Infrahub maker OpsMill raises $14M on bet that AI agents need better infrastructure data before they can work

Enterprises are racing to deploy AI agents that can configure, monitor, and fix infrastructure automatically. The problem is that those agents can't see clearly — the data about servers, networks, and cloud configs lives in a half-dozen disconnected tools that don't agree with each other. OpsMill has raised $14 million to be the layer that unifies it.

The Paris-based startup closed a $14 million Series A led by IRIS, the firm behind over 230 investments including some of Europe's larger infrastructure-software bets. Its product, Infrahub, is an open-source graph database purpose-built as an infrastructure control plane — a single place to query everything from VLAN assignments to device fingerprints, with built-in version control, CI workflows, and change approval built in.

The pitch is direct: enterprises have spent years building infrastructure automation with Terraform, Ansible, NetBox, and IP Fabric. Each tool has its own copy of the truth. When an AI agent needs to answer a question — which port is this server using, which customer VLAN does this device belong to — it typically has to query multiple systems and reconcile conflicting answers. Infrahub is the single unified view those agents can query instead.

"One common problem is data inconsistency," Cedric Grard, a senior cloud architect at Eurofiber, explained in a case study on OpsMill's site. Eurofiber, which operates data center and infrastructure-as-a-service across France, cut service deployment from five days to 15 minutes after replacing its patchwork of tools with Infrahub plus native Terraform and Ansible integrations. The company still manages everything in NetBox, but Infrahub sits above it as the queryable federation layer.

The mechanism that makes Infrahub relevant to the AI-agent wave is a Model Context Protocol server — the same standard that lets AI assistants like Claude and Cursor share tools across different model providers. Infrahub's MCP server exposes the validated, relationship-rich data in its graph to agents, so they can ask infrastructure questions and get consistent answers. The GitHub repository shows an active Apache 2.0 project with SDK bindings, Ansible and Nornir plugins, NetBox sync connectors, and a GraphQL interface alongside the REST API.

IRIS leading the round is the institutional signal worth noting. The firm has backed more than 230 companies and invests at the Series A stage in infrastructure software. For OpsMill, the money buys runway to hire the enterprise sales and partner engineering team needed to move beyond open-source adoption into structured procurement deals — the company has said it will use the funding to expand integrations with cloud vendors and network equipment providers, the two vendor categories that determine whether Infrahub gets embedded into enterprise procurement workflows or stays in the devops tooling drawer.

Infrahub's graph schema is where the control-plane claim gets concrete. The platform maintains structured models for devices, services, topologies, IP address blocks, VLAN assignments, and the relationships between them — synchronized from NetBox, Nautobot, IP Fabric, and Terraform state — so that any query gets a consistent answer rather than whatever the last tool to write to the database happened to return. That federation layer is what makes it useful as an agent backend: an agent can ask "which VLAN is this device on" and get a specific answer, not a list of conflicting values from four tools that haven't talked to each other in months.

The implications extend beyond one company's pitch. If the agent-infra convergence thesis holds, every AI agent framework needs a visibility layer beneath it — a way to get consistent data about the infrastructure it's orchestrating. That creates pressure on incumbents like NetBox and Nautobot, which already hold infrastructure data but were built for human operators querying a UI, not for agents that need consistent, machine-readable answers across multiple systems at once. It also opens a new category between the agent layer and the hardware: the agentic infrastructure stack, where Infrahub is competing for position alongside older CMDB vendors and newer federation tools. The thesis shows up in product announcements and funding rounds before it lands in published research — the pattern is visible to anyone watching the agent-infra convergence, but no major analyst firm has published dedicated coverage of this layer yet.

The counterforce is real. NetBox, the open-source infrastructure tracking tool Infrahub competes most directly with, has over 20,000 GitHub stars and a mature ecosystem. Enterprises already know NetBox; convincing a network team to add a graph-backed federation layer on top of it is a harder sell than the demo suggests. The $14 million is small for a 2026 infrastructure Series A, which suggests investors are buying the thesis — the agent-infra convergence — more than the current revenue proof. OpsMill's co-founders, CEO Damien Garros and COO Raphael Maunier, have backgrounds in large-scale infrastructure automation, according to the company's about page. That track record is the best evidence the product works in the wild.

What to watch: whether OpsMill converts design wins at larger enterprises into a reference cohort. The most likely early adopters — aside from pure devops shops — are managed service providers and telecom operators running multi-vendor network environments, where data inconsistency across vendor tools is a chronic problem. The test is whether the next named customer is a Fortune 500 with a documented deployment story, not another reference to the Eurofiber case study. On the competitive side, NetBox's community is actively discussing MCP compatibility; if NetBox ships an official MCP server before OpsMill reaches critical mass, the window for Infrahub as the agent-native layer narrows considerably.