Physical AI
Physical AI is intelligence in machines that move, sense and act: vehicles, robots, drones, instruments, medical devices. CyberSwarm builds the processor it runs on, one that computes inside its own memory, so a machine can perceive and decide by itself on a budget of milliwatts.
Backed by
The gap
The last decade of artificial intelligence happened in data centres, and it worked because a data centre has unlimited power, active cooling and no deadline that a few hundred milliseconds cannot meet. A machine in the physical world has none of those. It has a battery, a sealed enclosure, and a control loop that has to close before the thing it is steering hits something.
So the intelligence stays in the building and the machine keeps a thin connection back to it. That is workable for a thermostat. It is not workable for a car changing lane, a surgical instrument, a satellite over an ocean, or a robot on a factory floor where the network is one more thing that can fail.
The blocker is not the models. Those already exist and they are good enough. It is that no processor built to date can run them inside the power and time budget a physical machine actually has. That is the gap CyberSwarm was built to close.
“AI adoption worldwide is at five or six percent, and we already do not have enough energy. By the time we reach fifty percent, there is simply nowhere left to take that energy from.”
Mihai Raneti · Founder & CEO, CyberSwarm
Led by Falanga Invest
Device, array and system
Nine years of device work
What changes
Stated as outcomes rather than architecture. The engineering behind each one is on the technology page.
01
Perception and decision both complete on the device. A product works the same in a tunnel, in a basement, at sea and in a factory that will not put its line on the public internet.
02
Computation happens where the data already sits, so the energy that used to go into moving numbers is simply not spent. That is the difference between a feature that fits in a battery-powered product and one that does not.
03
A decision reaches the actuator in microseconds rather than after a round trip. For anything that steers, brakes, grips or corrects, that is not a nicety. It is the whole requirement.
04
A camera that recognises a driver without sending the driver anywhere changes what is legally and commercially possible. Privacy stops being a policy and becomes a property of the hardware.
See it work
This is the engine, live. Hover a sensor channel to feed it a signal, then activate it to send the whole input through at once and watch the answer appear along the bottom. What takes a conventional processor millions of operations happens here in a single pulse, because the components are doing the arithmetic themselves.
Why us
Everyone else is making the existing architecture more efficient. We removed the part that costs the energy: the memory and the arithmetic are the same physical component. That is not a tuning gain, and it is not something a competitor reaches by iterating.
The portfolio starts at how a resistive state is written into a thin film and runs up to the full network. Anyone building this way meets our filings at the device layer, which is the layer that is hardest to design around.
Partners license the architecture into their own silicon or take it as a module. That fits how automotive, aerospace and medical companies already buy, and it means one design win can carry a platform rather than a product.
Work is spread across many small arrays, so damage costs precision in one region instead of taking the system down. In the sectors we sell into, that behaviour is a requirement before it is an advantage.
Where it goes
From a driver-monitoring camera to an orbital imaging payload, the same constraint keeps reappearing: fixed power, no network, hard deadline.
Automotive
Sensor fusion, driver monitoring and predictive diagnostics that run inside the ECU on a power budget a vehicle can actually spare, with a wake-up that costs nothing, because the weights never left the array.
Aerospace & Defence
On-board processing for airborne and orbital systems where the downlink is narrow, contested or absent, and where a single-event upset must degrade the answer, not end the mission.
Robotics & Physical AI
Reflex-rate perception and control for manipulators, mobile robots and drones, where a decision that arrives 200 ms late is not a decision at all.
Healthcare & Medical Devices
Wearable and implantable devices that classify physiological signals on the device, where privacy is a property of the architecture rather than a clause in a policy document.
Industrial & Energy
Retrofit intelligence for factory floors, grid assets and remote infrastructure. Nodes that run for years on a harvested or battery supply and report an event, not a stream.
Consumer & IoT
Wake-word, presence, gesture and context awareness that stay on continuously without the power draw or the privacy cost. Because the array is transparent, it can sit behind the display.
For partners
Evaluation, integration and production paths, with the toolchain, the reference hardware and the engineers behind them.
For investors
Nine years of device work, a $50 million Series A, and the market thesis behind it, including an honest account of the risk that remains.
Newsroom
19 May 2026 · Funding
Forbes covers the Series A and the shift from research into commercial neuromorphic computing, with the full Raneti statement on why the architecture sits between classical and quantum.
18 May 2026 · Feature
The longest feature on the company to date: why running AI in the circuit rather than in software is the answer to the energy problem, and what that means for humanoid robots and autonomous vehicles that cannot depend on a data centre.
18 May 2026 · Coverage
Reports the first commercial agreements signed and an advanced partnership in the automotive sector, alongside Raneti’s description of intelligence emerging from the continuous interaction between hardware and its environment.
Whether you are evaluating the architecture for a product or the company for a portfolio, the conversation starts the same way.