Former F1 engineers raise $55M to deploy AI robots inside factories

Why it matters

The round backs a deployment-layer strategy for robotics: adapting available machines and models to factory-specific work, where reliability and customer data remain the bottlenecks.

microagi raised a $55 million seed round led by Hummingbird Ventures to expand Atlas, its system for collecting factory-floor data, training robot control policies and deploying them on existing hardware.

Northzone, LocalGlobe, Village Global and redalpine participated in the round, which microagi announced on July 16th. The company calls it the largest seed financing raised in Germany, a claim repeated in a July 20th thread on X and attributed to microagi in independent reporting. The valuation was not disclosed.

Founder and CEO Bercan Kilic (@bercankilic) started microagi in September 2025 after working as an aerodynamics engineer at Red Bull Racing. His co-founders include former Mercedes-AMG Petronas engineer Yoan Iliev, former Alan Turing Institute researcher Anton Poletaev, RWTH Aachen engineer Nico Nussbaum and Artjem Weissbeck, who previously co-founded the WhatsApp commerce software company Charles.

That group gives microagi an unusual starting point: engineers accustomed to optimizing complex physical systems, paired with researchers and operators building the data pipelines needed to train machines outside a controlled lab. The round arrived about 10 months after the founders began working from what they described as a Munich hacker house.

Atlas targets the deployment layer

Atlas is built around the gap between a robotics demonstration and a machine that can reliably finish a production shift. According to microagi's funding announcement, recording equipment placed on production lines captures demonstrations from experienced workers. Control policies are trained on that data in simulation, then deployed on a robot selected for the job.

Microagi says Atlas is hardware- and model-agnostic. Its business sits in the deployment layer: collecting customer-specific operational data, adapting existing robotics models and putting the resulting system into production. That approach lets microagi work with machinery a manufacturer already owns or source robots from outside vendors.

Semafor reported that microagi commonly buys hardware from Chinese robot manufacturers Unitree and UBTech, leases the machines to industrial customers and installs software fine-tuned for each site. Microagi also consults with manufacturing and logistics groups, recording live operations before identifying work that can be automated.

The setup gives microagi exposure to several parts of a deployment, including data collection, model adaptation, hardware procurement and on-site integration. It also makes execution more demanding than selling a standard software subscription. Each factory brings different machinery, work practices, safety requirements and edge cases.

Microagi's public announcement refers to deployments with large industrial companies without naming customers, publishing revenue or disclosing how many robots Atlas currently operates. The $55 million round therefore funds a strategy whose commercial scale remains difficult to measure from the outside.

Human work supplies the training data

Microagi has also built a consumer-facing data operation called Shift. In May, Shift offered free apartment cleanings in New York in exchange for permission to record the work. Operators wore cameras while completing household tasks, producing first-person footage that could be prepared for robotics training.

Kilic told Semafor that microagi had licensed some of that data exclusively to frontier AI laboratories. The same material can help shape control systems for robots deployed in factories, giving microagi a second source of training examples beyond industrial production lines.

The Shift program shows how Kilic is trying to address one of robotics' central constraints: useful physical-world data is expensive to capture, fragmented across tasks and difficult to reproduce at scale. Text and image model developers could train on vast quantities of existing internet material. Robot developers need demonstrations of hands, tools and bodies interacting with specific objects under real-world conditions.

Microagi's website says Shift can turn records of physical work into training-ready data without teleoperation or interruptions to a customer's production line. The company has not published enough technical detail to compare the resulting datasets with robot demonstrations collected through teleoperation, motion capture or dedicated training facilities.

A large check for an integration-heavy model

The new capital will fund additional Atlas deployments, engineering hires in Munich, Zurich, London and Istanbul, and computing capacity, according to microagi and redalpine's legal adviser. Microagi also operates a research center near ETH Zurich and has maintained operations in New York.

The investor group is financing a specific view of the robotics market: factories can buy capable hardware and access increasingly sophisticated models, yet still lack the data and integration work required to make those systems dependable on a particular line. Atlas packages that work into a platform supported by forward deployment.

For Kilic, the funding buys the time, hardware and compute needed to prove that process can scale across factories. The decisive metric will be the number of deployments that continue operating after engineers leave the site, not the size of the seed round or the volume of recorded demonstrations.