Kumar attributes this decline in margins to rising inference and review costs, saying AI startups cannot rely on the kind of high margins traditionally associated with SaaS businesses.
“(There is) just (one) disappointment that AI companies will not have the luxury of what earlier SaaS companies had (of) higher cost margins because of the inference and the review costs,” Kumar said during a panel discussion at TiE Delhi NCR India Innovation Day 2026.
ETtech had previously reported that SaaS firms typically enjoyed high gross margins of 70–85%. Per a 2025 JM Financial report, their fixed costs, such as investments in technology, do not rise in proportion to revenue, making them more profitable as they scale. Meanwhile, according to the State of AI 2025 report by VC firm Bessemer Venture Partners, gross margins for AI companies are much lower, at around 60%, with some companies reporting margins as low as 25%.
Speaking about the evolution of AI-native businesses, Kumar said companies built “from the ground up” with AI are fundamentally different from firms that merely add AI features onto existing products. He argued that AI-native companies must embed themselves deeply into enterprise and consumer workflows, with customers ultimately paying for measurable outcomes and return on investment rather than for software access alone.
“For us, whatever we call AI native is something that is built from the ground up with AI, and it’s not an add-on at the end,” Kumar said. He added that enterprises and consumers alike are “eventually going to pay for the outcome” generated by the product or service.
Kumar pointed to portfolio companies working with asset managers as an example of the shift underway. Repetitive analyst tasks, Kumar argued, such as preparing reports, investment memos and financial models are increasingly being automated through AI systems. He described this as a case where startups focus narrowly on a specific domain problem and use AI to directly improve operational efficiency and productivity for enterprise clients.
However, Kumar cautioned that while generative AI tools such as Anthropic’s Claude and coding platform Cursor have made it easier for startups to rapidly build minimum viable products, scaling those businesses sustainably remains a challenge.
What should startups then do?
Kumar hence highlighted that founders will need to evolve their business models and prove strong unit economics early as they scale.
“Your business models will (this,) have to evolve so that you are able to first, build more defensibility, second, prove out the unit economics, as you start to scale,” Kumar added.
Kumar also argued that startups operating in highly specialised domains with differentiated workflows or pricing models are more likely to build durable businesses, particularly as competition intensifies and foundational AI capabilities become increasingly commoditised.