Ubitium Raises $14 Million to Turn Working Silicon Into a Universal Processor for Physical AI

Most autonomous machines still carry three separate brains: a digital signal processor for sensing, a neural processing unit for AI inference, and a CPU for control. A German startup called Ubitium raised $14 million on Tuesday to replace that stack with a single reconfigurable processor — and unlike most pitches in this category, it already has working silicon to show for it.

The seed round was led by Ibex Investors, with participation from OpenOcean and existing investors KBC Focus Fund, Runa Capital and Inflection, along with a personal investment from Florian Seibel, co-founder and co-CEO of European defense technology company Quantum Systems. It brings Ubitium's total seed funding to $18 million since the company was founded in 2024 in Düsseldorf.

The company's argument is that physical AI is a shifting workload, not a static one. A drone, a warehouse robot or an inspection vehicle has to perceive its surroundings, decide and act continuously, often simultaneously, on battery power and in real time — and the balance between those three demands changes as the machine operates. Ubitium's architecture is a single array of programmable processing elements that reallocates resources at runtime across signal processing, AI and control, paired with unified memory and standard RISC-V development tools so manufacturers get one processor and one toolchain instead of several.

That thesis is already proven in silicon, the company says. Its first chip, the UB108, packs 256 processing elements on Samsung's 8nm process and was taped out roughly a year after founding; Ubitium says it has demonstrated signal processing, AI inference and real-time control running on the same reconfigurable array. "Physical AI machines bring a whole new set of computing challenges, and the inference isn't the hardest one," said CEO and co-founder Hyun Shin Cho. "Sensing the real world and controlling the machine in it are just as hard, and today those still run on separate chips. We built one Physical AI processor that does all three, proved it in silicon in a year, and this funding turns it into a product."

The next step is commercialization. Ubitium plans to tape out its first commercial processor, the UB410 with 4,096 processing elements, and deliver evaluation units in 2027. The initial customers are building physical AI's sensing layer — high-resolution radar and audio systems, where demanding signal processing and AI already run side by side and today's embedded chips sell in volume. Autonomous systems that need the complete perceive-think-act loop, including drones, are the longer-term target.

The team's credentials are unusual for a company this size. CTO and co-founder Martin Vorbach has spent more than three decades on reconfigurable computing; Ubitium holds 17 granted patents on its core technology, its staff carry more than 250 career patents and more than 350 peer-reviewed publications, and alumni lists include Intel, Nvidia, Apple and Texas Instruments. Seibel's interest is practical: an autonomous machine has to keep working when communications are unavailable, and today that means integrating "a stack of chips that were not designed to work together," each costing weight, power and integration time.

The market Ubitium is aiming at is large and still undefined. McKinsey's base case puts general-purpose robotics at $370 billion by 2040, and every robotics and drone developer currently absorbs the cost of a multi-chip stack that was never designed as a system. Ubitium's performance claims are self-reported and have not been independently benchmarked, and a reconfigurable architecture must still prove it can match specialized silicon on efficiency at production scale. But the direction of the bet is notable: while data-center AI races toward ever more specialized accelerators, physical AI is arguing the opposite case — that the edge needs one adaptable processor, not three specialized ones.