Sereact’s $110M Bet: The AI Brain Powering the Future of Robotics

Building hardware is brutal, but giving that hardware a brain is the hardest problem in tech right now. Most robotics startups spend their entire existence in a sterile lab. They film a great demo. They put it on the internet. And then their systems completely break down the second they hit a messy, unpredictable warehouse floor. The reality is, production environments do not care about your demo. Things break. Lighting changes. Products arrive in the wrong boxes. Sereact understood this from the start. They chose to solve the messy problems. So, it makes total sense why the market is aggressively backing them right now.

Sereact Raises $110M in Series B Funding

You do not raise a massive Series B in this climate unless you have something that actually works in the wild. Sereact just locked down 110 million dollars. Headline led the round. They brought in new investors like Bullhound Capital, Daphni, and Felix Capital. Bullhound manages over 1 billion Euros and has backed massive winners like Spotify and Revolut. And the existing believers came back for more. Air Street Capital, Creandum, and Point Nine all doubled down. They also have former Formula 1 world champion Nico Rosberg backing them.

This is a massive war chest. It is a four-times step up from their 25 million Euro Series. A just fifteen months ago. We are talking about serious acceleration. They started with a 5 million dollar seed round in 2023, and now they are holding over 140 million dollars in total funding. The valuation is kept quiet, but the intent is loud. They are using this capital to scale their new robotic brain and to push hard into the United States. Venture capitalists poured over 27.6 billion dollars into robotics in 2025 alone because they know this is the next massive wave. But Bullhound Capital made Sereact the very first investment of their Fund VII for a specific reason. The founders, Ralf Gulde and Marc Tuscher, built a software operating system that retrofits directly into the massive fleets of industrial robots that already exist. It is a pure software play.

Introducing Cortex 2.0: The AI Brain for Robots

Traditional robots are completely blind. They follow rigid, programmed sequences. If a box is an inch to the left, the robot crashes. Sereact decided to fix the brain instead of building another metal arm. The goal of this new funding is to scale Cortex 2.0. This is their next-generation vision language action model.

They did not just arrive at Cortex 2.0 overnight. It took grinding. They built Cortex 1.5 to handle failure recovery by letting a human operator give a quick remote correction, which the system then absorbed to update its local policy. Then came Cortex 1.6, which introduced the Process Reward Operator. This module extracted learning signals from the daily operational process itself, learning from sliding objects or collision risks. Now, Cortex 2.0 takes that foundation and adds a world model. It thinks first. Then it acts. The system generates candidate future trajectories based on the current state. It runs these candidates against a learned model of physics. The robot literally predicts the results, scoring them for stability, risk, and efficiency. It only executes the absolute best plan.

How Robots Predict the Future Before Acting

Most reactive policies out there operate on a try and see basis. If they drop an item, they just try the exact same motion again. That compounds the failure and causes massive bottlenecks. Sereact forces the shift to plan and try.

The robot dreams in latent space. It evaluates potential futures in its visual latent space before the physical arm even moves. It runs candidate actions against a learned model of physics, and our Process Reward Operator evaluates the predicted final state. It predicts the probability of a failure event occurring, penalizing paths that lead to high-speed contact or surface scraping. It anticipates if an object will slip or if a collision is about to happen. We tested this against state of the art baselines like diffusion policies on incredibly hard tasks. Take sorting highly reflective metal screws scattered on a table. Or unpacking a shoebox, where the robot has to open the lid without tearing the cardboard, remove the paper, and extract the shoes. Cortex 2.0 destroyed the baselines because it evaluates the consequences of its actions before committing to them. Small errors do not compound. The robot avoids the bad branches. It is predictive intelligence at its finest.

One AI Software for Any Robot Hardware

Hardware is becoming a commodity. Building custom metal for every task is a terrible business model. Sereact recognized that the actual value is in the intelligence layer. So they built a hardware-agnostic brain.

Cortex 2.0 plans in visual latent space. Pixel data encodes rules about objects and motion that transfer perfectly across different physical bodies. The exact same brain runs a single-arm picking cell. It runs a dual arm returns station. It even runs on the new wave of bipedal humanoid robots. The system uses an action mapping module to translate these high level visual plans into the specific joint commands needed by whatever physical machine it is controlling. You can deploy this software onto Universal Robots, gantry systems, or cutting edge humanoids without retraining the model from scratch. This universal approach is exactly why they are positioned to capture the intelligence market as humanoid production ramps up globally.

1 Billion Real-World Picks and Counting

The reality is that synthetic lab data looks great in a pitch meeting, but it dies on the factory floor. You cannot simulate the weird reflections of a plastic wrapper or the exact friction of a crushed cardboard box. Sereact took the hard road. They put their robots into live fulfillment centers.

They created a massive data flywheel. To date, their systems have completed over 1 billion real production picks. The company has built the largest real-world robotics dataset on earth, holding 507 million samples across 6.1 petabytes of data. Every time a robot picks an item, drops an item, or needs a recovery, that data is captured continuously at 30 Hertz. They record the camera observations, the robot state, and the gripper force feedback. This data flows back to continuously retrain the centralized model. And it works. It really works. They have pushed their autonomy rate to the point where human intervention is only needed once in every 53,000 picks. Competitors are raising billions to train models on video games and staged lab demos. Sereact is training on the night shift, during peak hours, handling messy items that no simulator could ever dream up.

Sereact Expands Operations to the US Market

Armed with 110 million dollars, you have to expand. Europe is a great proving ground, but the United States is the target. Sereact is officially crossing the Atlantic. They are opening their first US office in Boston, Massachusetts.

Boston is the absolute center of gravity for robotics engineering. They are hitting the ground running and actively hiring local commercial, engineering, and application staff. They need aggressive business development managers with at least five years of experience who can handle account based market development. We are talking about building top of funnel pipeline across retail, third party logistics, and heavy manufacturing. The US market requires discipline. It requires multi channel prospecting and deep CRM knowledge using tools like HubSpot, Salesforce, and Apollo. The sales cycles for enterprise automation are brutal and complex. But Sereact is showing up with a product that can be deployed within a single day to deliver immediate cost savings. That is a massive competitive advantage when pitching to executives who are tired of long, expensive integration projects.

Partnering with Major Brands Like BMW and Rohlik

You do not hit a billion picks by running a pilot program in a basement. Sereact systems are live in over 200 production deployments. These are heavy duty, critical operations. The software is currently powering machines for massive automotive manufacturers. BMW, Mercedes Benz, and Daimler Truck all trust Sereact to handle their components. When an automaker lets your AI control a robot on their floor, your technology is legitimate.

But it goes way beyond automotive. They are deep into logistics and e-commerce. They are powering operations for PepsiCo, Austrian Post, bol, and Active Ants. And they just announced a massive rollout with the Rohlik Group. Rohlik is a major player in European online groceries. They are deploying an initial fleet of 24 AI enabled robots across fulfillment centers in Berlin and Vienna. This will quickly scale into Frankfurt and could hit triple digits soon. These robots are integrated into Rohlik’s Veloq platform. They are handling both chilled and ambient grocery products at an industrial scale. Multi temperature picking is an absolute nightmare to automate due to high volume demands and varying product fragilities. But Cortex 2.0 handles it.

The era of blind, stupid robots is over. We are putting true intelligence into the physical world. And with a massive data moat and the capital to scale, Sereact is leading the charge.