Unrealized equity-stake marks break the headline P/E on megacaps

⚠️ FOUR CONFIRMED INSTANCES IN A SINGLE DAY (2026-08-04)

This is not an anomaly to check for occasionally. In August 2026 it is the default state of any company holding a private AI stake, and it spans market caps from $122B to $4.6T. CNBC ran the cross-market story on 2026-08-03. Check every name before quoting a P/E.

Ticker Headline P/E Real (ex-marks) Non-operating income Stake

GOOGL 18.9× ~37.6× $148.8B ttm ($97.8B in one quarter) SpaceX, Anthropic

AMZN 22.2× ~40.6× ~$52.6B ttm ($53.4B in Q2 alone) Anthropic, OpenAI

NOW 72.0× n/m — mark took GAAP net income from ~break-even to $298M $273M in Q2 private AI stakes

AMAT — ~26% of "record" EPS was non-operating $685M 9% of BESI

Ticker	Headline P/E	Real (ex-marks)	Non-operating income	Stake
GOOGL	18.9×	~37.6×	$148.8B ttm ($97.8B in one quarter)	SpaceX, Anthropic
AMZN	22.2×	~40.6×	~$52.6B ttm ($53.4B in Q2 alone)	Anthropic, OpenAI
NOW	72.0×	n/m — mark took GAAP net income from ~break-even to $298M	$273M in Q2	private AI stakes
AMAT	—	~26% of "record" EPS was non-operating	$685M	9% of BESI

Claim

ASU 2016-01 requires equity securities with readily determinable fair values to be marked to market through net income, not through OCI. A privately held stake that IPOs, or a private company that raises at a higher valuation, therefore lands in GAAP earnings as a large non-operating gain. Screeners divide price by that inflated EPS and display a multiple that is roughly half the real one.

These gains are non-cash, non-recurring, frequently unmonetizable (lockups), and they reverse. They belong in book value, not in the earnings multiple.

The one-second diagnostic

If reported net margin > operating margin, stop and find the non-operating line.

Operations cannot produce a net margin above the operating margin. Any company showing this is being flattered by something below the operating line — equity marks, a legal settlement, a tax item, or a divestiture.

Evidence — Alphabet, ttm to 2026-06-30


ttm revenue	$445.88B
ttm operating income	$147.63B → 33.1% operating margin
ttm reported net income	$244.20B → 54.8% net margin ← impossible from operations
ttm non-operating "other income"	$148.80B — 50% of pretax income
— of which the Jun-2026 quarter alone	$97.83B

Cause: SpaceX IPO'd 2026-06-12, forcing Alphabet's cost-method stake (~$900M invested in 2015) to be reclassified to fair value at $94.0B; and Anthropic revalued from $350B to $965B. The Q2 gain was $99.0B pre-tax / +$6.26 of the $9.11 quarterly EPS.

Basis	ttm EPS	P/E at $377.65
Headline GAAP (every screener)	$19.91	19.0×
Core, ex non-operating	~$10.04	~37.6×

The displayed multiple was wrong by a factor of two. Three independent methods (explicit quarterly reconstruction, Yahoo's own "Normalized Income" ttm, and the company's disclosed per-quarter adjustment) converge on 36–38×.

It reverses — and that is the second-order trap

A portfolio-specific passage was removed from the public build.

Second confirmed instance

A portfolio-specific passage was removed from the public build.

How to apply

Compare net margin vs operating margin on every name before quoting a P/E.
When they invert, rebuild EPS from operating income + net interest, taxed at the effective rate. Never use the headline.
State both multiples in the report and say which one the verdict uses.
Expect this specifically in companies holding large private AI stakes — the 2025–26 wave of private AI valuations is running through megacap income statements.
Treat a subsequent mark-down-driven "loss" as an optical event, and check whether it has created an entry.

Related: [[principle-primary-source-beats-vendor]] · [[pitfall-roicai-fcf-field-returns-ocf]] · [[principle-down-a-lot-is-not-cheap]] · [[pitfall-yahoo-misclassifies-strategic-investment-gains]] — the sibling claim. That note covers where the vendor files the mark (Yahoo books it under "Interest Income"); this note covers what the mark does to the multiple. Same underlying accounting, two independent failure modes: one corrupts the line item, the other corrupts the P/E. Check for both.

History

2026-08-04 — Filed during /analyze GOOGL, MU. Verified against Alphabet's quarterly income statement; the ttm reconstruction reproduces Yahoo's reported $19.91 EPS exactly, confirming the statement data is sound and only the derived ratio misleads. static because it is a permanent consequence of ASU 2016-01, not a point-in-time error.