Sapiom raised $35 million in a Series A round, announced on August 5, 2026 and led by Dragonfly Capital. The round comes as inference cost, not model capability, is becoming the operational bottleneck for companies running AI agents at scale.
Here's the full breakdown of the raise, who backed it, and what it means for the emerging inference-routing category.
What is Sapiom?
Sapiom is an AI infrastructure platform that provides an OpenAI-compatible router, automatically selecting the most cost-efficient and performant AI model for each individual request. Founded in 2025 by Ilan Zerbib, the company's pitch is that as businesses run more of their operations through AI agents, routing requests to the right model — rather than defaulting to the most expensive one — becomes a direct lever on cost.
Its customers include Polsia and more than 100 agent-building teams, reflecting a customer base concentrated among companies actively building and running AI agents in production.
The raise: $35M Series A
The headline numbers:
- Amount: $35 million
- Round: Series A
- Valuation: not disclosed
- Announced: August 5, 2026
- Lead investor: Dragonfly Capital
Sapiom previously raised a $15 million seed round from Accel in early 2026. Accel returned to participate again in the Series A alongside a new group of investors, which is a notable vote of continued confidence from an existing backer just months after the seed.
Who invested in Sapiom?
The Series A was led by Dragonfly Capital. Additional participants included:
- Accel
- Gradient
- Coinbase Ventures
- Okta Ventures
- Menlo Ventures
- Anthropic
- Array Ventures
- Operator Collective
- Formus Capital
- VanEck Ventures
Anthropic's participation as an investor is notable — a major AI lab backing an infrastructure company whose entire product is routing traffic away from any single model provider toward whichever is most cost-efficient for a given request.
What Sapiom will do with the money
Sapiom said the funding will go toward scaling its AI agent infrastructure and expanding model routing capabilities.
The company says it has processed more than 270 million transactions and delivered roughly a 75% cost reduction per customer — figures that suggest the core routing product is already working at meaningful volume, and that this round is about extending that infrastructure rather than proving the concept from scratch.
Why it matters
- Inference cost optimization is becoming its own infrastructure layer. As more companies move from experimenting with AI to running agents continuously in production, the cost of every model call compounds — creating demand for a routing layer that sits between applications and model providers, similar to how Baseten has built a business around serving models efficiently rather than training them.
- A 75% cost reduction claim, if it holds at scale, reframes where AI margin actually comes from. Model capability gets the attention, but for companies running agents at high volume, the execution layer — routing, caching, cost arbitrage — is where real operating leverage shows up.
- AI labs investing in routing infrastructure is a subtle but important signal. Anthropic backing a multi-model router suggests labs see value in being one option among many for cost-conscious agent builders, rather than insisting on exclusive integration.
Source: Sapiom Raises $35 Million Series A to Power the Next Trillion AI Agents
Following AI funding? Wortins tracks the biggest raises, valuations, and acquisitions daily in the AI Funding Tracker.