Salesforce: Reacceleration Is Real, but the EPS Double Was Borrowed

Salesforce spent 2026 as Wall Street's best-known test of whether artificial intelligence would make traditional software obsolete. The shares fell from a 52-week high above $269 to a low near $146 — a 45% collapse — with fresh lows arriving almost monthly through the spring and summer while the S&P 500 rose double digits. Then one earnings report changed the debate. After Salesforce reported fiscal-second-quarter results on the evening of August 26, the stock jumped 14% in extended trading, climbed another 22% the next day, and now trades near $252, erasing most of its 2026 loss in a single sitting.

Both sides came out of that night claiming victory. Believers saw proof that AI strengthens Salesforce rather than replaces it. Skeptics saw an earnings number that "more than doubled" and wondered how much of it was real. The quarter actually contained both an answer and a trap, and telling them apart is the entire investment question at $252.

The panic priced in the worst of it

Salesforce's 2026 problem was never its reported performance. It had beaten estimates for four straight quarters and grown revenue at double-digit rates while the stock sank — the classic profile of a valuation resetting faster than the business deteriorated. By early August the shares were around $184, down roughly 30% for the year, even as the upgrade attempts kept coming. J.P. Morgan initiated coverage on August 13 arguing the AI fears were "overblown" and that the stock was priced for further deceleration. A few days later Citi lifted its target on what its checks called "modest improvement" in Agentforce deployments. Jim Cramer made the bull case explicit on August 19 with the shares near $196 — about 14 times forward earnings — declaring that "the selling goes too far" and that the AI threat to a proprietary platform like Salesforce was essentially theoretical. CEO Marc Benioff was publicly pushing the same line, insisting the fear that "AI will kill Salesforce" was dead wrong.

The message of the weeks before earnings, when the stock drifted from roughly $184 to $205, was that a panicked market had already done the downgrading. Prediction markets had gone further, assigning a 91% probability of a beat to the August 26 report.

The bookings proved the story

The operating heart of the quarter was its forward indicator. Current remaining performance obligation — contracted revenue due within the next twelve months, the cleanest gauge of where revenue growth is heading — grew 14% in constant currency to $33.5 billion, ahead of the Street's $33.2 billion estimate and the fastest bookings growth Salesforce has reported in four years. For a company whose stock was priced for terminal deceleration, that one number was the vindication, and it holds even before counting the acquisitions still being integrated.

The AI line items backed it up. Agentforce annual recurring revenue topped $1.5 billion, up more than 240% — with an honest caveat that the metric was broadened this quarter to include Slackbot and Headless 360, so part of the jump is definitional. Data 360 ARR reached nearly $3.9 billion, up over 210%. Usage scaled as well: 3.2 billion agentic work units delivered in the quarter, up 97% sequentially, and bookings for the premium Agentforce bundles more than doubled quarter over quarter. Slack posted its fastest net-new-order growth since acquisition, and customer attrition stayed near record lows. Financially, free cash flow rose 81% to $1.1 billion against a consensus near $643 million, on a 34.1% non-GAAP operating margin. The quarter also brought a $1.6 billion contract from the U.S. Department of Veterans Affairs and an expanded partnership with Anthropic — a "Claudeforce" plugin that embeds Claude into Salesforce. The very company whose rise was the bull's biggest bear case now sells inside the product. That is the coherent story: customers are not replacing Salesforce seats with agents; they are paying Salesforce for agents on top of seats.

The EPS double was borrowed

Here is the trap. Adjusted EPS printed at $5.90, up 103% from $2.91 a year earlier and about 80% ahead of the Street — the headline that made the reaction violent. But a primary driver of the beat was a $2.6 billion one-time gain on Salesforce's strategic investments, chiefly its stake in Anthropic, which was valued at $965 billion in a May funding round. On roughly 820 million diluted shares, that gain is worth about $3 a share — essentially the entire $2.99 year-over-year increase in adjusted EPS. Back it out and the underlying bottom line was roughly flat, a normal quarter rather than a doubling.

The guidance raise carries the same fingerprint. Full-year revenue guidance inched up only $200 million, of which just $100 million is organic, and the operating margin outlook held at 34.3%. Yet full-year adjusted EPS guidance jumped from roughly $14.06–$14.12 to $16.67–$16.71 — a $2.60-a-share lift about the size of the one-time gain, helped by a buyback that has already retired 103 million shares under a $25 billion accelerated repurchase program. Management's earnings outlook moved up for reasons that have nothing to do with operations improving. The market cheered a bottom line that was partly borrowed from the asset side of the balance sheet.

This is not an argument that the quarter was bad; it was genuinely good. It is a question of what you are paying for the durable part.

What $252 is asking

At the August trough, near $185, the stock traded around 14 times forward earnings with expectations set for deceleration — the fear had already done the discounting, which is why the Cramer-and-upgrade view worked. At $252 the cheap bridge is spent: roughly 15 times the raised, gain-flattered forward EPS, or about 18 times once you normalize the one-time gain, against an enterprise value-to-EBITDA of around 18.5 times. That is not expensive next to ServiceNow, at roughly 52 times, or Oracle, near 26 times trailing earnings — but those grow faster, and Salesforce's organic revenue is still up only around 7% once Informatica's roughly four-point contribution is stripped out. The multiple now charges in advance for a reacceleration that bookings point to but the income statement has not yet delivered.

That is the good-company/good-stock split in one line. The business is better than 2026's price ever implied. The easy part of the stock — buying the fear — was August. What remains is a reacceleration option that has to keep proving itself.

The next proof window

The test is the fiscal-third-quarter report, roughly three months out. If cRPO holds at or above 14%, the story stays intact. If it drifts back toward high single digits, the 15-to-18-times multiple has room to compress even if nothing looks visibly wrong, because the price for proof has already been paid. The specifics to watch: whether the bookings strength converts into organic revenue, whether Agentforce ARR keeps compounding from a base that one analysis puts at only about 5% of eligible users, whether the pending Contentful and Fin deals land inside the raised guidance, and whether future marks on the investment portfolio keep flattering net income or start hitting it.

The August print settled 2026's argument, and it settled it for the people who bought the panic. That verdict is worth respecting. The next two quarters settle a different argument — whether a reacceleration that is visible in bookings becomes one you can see in the income statement, without the assist from the asset side of the balance sheet.