Cortical Labs says it has built what may be the world’s first biological data centers, swapping traditional silicon-only compute for living human brain cells integrated into chips. The idea is a radical alternative to today’s GPU-heavy AI infrastructure: instead of relying on power-hungry hardware designed for brute-force processing, these systems use lab-grown neurons that naturally adapt and reorganize themselves in response to input.
At the core of the concept is the CL1, a “wetware” biological computer first revealed at MWC 2025. Each CL1 unit contains about 800,000 neurons grown from human stem cells. Those neurons sit on silicon and are kept alive in a carefully controlled environment. A nutrient-rich solution sustains the cells, while temperature, gas levels, and filtration are regulated to manage waste and maintain stable conditions—allowing the neurons to remain viable for up to six months. A multielectrode array sends electrical signals to the neural tissue and records the responses, effectively acting as the interface between living biology and computing tasks.
Cortical Labs positions power efficiency as one of the biggest advantages. A single CL1 chip reportedly draws as little as 30 watts, and an entire rack of units is said to consume roughly 850 to 1,000 watts. The company contrasts that with the enormous electricity demand associated with modern AI data centers and their accelerator-based hardware. As energy costs climb and grid capacity becomes a real constraint in many regions, the pitch is simple: wetware computing could offer a much more sustainable path for certain kinds of workloads.
What makes the approach especially intriguing is how the “training” works. Instead of being trained like conventional machine learning models, the neurons adapt on their own, rewiring in response to stimuli much like biological brains do. Earlier experiments from the company showed a prototype learning to play simple games such as Pong. Now, Cortical Labs claims the latest CL1 system can handle more complex gameplay, including navigating a game like Doom on its own—meant to demonstrate how these living neural networks can learn behaviors through interaction rather than standard AI training pipelines.
The startup calls the broader concept Synthetic Biological Intelligence (SBI). It runs through a proprietary Biological Intelligence Operating System (biOS), which is designed to translate tasks into signals the neurons can respond to, then interpret the neurons’ output back into usable results. To make the platform accessible beyond its own lab, Cortical Labs has also launched remote access through the Cortical Cloud, offering what it describes as Wetware-as-a-Service. Researchers can rent access at around $300 per unit per week to deploy experiments directly onto living neural networks and measure performance remotely.
For now, Cortical Labs has opened a facility in Melbourne as a proof of concept. A much larger biological data center is also in development in Singapore through a partnership with local provider DayOne Data Centers. While the company acknowledges its neuron layers are still extremely simple compared to the full 3D complexity of a real human brain, the project highlights a fascinating direction: computation based on neurons rather than transistors, aimed at exploring new kinds of adaptive processing with potentially far lower energy costs.
