AI-first software companies posted faster recurring-revenue growth in 2025 but lower margins and less annual recurring revenue per employee, highlighting the cost pressures behind a fast-growing business model, a KeyBanc benchmark found.
KeyBanc Capital Markets announced the 17th edition of its software benchmark Nov. 13. Median ARR growth was 34% among companies selling AI as their main product, 22% at firms adding AI to broader software and 9% among those still developing customer-facing AI.
Growth did not translate into stronger operating economics. The median EBITDA margin was -12% for AI-first companies and -1% for embedded-AI firms. AI-first businesses recorded $117,000 in ARR per full-time employee, compared with $201,000 for companies integrating AI into broader offerings, KeyBanc said.
AI expenses also exceeded plans for many respondents. Forty-four percent said their AI spending was over budget year to date, and 24% said it was more than 20% over. The most common explanation was token usage running above forecasts. Companies devoting more than 5% of revenue to AI had a 66% median gross margin in 2025; among those spending less than 2%, the median was 78%.
Separate AI fees corresponded with 27% median ARR growth in 2025; the median was 17% among companies without a separate charge. KeyBanc said that growth advantage had not yet led to higher net retention.
The benchmark showed AI-driven ARR growth of 74% in 2025 and projected 128% in 2026, when AI-driven ARR was expected to make up 12% of total ARR. Across respondents, median ARR growth was projected to ease to 23% in 2026 from 24% in 2025. Participants’ plans put growth at 40% in 2027.
Scott Peterson, managing director of software investment banking at KeyBanc Capital Markets, called 2026 “a proving year for AI-driven revenue.” KeyBanc also reported that 92% of respondents said AI shortened product development cycles, while 91% used AI internally in engineering. Peterson said companies were reinvesting gains from AI coding tools in faster product development.
Staffing expectations varied between past experience and the outlook. Sixty-five percent reported no AI-related cuts or reallocations in 2025; 34% expected no such changes over the next 12 to 24 months. Respondents identified research and development (45%) and customer service (37%) as the functions most likely to be affected. The benchmark, previously called the SaaS Survey, expanded this year to examine AI’s effects on software strategy, monetization and operating models.
