San Francisco-based Sapiom announced on August 5, 2026, that it has closed a <cite index="8-5,8-6">$35 million Series A led by Dragonfly, with participation from Accel, Gradient, Coinbase Ventures, Operator Collective, Formus Capital, and VanEck Ventures, alongside continued backing from existing investors including Okta Ventures, Menlo Ventures, Anthropic, and Array Ventures.</cite>
<cite index="8-7">The Series A comes six months after Sapiom's $15 million seed round and just 11 months after the company was founded, bringing total funding to $50 million.</cite> Dragonfly Managing Partner Haseeb Qureshi will <cite index="3-11">join Sapiom's board of directors as part of the financing.</cite>
The Problem: Production Economics for AI Agents
<cite index="12-12,12-13">Sapiom, founded in 2025 by Ilan Zerbib, builds infrastructure that helps engineering teams get Artificial Intelligence (AI) agents into production, targeting the specific problem that most AI agents work well as demos but fail when companies try to run them continuously, at scale, and within a budget.</cite> The broader market context is acute: <cite index="14-20,14-21">Gartner forecasts that companies will cancel more than 40% of agentic AI projects by the end of 2027, with escalating costs among the leading reasons.</cite> Industry survey data cited by Semafor reinforces the pressure — <cite index="13-4">Forrester predicted enterprises will postpone a quarter of planned AI spending to 2027, and a KPMG survey of more than 2,100 senior executives in June found only 7% could point to established returns from using AI.</cite>
How the Platform Works
<cite index="15-1,15-2,15-3">Sapiom sits at the point of execution, turning fixed infrastructure choices into real-time decisions. For every action, it selects the best allowed path across models, compute, tools, and services based on the requirements of the task, cost, quality, latency, reliability, and company policy. Budgets and permissions are enforced before execution, and every action is metered and recorded in a complete audit trail.</cite>
A key differentiator is ownership of the underlying compute. <cite index="4-7">Sapiom runs open-weight models on its own server racks in a San Jose data center and charges directly for compute rather than adding a markup.</cite> CEO Zerbib has stated that <cite index="13-6">"in 95% of cases, it doesn't make sense to go to a very expensive frontier model."</cite>
Three Products Launching Alongside the Round
<cite index="15-4,15-5,15-6,15-7">Alongside the financing, Sapiom is launching Sapiom Router, Sapiom Agent Studio, and Sapiom Runtime. Router matches each model call to the right Large Language Model (LLM) instead of defaulting to the most expensive one. Agent Studio gives engineering teams a local environment to build, test, inspect, and deploy agents with the context of their existing codebase. Runtime provides the managed production infrastructure where those agents execute with the access, routing, recovery, step-level visibility, and controls required to operate at scale.</cite>
Early Traction and Customer Evidence
<cite index="8-1">Since launching six months ago, the platform has processed more than 270 million transactions across more than 100,000 agents running daily.</cite> The most cited customer case involves AI agent startup Polsia: <cite index="4-10,4-11">Polsia was spending $1.2 million a month on Anthropic tokens as its projected Annual Recurring Revenue (ARR) lurched from $100,000 to $10 million; Sapiom claims it got that bill down to roughly $100,000 a month.</cite>
Competitive Landscape and Investor Dynamics
<cite index="5-4,5-5">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.</cite> Sapiom's compute-ownership model is presented as its primary moat.
One notable tension: <cite index="5-1,5-2">Anthropic, whose token bills Sapiom is built to shrink, is an investor. Zerbib frames it as aligned — cheaper inference lets companies build more agents, some of which will still need frontier models.</cite> Proceeds from the round will be directed toward scaling the platform and hiring across engineering, machine learning (ML), and go-to-market functions, <cite index="6-10">with a particular focus on routing policy, inference systems, and post-training.</cite>