'Infrahub' maker OpsMill raises $14M to give AI agents a single view of

Your AI agent executes a network configuration change at 2 AM. Nobody’s watching. The change looks correct based on the data the agent was given — except that data was pulled from a spreadsheet last updated eight months ago, cross-referenced with a script someone wrote in 2019, validated against a configuration database that doesn’t account for three infrastructure changes made in Q4. The agent proceeds confidently. By 6 AM, a production line is down.

Now contrast that with an agent that runs against a single, structured, relationship-aware map of every infrastructure element — physical, virtual, and cloud — with every proposed change validated before it touches production. Same agent. Radically different outcomes.

That gap — between agents flying blind and agents with trustworthy data — is exactly what a Paris-based startup just raised $14M to close. And for anyone serious about deploying AI agents on real infrastructure, the implications run deeper than the funding announcement suggests.

What OpsMill Announced Today

OpsMill, founded in 2023 by Damien Garros and Karen Gallantry, announced a $14 million Series A led by IRIS Capital Management, with participation from Benhamou Global Ventures, Serena Capital, and Partech Partners. The funding will expand engineering and product teams behind Infrahub, the company’s flagship infrastructure data management platform.

Infrahub is available in two versions: a free open-source community edition and a paid enterprise edition — the same model GitLab uses. The open-source community already includes hyperscalers like TikTok. The enterprise edition is deployed across retail, insurance, manufacturing, and fintech customers in Europe and North America.

The numbers that actually matter: European cloud services provider Eurofiber Group cut service deployment times from five days to 15 minutes after deploying Infrahub. The broader market context: enterprise network automation demand has tripled since 2023, in a market valued at $220 billion.

The Real Bottleneck Isn’t the AI — It’s the Data Underneath It

Here’s the thing that most AI automation coverage misses: writing code to automate infrastructure has been solvable for years. The hard part — the part that actually kills projects — has always been data quality.

Garros put it directly: “Automation is ultimately a data problem and if you only have a partial view of your network, you’re flying blind. Writing the code for automating infrastructure was never the problem, the challenge has always been maintaining it and being able to trust it in production.”

Most enterprise infrastructure data today lives in spreadsheets, bolted-on scripts, and configuration management databases (essentially, organized asset lists) that were never designed to guide AI. These tools treat infrastructure as a static table of rows: server A exists, firewall B exists, switch C exists. What they don’t capture is the relationships — that server A depends on firewall B’s rules, that switch C is the path between them, and that changing any one of those without understanding the others can bring down a production system.

Infrahub approaches this differently. It uses a graph database — software that maps connections rather than just lists — to represent infrastructure as a dynamic web of relationships. Every element knows its context: how it connects to everything else, what it depends on, what depends on it. When an AI agent queries this system before making a change, it’s working from a structured, validated picture of reality rather than a patchwork of stale files.

The platform also adds a governed approval layer: every proposed change — whether from a human engineer or an AI agent — gets validated before deployment. Think of it as a safety gate that checks “does this change make sense given everything we know about this infrastructure?” before anything touches production.

What This Means for Anyone Building or Running AI Agents

Zoom out. This funding round is a data point in a much larger pattern emerging across the AI agent ecosystem. The problem OpsMill is solving isn’t unique to OpsMill — it’s the same problem showing up everywhere agents are being deployed on real infrastructure.

The AI agent space has spent two years celebrating capability: what agents can reason about, what tools they can call, how many tasks they can handle in parallel. That was the right conversation for 2024. In 2026, the conversation has shifted. The ceiling isn’t intelligence anymore — it’s data quality. Julien-David Nitlech, managing partner at IRIS, said it plainly: “Without clean, structured, trustworthy infrastructure data, AI-driven operations simply cannot function at scale.”

For teams exploring agentic AI in enterprise environments, this surfaces a prerequisite that often gets skipped. The question isn’t just “can our AI agent handle this workflow?” — it’s “does our AI agent have accurate enough data to act on this workflow safely?” Those are different questions, and the second one matters more.

The stakes are especially high in regulated environments. OpsMill specifically calls out financial services, where a misconfigured firewall rule doesn’t just cause an outage — it can trigger regulatory penalties. In manufacturing, unreliable automation can shut down production lines entirely. The downside of getting this wrong isn’t a failed API call. It’s a $300,000-per-hour hole in your operations.

This also signals something important for the broader AI automation market: the infrastructure layer is becoming a first-class concern, not an afterthought. The wave of agent platform investment we’ve tracked — from Sierra’s $950M round to emerging platforms across the stack — is converging on the same realization. Agents are only as reliable as the systems they operate against.

Some things only make sense once you have the whole picture — Beacon knows that one light, pointed in the right direction, changes everything.

What to Do About It

• Audit your infrastructure data before deploying agents. If your team is planning to give AI agents operational access to your network or cloud environment, start by mapping where your infrastructure data actually lives — spreadsheets, CMDBs, scripts. If the answer is “multiple disconnected places,” that’s your bottleneck, not your agent.
• Treat data quality as a prerequisite, not a Phase 2. The pattern that keeps producing $300K/hr failures is deploying agents first, then discovering the data problem. Establish a single source of truth before granting agents operational access.
• If you’re in financial services or manufacturing, prioritize this now. Regulatory penalties and production line shutdowns are the failure modes OpsMill specifically calls out. The risk calculus in those industries makes this a compliance issue, not just an engineering one.
• Evaluate Infrahub if you’re managing complex infrastructure. The open-source community edition is free and already used by hyperscalers. If your team manages hundreds or thousands of interconnected infrastructure elements, it’s worth understanding what relationship-aware infrastructure data looks like in practice — versus a standard configuration management database.
• Watch what $14M buys in the next 12 months. OpsMill is explicitly growing its engineering and product teams. How they expand Infrahub’s integrations — particularly with major AI agent orchestration frameworks — will signal whether this becomes foundational infrastructure for the agent ecosystem or stays a specialist tool.

What the OpsMill Round Signals for the Agent Ecosystem

• OpsMill raised $14M in a Series A on May 7, 2026 to expand Infrahub, its open-source infrastructure data management platform — evidence that investors see data quality as a hard prerequisite for enterprise AI automation.
• The code was never the bottleneck. Infrastructure automation has been technically possible for years. The actual failure point has always been fragmented, untrustworthy data — scattered across spreadsheets and outdated scripts — that agents act on at machine speed.
• Cascading failures cost an average of $300,000 per hour. When AI agents operate on bad infrastructure data, they don’t produce isolated errors — they propagate mistakes across hundreds of connected systems faster than humans can intervene.
• Eurofiber cut deployment times from 5 days to 15 minutes after deploying Infrahub — a 480x improvement that illustrates what agents can do when the data layer is clean and relationship-aware.
• The pattern to watch: Enterprise network automation demand has tripled since 2023. The companies that invest in data infrastructure now will be the ones whose agents work reliably at scale. The ones that skip it will be the case studies in the next wave of AI failure post-mortems.

The teams that solve the data problem first don’t just avoid the $300K/hr failures — they build a compounding advantage. Every reliable agent deployment earns more trust, unlocks more automation, and creates more distance from competitors still untangling spreadsheets. The technology for capable AI agents is largely solved. The infrastructure beneath those agents is the new frontier. OpsMill just got $14M to prove it.