To Scale Cortex 2.0 and Enter the US

Sereact has raised a $110 million Series B round led by Headline, with participation from Bullhound Capital, Daphni, and Felix Capital. Existing investors Air Street Capital, Creandum (lead of Sereact's 2025 Series A), and Point Nine all returned for this round.

The round funds two priorities: scaling Cortex 2.0, the next generation of Sereact's robotic brain, and entering the United States. Sereact opens its first US office in Boston and hires commercial, application, and engineering staff locally.

At a glance:

• $110M Series B, led by Headline
• 200+ Sereact systems live across Europe and with that the most deployed AI picking robot company in the world.
• 1B+ real production picks completed on Cortex
• 1 in ~53,000 picks needs remote human help
• First US office in Boston

Cortex 2.0 - The robotic brain that anticipates outcomes before it moves.

Today's Cortex sees and picks. Cortex 2.0 thinks first, then acts.

Cortex 2.0 augments a vision-language-action (VLA) model with a world model. From the current state, it generates a set of candidate future trajectories, runs them against a learned model of physics and object behavior, and scores each one for stability, risk, and efficiency. The robot commits only to the best-scored branch - and updates the rollout in real time as the scene changes. World models are the next frontier in AI, and most of that work is happening in research labs on synthetic data. Cortex 2.0 is the one trained on more than a billion picks of real production.

The shift is from try-and-see to plan-and-try. Today's reactive policies, when they miss, tend to repeat the same motion and compound the failure. Cortex 2.0 evaluates several outcomes first and rules out the bad ones before the arm moves - which is exactly what's needed for the kind of work where contact matters: assembling a component under tension, placing a windshield wiper without scratching it, kitting parts that have to land in exactly the right orientation for the next station. That's the next market Sereact is going after, and Cortex 2.0 is how it gets there.

Why it generalizes

Cortex 2.0 plans in visual latent space. While joint commands are tied to a specific robot's kinematics, pixels encode regularities about objects, contact, and motion that transfer across embodiments. The same brain runs single-arm picking cells, dual-arm returns stations, humanoid robots, and fixed cells.

Planning compute is tunable per task. More foresight when failure is expensive (parcel packing, kitting, fragile placement) and less when recovery is cheap (a regrasp on a missed pick). Cortex 2.0 spends planning budget where it pays back.

Every pick across every site goes back into the model.

Cortex 2.0 sits on top of an infrastructure Sereact has been building for five years: a closed loop where every robot in production is also a data source, and a single centralized model is continuously retrained and redeployed across the fleet.

Every successful pick, every failure, every recovery is captured with synchronized observations, robot state, gripper force feedback, and outcome - then filtered, prioritized by novelty and uncertainty, and used to update the model. Updated policies pass automated regression checks and roll out to the fleet. The loop closes. Data compounds. Coverage of the long tail expands.

This is what makes the gap structural. Competitors are raising billions to train on simulated data and lab demos. Sereact has spent five years training on real operations. At night, at peak, on the messy items that don't look like anything the robot has seen before. To Sereact's knowledge, no other industrial robotics company outside of self-driving has a learning loop running at this scale.

The model on a Sereact robot today is not the model that was on it last month, and won't be the model on it next month. Every shift moves it forward.

Ralf Gulde (left), Franz Cortex-powered (middle) and Marc Tuscher (right)

Why warehouses first.

Sereact builds AI for robots that work in the physical world. Warehouses were the first deployment because no other environment provides the same combination of data points: billions of real interactions, every object shape imaginable, hard throughput constraints, and consequences when the robot gets it wrong.

Every shift, across every customer site, Cortex learns from real picks on real items in conditions no simulator can reproduce. That data is what moves the model, and the model moves with the robot: mobile, humanoid, fixed cell - whatever comes next.

More than 200 Sereact systems are already live across Europe, making Sereact the most deployed AI picking robot company in the world. Those systems, all running the Cortex brain, have completed over 1 billion real production picks for customers including Active Ants, Austrian Post, BMW, bol., Daimler Truck, DeltiLog, Mercedes-Benz, Monta, MS Direct, PepsiCo and Rohlik Group.

One pick in roughly every 53,000 needs remote human help. Everything else, the robot handles on its own: that ratio is a critical differentiator. Every pick Cortex does goes back into the model. Competitors are raising billions to train on simulated data and lab demos. Sereact has spent five years training on real operations, at night, at peak, on the messy items that don't look like anything the robot has seen before. It's a gap that widens with every shift.

About Sereact

Sereact builds physical AI for warehouses and manufacturing. Its Cortex brain runs across single-arm picking cells, dual-arm returns stations, humanoid robots, and Sereact Lens, a 3D perception system for inventory and quality control. Founded in 2021, headquartered in Stuttgart. Customers include Daimler Truck, Mercedes-Benz, BMW, MS Direct, Active Ants, DeltiLog, Rohlik Group, and Austrian Post. Sereact has raised over $140 million to date.