Sapiom Raises $35M to Curb Spiraling AI Agent Costs
The San Francisco startup provides a financial routing layer to prevent AI agent expenditures from scaling out of control.
San Francisco-based startup Sapiom has raised $35 million in a Series A funding round led by Dragonfly to help companies manage the volatile costs of AI agents. The investment brings the company's total funding to $50 million, following a $15 million seed round led by Accel.
Sapiom operates a financial and routing layer that sits between AI agents and the large language models they utilize. The platform optimizes spending by routing requests to the most cost-effective model capable of completing a task and enforcing strict budgets before expenditures occur. The impact of this routing is significant; one customer, Polsia, reportedly saw its monthly Anthropic bill plummet from $1.2 million to approximately $100,000. Since launching six months ago, the platform has processed more than 270 million transactions.
The Crisis of AI Sustainability
This funding arrives as corporate leaders struggle with the financial sustainability of agentic AI. According to Gartner, more than 40% of agentic AI projects are predicted to be canceled by the end of 2027 due to a lack of established returns and escalating costs. Sapiom founder Ilan Zerbib argues that current pricing from frontier labs is unsustainable, noting that startups cannot deploy at those rates even when demand exists. Zerbib further claims that in 95% of cases, using an expensive frontier model for agent tasks is unnecessary.
A New Financial Layer for AI
Sapiom enters a competitive landscape of roughly 80 routing competitors, including industry giants like Microsoft Azure and Amazon Bedrock, as well as specialized services like OpenRouter. To differentiate itself, Sapiom serves open-weight models from its own data center racks in San Jose, charging users directly for compute rather than applying a markup.
This shift treats AI agents as financial actors requiring active budget management. Haseeb Qureshi of Dragonfly highlighted the current chaos in corporate AI spending, stating that agents are effectively becoming "employees with no manager and no budget," leaving CTOs to act as CFOs without visibility into where money is going. The strategic involvement of Anthropic as an investor suggests that lowering inference costs for simple tasks may actually drive broader agent adoption by increasing the total volume of requests across the ecosystem.
The Path to Agentic Scale
As companies move from the experimentation phase to full-scale deployment, the "token bill" has become a primary bottleneck for growth. Sapiom is positioning itself as the essential infrastructure to solve this problem. Looking ahead, the company is betting on a massive expansion of the sector; Zerbib believes that trillions of agents will be operating within the economy over the next three years. The industry will now watch whether Sapiom's routing efficiency can prevent the mass cancellation of AI projects predicted by analysts.