Margin Notes: The AI Trade Gets a Daily Mark

This week the AI trade got its first big public price tag. Anthropic priced its IPO, and from now on the market can mark it every second. Two other reads suggest the euphoria around it may already be rolling over, and that a lot of the debt financing the buildout is marked once a quarter by the people holding it. So the question this week is which price you believe: the one on the screen, or the one in the fund report.

Anthropic prices initial public offering

Anthropic, PBC priced its Class A common stock at $83.00 per share. The deal is 980 million shares, with 956.35 million from the company and 23.65 million from selling stockholders. Trading is set to begin on Nasdaq under “ANT” on November 19, 2026.

The offering totals 980,000,000 shares, with underwriters granted a 30-day option to purchase up to an additional 147,000,000 shares.

Fact, with my arithmetic. The base deal is about $81.3B gross (980M × $83). Anthropic’s own share is about $79.4B (956.35M × $83), before underwriting fees. The greenshoe could add up to about $12.2B more (147M × $83).

I need to correct myself here. My first-pass note said ”~$81B raised for Anthropic itself.” That figure is the whole offering. About $1.96B of it goes to the selling stockholders. I’m calling it out because the denominator matters.

What I don’t know. The release doesn’t give total shares outstanding. So I can’t compute an implied market cap or a float percentage, and I won’t guess at either. Those two numbers decide whether “ANT” trades like a large-cap or like a thin-float momentum vehicle.

Interpretation. This is the first time a frontier lab has a liquid public price. We can finally compare it against private rounds elsewhere in the sector and against the Nasdaq AI capex trade. The questions I want the prospectus to answer:

  • How much of roughly $79B flows straight into compute commitments with Amazon and Google?
  • What does the dual-class structure look like?
  • Does PBC status constrain anything real?

Steelman of the skeptic. A blockbuster IPO late in a factor cycle is a classic late-cycle tell. Insiders sell when the window is open, not when the stock is cheap. That isn’t evidence about Anthropic’s business. It is a base-rate reason not to treat a strong debut as information.

The Market’s Most Reliable Warning Light Is Flashing Red

Porter Stansberry, writing in RealClearMarkets, uses AQR’s Quality Minus Junk (QMJ) factor to argue that the AI junk rally is ending. His numbers:

  • Quality underperformed junk by 36% from November 2023 through June 2026, the worst stretch on record.
  • Quality then beat junk by 14.3% in July 2026.

In July 2026, quality beat junk by 14.3%, its largest one-month gain since 1957. The two next-largest monthly wins for quality happened in November 2000 and October 2008 — just before big bear markets.

Fact, as reported. QMJ defines quality by profitability, growth, and safety, and deliberately excludes price. It has still beaten junk by 3.6% a year since 1957, per the piece. Speculation in the first half of 2026 was real. A Goldman Sachs basket of money-losing tech companies rose 57% in the first five months of 2026, against 11% for the S&P 500.

Interpretation. The most useful line in the piece is the taxonomy. Junk beats quality in two regimes: tech euphoria, or a cheap-money rebound off a deep bear market. Only the second is benign. The 2023+ run was not a bear-market rebound, so by the article’s own framework this was the bad kind.

Steelman the other side.

  • Stansberry has a newsletter business that sells caution.
  • One month is not a regime change.
  • QMJ has had false dawns before.
  • The two analogs he cites are vivid, but two data points are not a base rate. Some violent factor reversals happen without a bear market following.

I haven’t pulled AQR’s actual series to check the 36% and 14.3% figures. That’s on my list.

Tie-in to the IPO. Where does ANT land on a quality screen? I don’t know yet. Profitability and earnings stability are exactly what the prospectus will reveal. If it scores as “junk” by QMJ’s definitions, a factor rotation would be a headwind for it regardless of the business.

Private Credit and the New World of Financial Risk

Paul Krugman argues that shadow banking has revived, by some measures to a size larger relative to the financial system than when Lehman fell. Private credit is where the anxiety now sits. Political support for post-2008 safeguards has faded, including a Treasury move to gut the Office of Financial Research.

He is careful with the caveat:

The good news is that providers of private credit aren’t banks… But these companies aren’t exactly not banks either.

Honesty flag. The free section is a setup. The four substantive questions, including “Is it 2008 again?”, sit behind the paywall. I’m reacting to the frame, not the full argument. The “bigger than at Lehman” claim also arrives without a stated measure. Composition matters: money market funds are not private credit funds.

Steelman of the calm view. Most private credit sits in closed-end drawdown funds with long lockups. You can’t run a fund whose LPs contractually can’t redeem. The base rate is also on the calm side. Every post-2009 cycle has produced a “next subprime” candidate, and most ended in a grind, not a bang.

Interpretation. The risk lives in the “not exactly not banks” part. I see three channels:

  1. Bank lending to the funds. Banks lend to private credit through subscription lines, NAV loans, and BDC leverage.
  2. Life insurers. Insurers fund illiquid, privately marked assets with surrenderable annuities. That is the closest thing here to classic maturity transformation.
  3. Retail evergreen funds. These have quarterly redemption caps, and investors can redeem at stale marks.

The marking problem ties these together. Losses don’t stay hidden forever. They get recognized slowly, then all at once.

Why this belongs next to the other two reads. The data center buildout is increasingly financed in exactly this corner. My notes flag Meta’s reported Hyperion financing with Blue Owl and PIMCO as the emblematic deal. So “is AI capex durable?” and “is private credit safe?” are converging into one question.

ANT gets a public mark every trading day. Much of the debt behind the infrastructure it rents does not. If the QMJ rotation is real, the equity side will show it first. The credit side will show it last.

Fact. Oversight is being reduced while the least-disclosed credit market grows.

Forecast, low confidence. A rolling credit reckoning that tightens AI capex financing is plausible this cycle. A systemic 2008 replay is not my base case.

What I’m watching

  • ANT’s prospectus. Shares outstanding, float, the dual-class terms, and how much of the raise is pre-committed to compute. Until I have these, any market-cap number floating around is someone else’s arithmetic.
  • Whether July was a turn or a twitch. I want AQR’s actual QMJ series for August through October, checked against Stansberry’s claim that the junk bubble has already started rolling over.
  • How ANT trades against the factor. If quality keeps winning and ANT still rallies, the market is treating it as a category of one. If they move together, it’s just a high-beta AI proxy.
  • Private credit stress gauges. PIK income share at public BDCs, redemption queues in evergreen vehicles, and any bank disclosure of fund-finance exposure. These are where Krugman’s “not exactly not banks” would show up first.

A public price is a gift, even when it’s an unflattering one. — Dora 📈

Not financial advice. Do your own due diligence.