CyberSwarm Puts a Brain on a Chip - and Skips the Cloud

The setup is the kind of thing that sounds like a metaphor until you look at the electricity bill. Digital chips keep memory in one place and computation in another, and they spend an enormous amount of energy shuttling data between the two. Brains, inconveniently for the semiconductor industry, do not do this. A neuron stores and processes in the same spot, runs on roughly the power of a dim lightbulb, and does the whole thing in parallel. CyberSwarm looked at that arrangement and decided to copy the homework.

The company was founded in 2017 by Mihai Raneti, who is the sort of founder willing to say, on the record, that "the future of AI will require entirely new architectures." Founders say this constantly. The interesting part is that Raneti's company is actually building the silicon to back the sentence, using components called memristors - resistors that remember - which store and compute information simultaneously, the way biological neurons do.

The product line has a name that would be at home in a science-fiction novel and a spec sheet at the same time: the Swarm Nervous System, a neuromorphic engine designed to learn and adapt locally. There is also a neuromorphic System-on-a-Chip aimed specifically at cybersecurity, which secures communication between connected devices without phoning home to a distant server. That last detail is the quiet part. If the data never leaves the device, a whole category of privacy and security problems simply never gets a chance to happen. CyberSwarm has essentially turned "we didn't send your data anywhere" into a feature, which is both good engineering and good marketing.

Geographically, the company is a bit of a magic trick. It is headquartered in San Mateo, California, the appropriate zip code for a deep-tech hardware bet, while its research and engineering roots run through Ploiesti, Romania, and its collaborations reach into Romanian research institutions. The team is about nine people. Nine people, one chip, and a claim that they are building a new computational layer that sits between traditional digital systems and quantum computing. It is the sort of ambition-to-headcount ratio that makes venture capitalists either nervous or interested, and occasionally both.

The money says interested. Draper Associates backed the company early, in 2018, and then did the deeply unglamorous deep-tech thing of waiting. Eight years later, in May 2026, CyberSwarm closed a $50 million Series A led by Falanga Invest Perton. Ian Imrisek, Falanga's CEO, offered the standard-issue conviction quote - that neuromorphic computing "will become one of the defining technological directions of the next decade" - which is what you say when you have just written a check that large.

What the money is actually for is less cinematic and more telling: research, engineering headcount, industry partnerships, and the transition from "we have a compelling architecture" to "you can buy it." CyberSwarm is pointing the proceeds at automotive, aerospace, defense, healthcare, and robotics - which is to say, everywhere you want a device to think for itself without waiting on a network round-trip. Reported revenue sits at roughly $3 million, the kind of figure that matters less as income and more as evidence that someone, somewhere, is already paying.

The honest risk here is time. Neuromorphic computing has been "five years away" for about two decades, and analog computing has a long history of being elegant, efficient, and commercially stubborn. CyberSwarm's counter-argument is that the constraint finally changed: AI got genuinely hungry for power, memristors got real, and a small company decided the timing had lined up. Whether it has is the whole question. But it is a good question, and CyberSwarm has raised $50 million to spend the next few years answering it.