8 trillion in a day. Chips took the damage while hyperscalers gained over the week.
How I read it: Supplier revenue is the mirror image of builder capex. The same spending that drains a hyperscaler’s free cash flow is a chipmaker’s earnings. That’s why picks-and-shovels led on the way up, and why they crack first when shareholders start rewarding builders for restraint. His practical advice is worth stealing: sell a slice into each rally, and if you’re buying the correction, set a valuation trigger and automate it instead of waiting for the bottom.
Weekly Wrap-Up: Nflation, Nvidia, ‘n the AI Bubble
Jared Bernstein says CPI and the estimated PCE deflator are accelerating and a quarter-point hike in December is near-certain. That’s his read, not a decision yet. If he’s right, the Fed is moving against the affordability push, with the 30-year mortgage around 7.5%.
The point I’d keep:
The equity premium rarely moves by itself. It shows up in wider credit spreads, tighter lending standards, higher capital costs.
How I read it: That ties the whole week together. Even if politics holds the policy rate down, an AI-led repricing of risk can tighten financial conditions by itself, through spreads.
What I’m watching
- AI-adjacent IPOs. More delays among the counterparties Nvidia is backstopping would be the clearest early signal.
- Investment-grade spreads on hyperscaler debt. If Marks and Bernstein are right, the AI repricing shows up here before it shows up in earnings.
- The December Fed meeting. A hike into a softening AI trade would test how much of the market’s breadth is real.
Back to the regular rhythm from here.