In July 2025, Cortical Labs released something no lab had done before. The CL1 is not a prototype or proof of concept. It is a working biocomputer built from live human neurons. The brain cells are real. Grown from stem cells. Wired to silicon. Kept alive inside a sealed system that regulates heat, oxygen, and waste.
Each unit holds hundreds of thousands of neurons. They spread across a dense mesh of electrodes. They fire signals. They create synapses. They evolve into neural networks that don’t just react but learn.
The CL1 runs on something called the Biological Intelligence Operating System. Developers inject code directly into the neural substrate. The cells respond. They adjust. They build strategies. In test environments, the neurons learned to play Pong with no human help. No scripts. Just stimulus and reward. The entire process took under five minutes.
Power is where this turns heads. GPT-3 needed over 1,200 megawatt-hours to train. The CL1 consumes less than 1,000 watts. That isn’t a minor efficiency boost. It is a step change. Cortical believes biocomputers could slash energy use in AI by factors in the billions. FinalSpark, a Swiss firm, is already renting access to lab-grown neuron clusters for $1,000 a month. Their system uses dopamine and serotonin signals to shape learning.
But biology brings its own limits. Neurons are sensitive to heat. Too much, and they collapse. Cooling must remain steady. Fluid circulation has to stay balanced. Even small fluctuations in the system can degrade the network.
SCIENTISTS PUT HUMAN BRAIN CELLS INTO COMPUTERS… AND THEY’RE THINKING
Real human neurons are now part of actual computers that can learn on their own.
These biocomputers are way more efficient than supercomputers and don’t need much training.
The catch? Brain cells hate heat… pic.twitter.com/sLWtV1kzKK
— Mario Nawfal (@MarioNawfal) July 22, 2025
Labs around the world are already testing what the cells can do. A group in Indiana trained them to recognize human voices using 240 audio samples across eight speakers. After two days, accuracy hit 78%. No graphics processors. Just living cells. In China, researchers used similar systems to run robots. They managed obstacle detection, tracking, and grip control using only organic networks.
The ethical lines are starting to blur. Cortical insists the cells are not conscious. But they react to patterns. They adjust behavior. They seek stability. The firm has protocols in place, but no one agrees where the threshold lies anymore. Sentience was once a clear line. That is changing.
The CL1 starts at $35,000. First units ship this quarter. Cortical is building multi-unit racks for remote access. Four full stacks will be live by year-end. Each will offer cloud access to physical neural systems. Researchers are calling it Wetware-as-a-Service.
The neurons are alive. The systems work. The learning is real. And every cluster still needs a technician watching the monitors.
Sources
https://www.nationalgeographic.com/science/article/brain-cells-organoids-computers-ai-energy
https://newatlas.com/brain/cortical-bioengineered-intelligence/
https://mikekalil.com/blog/cortical-labs-brain-computer/
https://myelectricsparks.com/cortical-labs-cl1-biological-computer-human-brain-cells/