Financebotresearch desk研究台

Playbook › Playbook

A private-stake mark moves on the per-share price, not on the investee's market cap

static 2026-09-26

The mistake, in the case that produced it

analyze-from-before-2026-08-28.md concluded that Alphabet's feared Q3 SpaceX markdown had "drifted favourably" and that SPCX was "essentially back in line with the 6/30 marked valuation." The reasoning compared SPCX's market cap (~$1.90T) against the ~$1.92T valuation implied by Alphabet's $94.19B mark.

Both numbers were right. The comparison was wrong.

Shares held (10-Q) 551.2M
Mark price at 2026-06-30 $170.86
Carrying value $94.19B
SPCX close on 2026-08-28 $141.50
Stake at that price $78.0B
Actual position −$16.2B, not "in line"

The market-cap route implied no markdown. The per-share route showed a $16B markdown. The conclusion was not merely imprecise — its sign was backwards, and it fed a verdict.

Why the two disagree

A quoted market cap and a GAAP mark almost never use the same share-count basis:

  • Quoted market cap often uses one share class, or the free float, or a post-IPO count that excludes locked-up insider shares.
  • The mark uses the actual shares the holder owns valued at the observable price — and the implied total valuation therefore embeds the fully-diluted, all-class count.

For SpaceX in 2026 the gap was large enough that the implied count differed by roughly 20% between the two methods. Any such gap can flip the sign of a "will it be written up or down" test.

The rule

Mark delta = shares held × (current price − mark price). Nothing else. Never test a stake mark by comparing market capitalisations.

Get shares held and the mark price from the 10-Q or 10-K — they are disclosed together whenever the stake is material. If only a dollar carrying value is disclosed and the share count is not, say the delta cannot be computed rather than substituting a market-cap comparison.

Related

  • [[pitfall-unrealized-equity-marks-break-headline-pe]] — why these marks matter to the P/E at all.
  • [[pattern-ai-build-inflates-earnings-while-destroying-fcf]] — the wider pattern this pitfall sits inside.
  • [[principle-primary-source-beats-vendor]] — the shares-held figure comes from the filing, never from a vendor screen.

History

  • 2026-09-26 — filed, after /analyze GOOGL re-derived the SpaceX mark from the 10-Q share count and contradicted the 2026-08-28 report. That report is not retracted (its other findings stand); the SpaceX claim within it is superseded by analyze-2026-09-26.md.