CyberSwarm has raised $55.5 million in total: $50 million in the current round led by the Czech fund Falanga Invest, $1 million from Draper Associates at the start, and $4.5 million in convertible notes. An acquisition approach valuing the company at $110 million was declined.
The team is nine people internally and around fifteen counting the extended group, which includes advisors from Intel. The portfolio stands at more than thirty patents, built up over eight years.
The company began around the memristor, a component whose resistance changes with the charge that has passed through it and which holds that value once the power is gone. A cybersecurity product turned into neural network hardware once it became clear the same component could both store a weight and compute with it.
“We store in the same way that we process,” Raneti says. There is no separate chip doing the arithmetic, because the component holding the value is the component doing the work. Conventional systems handle information one step after another, and nature does not.
For humanoid robots, the plan is to replace the processing side, so the machine gains autonomy, can navigate an unfamiliar space without prior training, and works longer on the same battery. The same argument applies to autonomous vehicles and to defence: the processing happens locally, so the machine is not tied to a connection.
The energy case is the reason any of this matters commercially. “Artificial intelligence adoption sits at five or six percent worldwide and we already do not have enough energy,” Raneti says. At fifty percent adoption the supply does not exist. Transistors, on his account, are switches, and switches are not enough for this work.
“Without new materials and without new hardware, the technology we have now will stagnate,” he says.
There is a constraint that money does not solve. Romania has lost much of the know-how for work of this kind, and the engineers still capable of it are comfortably over fifty five.