The Paper-Gain Ledger: Alphabet's $98B SpaceX Mark and Amazon's $53B Anthropic Mark — Now Facing IPO Price Discovery

The Paper-Gain Ledger: Alphabet's $98B SpaceX Mark and Amazon's $53B Anthropic Mark — Now Facing IPO Price Discovery

A fault line in the Mag-7 divergence story: the basket's headline profits are increasingly not operating profits. Alphabet's June-2026 net income of $112.19B included a ~$98B unrealized gain on its SpaceX stake; Amazon's net income of $62.65B included a ~$53B unrealized gain on its Anthropic stake.¹ Both marks are unrealized, reversible, and about to meet a public market test.

The Anthropic IPO now has a named, quantified bear case

Anthropic is barreling toward a Nasdaq listing at a reported ~$2T target valuation, and the valuation fight went public this week. New Constructs (David Trainer) called it "the most ridiculous IPO of 2026" and valued the company at $150 billion — a 92% discount to the target. From CNBC (Oct 7):

"While Anthropic offers more to society than WeWork ever did, at a $2 trillion valuation, its IPO presents far bigger risks and is positioned to be a far bigger rip off of U.S. capital markets."

The firm's math: to reach $2T, Anthropic would need to record "double the trailing year of profit for Nvidia," whose net income over the past four quarters topped $190B — while Anthropic's 2025 revenue was $4.6B against a net loss of $42B, per the leaked IPO prospectus (Reuters, via CNBC). The firm's blunt conclusion: "We don't think Anthropic has a viable business," citing mounting operating losses and open-source model competition.

The revenue quality behind the mark is also method-dependent. Reuters (Oct 8): "Anthropic pays cloud partners about 16% of every dollar earned through them, which last year accounted for half of revenue, a Reuters analysis showed" — i.e., roughly half of Anthropic's reported revenue runs through AWS and Google Cloud at a grossed-up accounting treatment (see The AI-Slowdown Debate Is Splitting the Basket: Warsh Fed Hikes Into AI Capex, Dispersion Widens for the OpenAI ARR gross-up episode that exposed this). And the "operating profit" claim is soft: Anthropic "has reportedly told investors it realized an operating profit in the second quarter, per the Financial Times, but those figures have excluded stock-based compensation" and "largely preceded an array of computing deals signed by the company that are set to cost hundreds of billions of dollars" (The Information counts $517B of compute deals in 11 months). Axios's verdict: "Beware of claims about 'profit' as well, at least until they come with the financial documents typical in an earnings release."

What it means for the basket

Amazon's ~$53B mark was struck at the $965B May funding round; a $2T IPO print roughly doubles the implied stake value, while a New-Constructs-style outcome ($150B) implies one of the largest single markdowns in corporate history flowing through AMZN's other-income line — on top of a quarter where Amazon's FCF was already -$8.82B (see Capital Allocation and FCF Divergence: Free Cash Flow Squeeze vs. Shareholder Returns — and Alphabet Now Selling Equity). Alphabet holds Anthropic too and carries the larger, SpaceX-side exposure (watch: any SPCX mark-down). The IPO is the first time these private marks get continuous public price discovery — the exact mechanism that either validates or deflates the basket's headline EPS. Meanwhile OpenAI's parallel repricing (a $30B raise at $1.4T, ~28x its corrected ~$50B ARR, with Polymarket's ≥$1.5T-IPO odds down from ~80% to 58%) shows the private AI marks are already moving. Anthropic's annualized revenue crossed $65B in July and is "set to reach $100 billion by year-end" (Reuters) — the growth is real; the question the IPO answers is what multiple of it, net of the partner-channel gross-up and the $517B compute commitments, public investors will pay.