Essay

Salesforce's AI quarter was real. More than half the profit was paper.

Salesforce grew GAAP earnings per share 119 percent last quarter, and $2.43 of the $4.29 it reported rode on one line: a $2,613 million gain on strategic investments. Net income itself grew 87 percent; the $25 billion buyback program stretched the rest across a shrinking share count. The press release carries the gain plainly and footnotes its per-share effect to the penny, against a year-ago quarter where the same line rounded to zero. The 10-Q gives the line its name: Anthropic, a privately held stake that grew from 22 percent of Salesforce’s strategic-investment portfolio in January to roughly 45 percent by July, about $5.1 billion of an $11.3 billion book, with $2.7 billion of unrealized Anthropic gain landing in this quarter alone. That is more than the whole net line: $296 million of impairments elsewhere in the book saw to the difference. Anthropic raised $65 billion in May at a $965 billion post-money, Salesforce’s fiscal quarter closed July 31, and the new mark flowed straight through the income statement. More than half the reported profit, before anyone touched a product.

Run the beat math and it gets starker. The Street’s consensus sat near $3.27 in adjusted earnings per share; Salesforce printed $5.90, and its own footnote assigns $2.53 of that to the investment gain. Strip the mark and the operating quarter cleared the bar by about a dime.

Nobody was fooled, and that deserves a proper concession. The morning after the print, the Motley Fool ran the subtraction in public under the headline “Thank Anthropic”: $5.90 minus $2.53 leaves about $3.37, from a publication that discloses it is long the stock. CNBC named the gain in its own headline while calling the 22.6 percent pop the second-best day in the company’s history. And the tape says the market was pricing something else anyway: the pop added tens of billions of market cap while the entire Anthropic stake carries at about $5 billion, so the mark cannot arithmetically be what got bought. What got bought: cRPO growth accelerating to 14 percent, a full-year revenue guide raised by $200 million, a revenue print all of $30 million over consensus, and a year of AI-eats-CRM fear dying on a single evening.

Claudeforce, announced the same day as the earnings, is the part the paper-gain story misses. Claude becomes a reasoning model inside Agentforce’s Atlas engine, powering Agentforce Coworker and Vibes by default, and the default model for Slack AI, while Salesforce ships the other direction into Claude as a plugin carrying 37 prebuilt sales skills, open beta in September. Set the stake beside the deal and the shape is familiar. Anthropic writes the code. Salesforce owns the distribution: direct sales, a partner channel, an enterprise renewal engine it has been building since 1999. Salesforce Ventures bought into the Series C in 2023 and every round through at least February’s Series G. Seed the lab early, size the position up, route its models through your own channel. I have argued for years that code without go-to-market is worthless; this quarter is that sentence executed with a balance sheet. Calling it an AI quarter is fair. The bet and the product paid in the same print.

The ledger is where the quarter earns its scrutiny, and it earns it twice. First count: one document, two stories. The financial tables put the gain on its own line and quantify what it did to both earnings figures. The highlights bullet above them reads “GAAP diluted net income per share of $4.29, up 119% Y/Y,” no attribution, beneath a CEO quote calling the quarter “outperforming across every key metric.” Disclosure was never the question; the number sits right there in the tables. The question is which of the two stories survives into the screeners, the reprints, and the memory of what AI earned in 2026. Second count: the one clean product number on the panel was rebased in the same breath. Agentforce ARR crossed $1.5 billion, up over 240 percent, and the bullet announcing it adds, “Effective Q2 FY27, Agentforce ARR includes our AI offerings, Slackbot and Headless 360.” A growth rate quoted in the very quarter its definition widened. The go-to-market is real. The instruments reporting it all belong to the seller. (I sell AI agents to revenue teams for a living, so grade my incentives on that reading however you like.)

Berkshire’s equity portfolio runs the same line, held by the opposite temperament. When the accounting rules began forcing equity marks through the income statement in 2018, Buffett warned the change would produce “truly wild and capricious swings in our GAAP bottom-line”, and a year later told shareholders to focus on operating earnings, paying “little attention to gains or losses of any variety.” Same line on the income statement, opposite highlights. One release steers you away from the mark it must report. The other headlines the growth rate the mark produced.

The fair pushback: if every analyst strips the mark inside a day, where is the harm? Mostly, I concede it. The professionals were never at risk, and the pop was never their confusion. Two things remain. The first is a choice: GAAP gives Salesforce no say about the gain, but non-GAAP is the company’s own definition, and its adjusted $5.90 keeps every dollar of the mark while its guidance assumes no investment gains at all, on the stated ground that nobody can forecast them. Counted when they land, never in the promise (and losses land the same way: over the same six months a year earlier this line ran negative and cost a nickel off both figures). The second is provenance: under the measurement alternative a private stake remeasures only when the issuer prints an observable transaction, so the $2.7 billion was minted by Anthropic’s May fundraise and priced by that round’s buyers. No public market has ever cleared that number. Every step is legal, disclosed, and standard. The headline just sits on a figure with no public price.

Benioff called it one of the best quarters ever, “outperforming across every key metric.” On the metrics that measure Salesforce, it was a good quarter. The metric that made it historic measured Anthropic.

Sources: