A London-based startup called Arrakis emerged from stealth in late July with $38 million in backing, a team of Palantir and Datadog alumni, and a clear thesis: most AI investment has been aimed at desk workers, but the real money is in the 70% of operations that run factories, logistics networks, and industrial infrastructure.

The company announced a $30 million Series A led by Blossom Capital, with participation from Accel (which also led the $7.5 million seed round). The round attracted notable individual backers including Olivier Pomel, founder and CEO of Datadog; Olivier Godement, OpenAI’s Head of Business Products; and Junaid Hussein, founder of Cambridge Aerospace.

Founded in January 2026 by former Accel investor Rafael Quintanilla alongside co-founders Haroun Beltaifa, Romain Fouilland, and Mikhail Galkov, Arrakis is building what it calls an AI operating system for industrial companies. The platform helps organisations deploy AI agents directly into mission-critical operational workflows rather than keeping them in demo mode.

The Problem With Industrial AI Deployment

The “last mile” challenge in industrial AI is well known to anyone who has watched ambitious AI projects stall before reaching production. Most large enterprises have piloted AI in some capacity. Far fewer have turned those pilots into agents that run real workflows with real accountability.

Arrakis is specifically targeting this gap. Its customers operate across aerospace, energy, logistics, manufacturing, construction, and telecommunications. These are industries where operational complexity is high, integration with legacy systems is non-negotiable, and the cost of failure is measured in real dollars, not demo metrics.

Early customer results are notable: deployments achieved a 90% reduction in procurement cycle times, which reflects the kind of operational leverage that justifies serious investment in AI infrastructure.

Why This Matters Beyond Industrial Sectors

The industrial AI story has broader implications for any business trying to move from AI experiments to AI employees.

The pattern Arrakis is addressing, piloting AI tools but struggling to get them into production workflows, is not unique to manufacturers and logistics operators. It shows up in professional services, finance, healthcare, and any organisation where core processes run on complex, interconnected systems built over many years.

What Arrakis is betting on is that AI deployment is a specialised discipline, not a consequence of buying the right model. The hard work is integration, not intelligence. Getting an agent to reliably handle procurement tasks in an aerospace company requires understanding the specific systems, data flows, exceptions, and compliance requirements of that environment.

The funding round adds credibility to this view. Datadog’s Olivier Pomel knows what it takes to build monitoring infrastructure for complex distributed systems. OpenAI’s Olivier Godement has visibility into where enterprise AI deployments succeed and fail. Their participation suggests this is not just a capital-needs story but a conviction bet.

What This Means for Business

For business leaders evaluating AI deployment, the Arrakis story reinforces a few things worth keeping front of mind.

AI agents are ready. The models are capable. The bottleneck is now integration and change management, not technology.

Deployment expertise is becoming a competitive advantage. Companies that figure out how to wire AI agents into real operational workflows will move faster than those still waiting for the technology to mature further.

Industrial sectors are moving. If you assumed AI productivity gains were reserved for knowledge workers, the Arrakis traction numbers suggest otherwise. Operations-heavy businesses are actively deploying and measuring results.

The “last mile” framing is a useful lens for any business: where are the pilots sitting idle that could become real workflows with the right deployment support?

Arrakis plans to use the funding to triple headcount and open offices in New York and the Middle East, targeting the industrial and infrastructure-heavy markets where operational AI deployment is both hardest and most valuable.

For organisations running complex operations, this is a signal that the deployment tooling for industrial AI is maturing fast. The question is no longer whether AI agents can handle these workflows. It is who gets there first.