OpenOcean has invested $4 million in Ubitium’s $14 million seed round, led by US-based Ibex Investors with participation from Quantum Systems co-founder Florian Seibel and existing backers KBC Focus Fund, Runa Capital, and Inflection. Ubitium is building a universal RISC-V processor for embedded systems and Physical AI: one chip that does the work of the CPU, MCU, GPU/NPU, DSP, and FPGA that today coexist inside most intelligent devices.
When we share our excitement about Ubitium with fellow investors, the reaction usually lands somewhere between amazement and disbelief: “Wait, one chip can actually do all of that?!” That was my first question too, and many more followed as we investigated who could build it and why the timing is right. Below are the insights that got us over the line.
A monster of a problem
Open up a drone, a conference-room microphone array, or an automotive radar module and you find four or five classes of processor bolted onto one board. A CPU for the operating system, a microcontroller for real-time control, a DSP for signal processing, a GPU or NPU for AI inference, often an FPGA for whatever is left, each from a different vendor with its own toolchain, power budget and software team. The industry has a name for the result: the Frankenstack.
Each of those chips is excellent on its own, but the friction lies in tying them together. It shows up as latency and energy burned on every hop, as headcount duplicated across separate software stacks, as supply risk spread over five sole-source vendors, and as design cycles measured in years. The Frankenstack has gotten uglier over time: a car of the 1970s ran on a couple of processors, whereas a premium car today carries well over a hundred.
The pain is sharpest in signal processing. While the industry pours trillions into AI, legacy applications and Physical AI alike are still bottlenecked by moving data rather than by crunching it. It takes a long chain of transforms to process and act on sensor input, split today across several classes of silicon where every handoff compounds latency and power usage.
One array, many jobs
The UB410 will be a single chip built around sixteen Universal Processing Arrays on only 25 mm² of silicon. Any array can reconfigure at runtime into an out-of-order CPU, a dataflow engine, a GPU-like unit, or a real-time controller. Different arrays can run different modes at the same time on shared memory, which enables the chip to run Linux on its own alongside signal processing and inference workloads, unlike earlier reconfigurable architectures that required a separate host CPU.
Beyond easing the Frankenstack's toll across the product life cycle, the chip should also deliver large cost and power efficiency gains. These result from keeping a small piece of silicon on a mature 8nm process busy and from reducing data movement. In some AI and signal processing workloads, those gains could reach one to three orders of magnitude versus incumbents while matching or even surpassing them on absolute performance.
So, what's the catch? Customer interest is already strong; the hard part is technical. It requires both hardware and compiler breakthroughs, delivered at a pace that convinces design partners and future investors alike that a European venture can own the next frontier in embedded computing. The first prototype is back from Samsung Foundry and running, and this round pays for the production-ready silicon.
The all-star team
Hyun Shin Cho, the CEO, lived the problem from the customer side. He spent thirteen years at ThyssenKrupp, ending as Head of Digital Transformation at TK Elevator, where he built a planet-scale IoT platform with Microsoft and a cognitive elevator with NVIDIA. Chairman Peter Weber made his name on the vendor side, with eight years at Intel Europe and turnarounds at Telefunken Semiconductors and Dialog as the latter became one of Apple’s key suppliers.
The concentration of reconfigurable computing talent is genuinely unique. CTO Martin Vorbach has pioneered the field for three decades with more than 200 architectural patents, and his inventions power today’s leading FPGAs. Around him is a group that has worked on exactly this problem before, with strong academic and industry credentials from Intel, Texas Instruments, Apple, NVIDIA, and the Karlsruhe Institute of Technology, among other centers of excellence.
The bet
We see this playing out across two main markets. The Frankenstack is the one that exists today across professional audio, automotive sensing, industrial and robotics, and defense. Then comes the growth from Physical AI: as perception, inference, and control move on-device and have to fuse in real time, the workload mix becomes less predictable, which is exactly the case for a future-proof universal chip. Regulation and sovereignty pull the same way, from European defense procurement favoring ITAR-free silicon to driver-assistance mandates in new cars.
While the total size of the opportunity is attractive, what matters more today are the winnable segments. We see the TAM reaching tens of billions of dollars by 2030, serviced gradually by Ubitium from today’s beachhead in real-time signal processing, where it has the strongest right to win. This will provide the first commercial validation of the technology and, from 2028, help finance both further expansion and product upgrades.
Today’s compiler tailwinds gave us the final push. Historically, chip startups have died on software more often than on silicon, because customers are reluctant to adopt a new toolchain. Maturing open-source LLVM and RISC-V ecosystems, plus LLM-assisted code tooling, shrink that problem to something a seed-stage company can attack. This remains a formidable challenge, but one we trust the team to tackle.
Delighted to be backing Hyun Shin, Martin, Peter, and the team.
About Ubitium
Ubitium is a German fabless semiconductor company developing a universal, runtime-reconfigurable RISC-V microprocessor built to cut computing complexity, cost, and power usage across embedded systems and Physical AI applications. Founded in 2024 by Hyun Shin Cho and Martin Vorbach, a reconfigurable computing pioneer whose patents were licensed by major semiconductor companies, Ubitium is headquartered in Düsseldorf with engineering in Karlsruhe. For more, visit ubitium.com.
