Cortical Labs Built a Computer Out of Human Brain Cells

It's not science fiction anymore. Cortical Labs, an Australian biotech startup, has built and is now selling what it calls the world's first code-deployable biological computer — a machine that runs on actual human neurons.

The device is called the CL1. Inside it: 800,000 lab-grown human brain cells cultivated on a silicon chip, kept alive for up to six months by an onboard life-support system, and capable of learning and adapting to solve real problems in real time.

The company is pricing the first 115 units at $35,000 each (or $20,000 in 30-unit server racks). They're also offering cloud access at $300 per week per unit, so researchers who can't afford the hardware can still rent time on Cortical's biological neural networks.

This is the moment biocomputing stops being a lab curiosity and becomes a product you can actually buy.

Why This Matters

Conventional AI runs on silicon. It's fast, it's power-hungry, and it operates on fixed logic gates. A biological neural network, by contrast, is adaptive by default. Neurons learn. They form new connections. They respond to their environment. What digital AI systems spend enormous computational resources trying to emulate — genuine learning and flexibility — biological neurons do naturally, as a side effect of existing.

Karl Friston, a theoretical neuroscientist at University College London, put it this way: "The real gift of this technology is not to computer science. Rather, it's an enabling technology that allows scientists to perform experiments on a little synthetic brain."

That's the key insight. The CL1 isn't positioned as a replacement for GPUs or CPUs. It's a research tool. A way to study how neurons process information, test drug compounds on actual human brain tissue, and explore questions about cognition and learning that silicon-based systems can't easily answer.

The Demo That Changed the Conversation

In 2021, Cortical Labs released DishBrain, an early biocomputer that successfully learned to play Pong. It took 18 months. It was impressive for the time, but Pong is simple — a 2D game with minimal variables.

This year, they leveled up. The CL1 learned to play Doom.

Doom is harder. The game has visual complexity, multiple enemies, and requires real-time decision-making. To make it work, the neurons had to "see" what a human player sees. Engineers solved this by converting visual information into electrical stimulation patterns that the neurons could recognize.

The kicker: an independent developer with minimal biocomputing experience completed the integration in about a week, using nothing but Python and the CL1's new programming interface.

The biocomputer isn't winning Doom tournaments. It plays better than random, but it loses frequently. That's not the point. The point is that it learned. It adapted. It got better. And it did so faster than comparable machine learning systems would have.

The Real Applications

Drug testing and toxicology are the obvious near-term use cases. Right now, pharmaceutical companies rely on animal testing or silicon-based simulations to understand how compounds affect neural tissue. Neither is perfect. Animal models don't always translate to humans. Simulations are abstractions.

A living human neural network on a chip? That's actual human tissue responding to actual compounds. No animals harmed. More relevant data. Faster iteration.

Beyond that, the applications get speculative but not unreasonable. Robotics. Biocomputers controlling robotic systems that need to learn and adapt in real time. Personalized medicine. Neural networks grown from a patient's own cells, used to test treatments before they're administered. Brain-computer interfaces. Assistive devices that can learn from user behavior.

The company is also exploring what it calls the Minimal Viable Brain (MVB) — essentially, the smallest, simplest neural system capable of complex information processing. If they can figure that out, they can scale it. Multiple CL1 units networked together. A biological data center.

The Elephant in the Room

Yes, there are ethical questions. These are human neurons. Lab-grown, not extracted from living people, but still derived from human tissue. Are there consciousness implications? Could these organoids develop some form of sentience?

Cortical Labs' answer is straightforward: regulatory compliance and bioethics oversight will be integral to commercialization. They're working with institutional review boards. They're not pretending the questions don't exist.

For now, the neurons in the CL1 are simple enough that consciousness seems unlikely. But as the technology scales and the networks get more complex, these questions will get harder to ignore.

The Bigger Picture

We're at an inflection point. Silicon-based AI is hitting scaling limits. Power consumption is becoming a constraint. The energy cost of training large models is becoming unsustainable. Meanwhile, the human brain runs on about 20 watts.

Biocomputing won't replace silicon. But it might complement it. Hybrid systems that use biological components for tasks that require genuine learning and adaptation, and silicon for tasks that require raw speed and precision.

Cortical Labs is not alone in this space. FinalSpark in Switzerland is building similar systems. Johns Hopkins researchers are exploring organoid computing. The field is moving fast.

The CL1 is the first commercial product. It won't be the last. And unlike most biotech breakthroughs, this one is available for purchase right now, at a price that's high but not astronomical for a research institution.

The future of computing might not be silicon at all. It might be wet.