Anthropic's valuation mark supplied 43% of Salesforce's adjusted quarterly profit

The composition is the interesting part. Salesforce's filing put $2.7 billion of unrealized gain against Anthropic in a quarter whose total net investment gains came to $2.6 billion [c5b]^[4], which means the other 450-odd holdings, carried at roughly $6.2 billion between them, netted out to about negative $100 million ^[3][9]. One private round wrote the whole line (the gain was 42.9% of the adjusted number and 56.6% of the GAAP one ^[1][2]).

A $5.1 billion stake against the $965 billion valuation Anthropic disclosed in May is about 0.53% of the company ^[5][7][5], and if Salesforce added nothing to the position during the three months, the holding entered the quarter carried near $2.4 billion ^[6]. More than half of today's mark is an accounting entry made since the start of May.

Underneath, revenue rose 11% to $11.3 billion with subscription and support up 12%, and non-GAAP operating margin was 34.1% ^[11][12]. GAAP operating income was flat at $2.3 billion while interest expense went to $473 million from $67 million, up about seven times ^[13][10]. With diluted shares down roughly 15% ^[10], flat operating income should have delivered something like 18% more per share; ex-gain GAAP earnings instead fell about 5%, to $1.86 from $1.96 ^[7]. The buyback bought eighteen points of per-share arithmetic and the financing line ate them.

This is probably wrong, but the more useful reading is that Thursday was a re-rating the contracted numbers already justified and the mark merely triggered: the stock went into Wednesday's report about 24% below its 52-week high ^[17], it still trades at roughly 16 times next year's expected earnings after a 23% day ^[18], current remaining performance obligation stood at $33.5 billion, up 14% in constant currency ^[14], and Agentforce with Data 360 added about $500 million of annual recurring revenue sequentially to reach nearly $3.9 billion, close to 15% in a single quarter ^[15][8]. Those are signed contracts and they do not depend on anybody's mark. The counter-thesis is that a company guiding to 11% to 12% revenue growth for the full year ^[16] has not earned a 23% day, and that between the Anthropic mark and the Claudeforce plug-in announced alongside the results ^[8], the market paid for the same relationship twice in one session.

What would settle it is cheap to observe. If the sequential ARR add holds near $500 million and cRPO stays in the mid-teens, the software business grows into the price and the portfolio entry was incidental to it. If Anthropic's next priced round comes in flat or below, the same mechanism runs in reverse and the 103% adjusted earnings increase ^[2] becomes a subtraction of comparable size, with revenue and margin unchanged either way. I hold this loosely: at 16 times forward earnings the price looks defensible on the contracted revenue alone, which is precisely why it is a pity the quarter will be remembered for a line in the investment schedule.