Ubitium raises $14M to bring universal computing to physical AI

Ubitium has pulled in $14 million in new seed funding to push its universal processor for physical AI out of the lab and toward a product. The startup says the round lifts its total seed funding to $18 million since it was founded in 2024. Ibex Investors led the deal, with OpenOcean joining in, along with existing backers KBC Focus Fund, Runa Capital and Inflection. Florian Seibel, who co-founded and co-runs Quantum Systems, also invested personally.

The pitch is simple enough, even if the engineering isn’t: robots, drones, vehicles and industrial machines all need to sense the world, infer what’s happening and react in real time. Today, those jobs are usually split across different chips, which means extra data transfers, more software complexity and more moving parts to manage. Ubitium wants to fold those functions into one reconfigurable architecture instead.

That architecture uses programmable processing elements that can shift compute resources between sensing, AI inference and real-time control depending on what the machine needs at the moment. Ubitium says it pairs that with unified memory and standard RISC-V development tools, so manufacturers can work from a common processor and software stack instead of stitching together several specialized chips. The company argues that should reduce both complexity and cost.

There is at least one piece of silicon behind the story. Ubitium’s first chip, the UB108, has 256 processing elements and was built on Samsung’s 8nm process. The company says it has already shown signal processing, AI inference and real-time control on the same reconfigurable array. Now it is moving on to the UB410, a commercial processor planned with 4,096 processing elements.

Ubitium says the first uses for the UB410 will be high-resolution radar and audio systems, with autonomous systems such as drones also on the list. The funding will help advance that chip toward commercialisation, with tape-out and evaluation units planned for 2027. Hyun Shin Cho, the company’s CEO and co-founder, frames the bet as a practical one: inference is only part of the problem, and the real world still has to be sensed and controlled on separate chips in most systems.