Sapiom raises $35M Series A to cut the cost of running AI agents

Sapiom, a San Francisco startup that sits between AI agents and the models they run on, has raised a $35 million Series A led by Dragonfly. The round drew Accel, Gradient, Coinbase Ventures, Operator Collective, Formus Capital, and VanEck Ventures, alongside existing backers Okta Ventures, Menlo Ventures, Anthropic, and Array Ventures.

The round lands just 11 months after Sapiom launched and six months after a $15 million seed, taking total funding to $50 million. It ranks among the largest 5% of Series A rounds for its sector and region on record.

Sapiom targets the cost of running agents in production. Its Router sends each model call to the cheapest capable model rather than the most expensive one, enforcing budgets before money is spent. Alongside the financing, it launched Router, Agent Studio — a local environment for building and testing agents — and Runtime, its managed production infrastructure.

One customer, AI startup Polsia, saw its monthly token bill on Anthropic hit $1.2 million as projected revenue jumped from $100,000 to $10 million in a year, Semafor reported. After Sapiom's evaluations, that bill fell roughly tenfold, to about $100,000. Since launching six months ago, the platform has processed more than 270 million transactions across over 100,000 agents running daily.

Dragonfly general partner Haseeb Qureshi is joining Sapiom's board. "Agents are becoming employees with no manager and no budget," he said, "and increasingly, the CTO is the one acting as CFO, allocating real money with no visibility into where it goes."

Routing is starting to look like a commodity. Amazon and Microsoft bundle it into Bedrock and Azure, open-source routers are free, and one tracker counts 80 active competitors. Sapiom's bet is what it owns underneath: it serves open-weight models from its own racks in a San Jose data centre and charges for compute directly, rather than adding a markup.

Cost is becoming the hard constraint on agentic AI. Gartner forecasts that companies will cancel more than 40% of agentic AI projects by the end of 2027, with escalating costs a leading reason. Zerbib is betting on volume: "we're talking about trillions of agents that will operate in the economy in the next three years," he said — most of which, on his numbers, do not need a frontier model.