Stuttgart's Sereact raises US$110 million led by Headline to sell one robot brain for any machine, starting with the US

Sereact, a Stuttgart company that sells robot intelligence rather than robots, said on 27 April 2026 that it had raised US$110 million in a Series B round led by Headline. Its chief technology officer put the strategy in one line in the announcement: the company does not build robots and does not sell services, it ships a model meant to run on any robot. The number behind that claim is what made investors pay. Sereact said its systems have completed more than 1 billion real production picks, and that only about one pick in 53,000 needs a remote human to step in.

New investors Bullhound Capital, Daphni and Felix Capital joined the round alongside returning backers Air Street Capital, Creandum and Point Nine. Sereact said the round brought its total funding to more than US$140 million. It did not disclose a valuation, and none appears here. On 20 July the company said Zalando had joined the Series B as a strategic investor, lifting the round to US$116 million and total funding to US$145 million. Zalando's cheque was not disclosed; by subtraction of the two round totals it is roughly US$6 million. The question the money has to answer is whether a European software company can win the brain of the robot when hardware makers and US model labs want the same seat.

What the money buys: Cortex 2.0 and a Boston office

Sereact said the capital would scale Cortex 2.0, the new version of its robot model, and fund its entry into the United States, starting with a first US office in Boston staffed with local commercial, application and engineering hires. The company later said that office had opened at Hood Park in Charlestown. In the same announcement it said more than 200 Sereact systems were live across Europe, and it described itself as the most deployed AI picking robot company in the world, a claim it did not benchmark against named rivals.

The customer list explains why the round was priced as a scale-up rather than a science project. Sereact lists Active Ants, Austrian Post, BMW, bol., Daimler Truck, DeltiLog, Mercedes-Benz, Monta, MS Direct, PepsiCo and Rohlik Group among the companies it works with. That mix of e-commerce fulfilment, parcel, grocery and automotive plants matters, because each site produces different objects, packaging and failure cases that feed the same model.

A picking specialist that has raised 22 times its seed in under three years

Sereact was founded in 2021 by Ralf Gulde, its chief executive, and Marc Tuscher, its chief technology officer. In August 2023 it raised US$5 million in a seed round led by Point Nine and Air Street Capital, when it was marketing a pick-and-pack product it called PickGPT and quoting speeds of 1,500 picks per hour. In January 2025 it raised EUR 25 million in a Series A led by Creandum, with angels including former Formula 1 world champion Nico Rosberg. The Series B is 22 times the seed in dollar terms and came about 15 months after the Series A,.

How a world model changes the pick

Cortex 2.0 pairs a vision-language-action model, which turns camera images and an instruction into robot motions, with a world model, a learned simulator of how objects behave. Sereact said the system generates several candidate future trajectories, runs each against its learned model of physics and object behaviour, and scores them for stability, risk and efficiency before the arm moves. Tuscher described this as the robot dreaming in latent space. Because the planning happens in a visual latent space rather than in one robot's joint coordinates, the company said the same model runs single-arm picking cells, dual-arm returns stations and humanoid robots.

The company kept shipping after the round. On 15 July it described Cortex 2.5, which learns a new short task from two or three video demonstrations shown at run time, without retraining. Sereact reported a 92.6% success rate for that in-context approach against 97.6% for a fully retrained model, and said the result covers only short, single pick-and-place tasks close to its training domain. On 1 June it announced a global partnership with Veloq, the Prague-based grocery fulfilment platform that grew out of Rohlik Group and is an AutoStore Global Partner, with joint expansion into the US named as a priority.

Competing for the robot's brain

Sereact's pitch puts it against two groups at once. The first is the general robot-brain labs. Skild AI raised US$1.4 billion in January 2026 at a valuation above US$14 billion, about 12.7 times Sereact's round, and Physical Intelligence is building similar cross-robot models in the US. The second is warehouse picking specialists such as Poland's Nomagic, which raised US$44 million in a Series B in February 2025, Covariant in the US and Mujin in Japan. In Germany the money has gone mostly to hardware: RobCo raised US$100 million in January 2026 for modular factory robots it rents out, and NEURA Robotics closed a Series C in June. Bullhound's Per Roman framed Sereact's edge as software that retrofits into the large fleet of industrial robots already in service.

The open questions are commercial. Sereact has not published revenue, unit prices, headcount or the share of its 1 billion picks that come from any one customer. Its intervention figure is self-reported and depends on how a pick and an intervention are counted. And in the US it will meet Amazon's in-house robotics and well-funded local rivals on their home ground.