Zoho cofounder warns AI costs may break SaaS economics

Zoho cofounder Sridhar Vembu has publicly questioned whether the cost of building AI-powered software can ever square with the economics of the SaaS model. In a post first reported by Inc42, Vembu argued that the compute and data expenses required to train and run large language models threaten to push gross margins below levels that investors or public markets will tolerate.

The timing is pointed. Workday’s launch of Sana, an AI agent for HR and finance workflows, came only two weeks after Anthropic’s leaked S-1 revealed an $8 billion loss on surging revenue. That filing laid bare the arithmetic behind AI’s scaling costs: every incremental dollar of revenue demands a disproportionate increase in cloud spend. PhonePe’s internal audit automation, which cut cycle time from four days to two hours, shows the efficiency gains AI can deliver, but Vembu’s post suggests those gains may not translate into sustainable margins for companies that sell software by the seat.

AMD’s push to move AI inference onto local endpoints hints at one possible escape valve. If enterprise workloads can shift from centralized cloud clusters to on-premise hardware, the cost curve might flatten. For now, though, the tension is unresolved. Vembu’s warning leaves SaaS founders weighing whether to chase AI differentiation at the risk of breaking their own economics, or to wait and risk falling behind.

Sources: inc42.com