This Swiss Startup's Robots Could Save AI Data Centers From a Dangerous Future

Jan de Vries · 2026-08-04

### The Problem That's Growing Faster Than We Can Staff It AI data centers are scaling at a breakneck pace. The demand for computing power is exploding, and the physical infrastructure that makes it all possible is being pushed to its absolute limits. But here's the catch: the people who can safely maintain these facilities are becoming scarce. It's a bottleneck that threatens to slow down the entire industry. Enter Exclaim Robotics, a Zürich-based startup that just stepped out of stealth mode with a $4.95 million pre-Seed round co-led by Founderful and Playfair. The company's mission? Build autonomous robots that can handle the dangerous, repetitive maintenance work inside AI data centers. It's a bold idea, and the timing couldn't be more critical. ### Why Human Maintenance Is Becoming Too Dangerous To understand why this matters, you need to look at what's changing inside these facilities. The next generation of AI data centers will distribute power at 800 volts DC directly inside the rack. That's not just a technical upgrade; it's a fundamental shift in risk. At that voltage, even a simple task like opening a rack to replace a failed drive carries a serious arc-flash hazard. A single mistake could be lethal. Helen Oleynikova, CEO and Founder of Exclaim Robotics, puts it in stark terms: "The speed of innovation in data centre infrastructure is reminiscent of the space race: power density is pushing the whole industry to the edge of the laws of physics. Liquid cooling, 800V DC across the whole facility, possibly superconducting bus bars cooled with liquid nitrogen. Maintaining that by hand gets more dangerous every year, and that's the part robots should take." ### Meet the Team Behind the Robots Oleynikova isn't new to this game. She spent 15 years working in robotics at Willow Garage, Google, Microsoft, NVIDIA, and ETH Zürich, and holds a PhD in mapping and planning for cluttered, unstructured environments. She also co-created nvblox, a 3D mapping library now used by robotics teams worldwide. She's joined by founding engineer Patrick Pfreundschuh, who also holds a PhD from ETH Zürich. He created the BIEVR-LIO LiDAR-inertial odometry system and was part of the winning team at the DARPA Subterranean Challenge. Two more founding engineers are set to join in 2027. This is a team with serious credentials. ### What These Robots Will Actually Do So, what does a data center robot actually do? The company is starting with the highest-volume break-fix tasks: replacing and cleaning optics and SFPs, handling copper network cabling, and swapping out failed drives. These are the mundane but essential jobs that keep the lights on. The robot needs to navigate the data hall on its own, position itself at a rack, and work the front face from the floor to the top rail. Crucially, it has to do this without any changes to the building. No new rails, no special markers, just adapt to the existing environment. ### Training in Simulation Before Reality One of the smartest parts of Exclaim's approach is how they train these robots. They start in simulation, where a robot can practice a task thousands of times more than is possible with real hardware. Only after the virtual reps are mastered does the robot move to real racks with real sensors, repeating the work until it's reliable enough to trust. But the company is realistic about the challenges. "The harder problem is the environment rather than the motion: racks are dense with cabling, and a robot has to understand what it is looking at before it can safely touch anything," the startup noted. "Where its confidence in a task is low, it will hand off to a remote operator rather than attempt it." ### The Bigger Picture and What's Next This isn't just about fixing a few servers. The shift toward 800V DC power is going to change everything about how data center infrastructure gets run. Antonia Albert, Principal at Founderful, sums it up well: "AI data centres are scaling faster than anyone can staff or safely maintain them, and the 800V transition will change everything about how that infrastructure gets run. Helen and her team from ETH Zürich, NVIDIA, Microsoft, and Google are the people to solve it." Exclaim plans to use this capital to build its first prototypes, acquire robotics hardware, and grow its engineering team. They're currently hiring robotics and hardware engineers in Zürich. Longer-term, the goal is to apply the same capability wherever infrastructure maintenance is hazardous, remote, or hard to reach, with high-voltage and energy infrastructure as the next target. It's an ambitious vision, but with the right team and the growing urgency of the problem, it's one that could reshape how we think about data center maintenance for years to come.