Sapiom Raises $35M Series A to Cut the Cost of Running Agents
Sapiom announced a $35 million Series A on August 5, led by Dragonfly with participation from Accel, Gradient, Coinbase Ventures, Operator Collective, Formus Capital and VanEck Ventures, plus existing investors Okta Ventures, Menlo Ventures, Array Ventures β and Anthropic. The round comes six months after a $15 million seed and eleven months after launch, bringing the total to $50 million in under a year. Founded in 2025 by CEO Ilan Zerbib (sapiom.ai).
Sapiom sits between AI agents and the models they run on: an infrastructure layer for deploying, operating and scaling agents in production, with cost routing as the sharp edge. The traction numbers are what pulled a round this size this fast: over 270 million transactions processed since launch, more than 100,000 agent runs per day, and one customer cutting inference costs 75 percent after moving its agents onto the platform.
After weeks where the funding window returned nothing agent-related, two things about this round stand out. First, the thesis: everyone funded agent building in 2025; the 2026 money is going to the unglamorous layer that makes agents economically survivable. A 75 percent inference-cost cut is the difference between an agent product with a margin and one without. Second, the cap table: Anthropic invested at seed and came back for the A. The lab whose models generate those inference bills is backing the company that shrinks them β which tells you the labs believe agent workloads only scale if someone solves the bill.
← Back to all articles
Sapiom sits between AI agents and the models they run on: an infrastructure layer for deploying, operating and scaling agents in production, with cost routing as the sharp edge. The traction numbers are what pulled a round this size this fast: over 270 million transactions processed since launch, more than 100,000 agent runs per day, and one customer cutting inference costs 75 percent after moving its agents onto the platform.
After weeks where the funding window returned nothing agent-related, two things about this round stand out. First, the thesis: everyone funded agent building in 2025; the 2026 money is going to the unglamorous layer that makes agents economically survivable. A 75 percent inference-cost cut is the difference between an agent product with a margin and one without. Second, the cap table: Anthropic invested at seed and came back for the A. The lab whose models generate those inference bills is backing the company that shrinks them β which tells you the labs believe agent workloads only scale if someone solves the bill.
Comments