Doubling a half-year is a defensible convention right up to the moment one transaction is half the half. Strip Cursor out and 2026 annualizes to roughly $120 billion of announced software M&A value [6], which is 28% of the $429 billion the sector printed in 2021 [7][1], while the deal count runs at about 91% of last year's [2]. Divide the ex-Cursor dollars by the ex-Cursor count and the average disclosed price is around $45 million [3]. Nearly as many buyers as 2025, each writing roughly half the check. (For scale on the concentration, the previous record holder was Q2 2022, when Broadcom/VMware and ICE/Black Knight together managed only 50% of the quarter [9].)
The valuation half of the report is the part that should reallocate someone's planning hours. Cyber Security and Marketing both post a 38% Rule of 40 and trade at 5.2x and 2.1x forward revenue [13], a 148% premium for identical arithmetic [4], and Kroll declines to say what accounts for it [16].
SaaStr, reading the same pages, offers two candidates. The dull one is revenue durability: Kroll's Engineering bucket is Autodesk, Cadence, Synopsys, Dassault, PTC and Trimble, its HCM bucket is ADP, Paychex, Paycom and Workday [15], and twenty-year switching costs in EDA and CAD set against steadily commoditizing payroll will carry a 73% multiple gap without anyone invoking agents [17]. The other, which SaaStr is careful to call consistent with the data rather than demonstrated by it, is that buyers are marking down the categories whose work an agent can absorb and marking up the systems of record an agent has to read from [18].
This is probably wrong, but I take the durability reading, or rather the sharper version of it: the Rule of 40 was never a valuation model, it was a proxy for how long revenue lasts, and it stopped predicting price when categories stopped churning alike. That does not much help a seller, because the money sits in the framing under either story. Salesforce paid 9.5x for Fin inside a comp set trading at 1.8x, having got itself read as an AI agent company rather than as customer-service software [19], which is 5.3 times its category's median [6], and Autodesk paid 26.7x ARR for MaintainX [20], about five times the Engineering revenue multiple even allowing for the different denominators [7].
What would break this read is disclosure. Announced value counts only deals with a published price, and the report summary does not say what share of the near-record count carried one, so a year full of quietly priced small deals would leave the ex-Cursor figure measuring what buyers said rather than what they spent. Until that is itemized, treat the count as measured and the aggregate as estimated.
