World’s first data center rack with living neurons unveiled

What if a server rack were made up not of chips, but of living human neurons? That is exactly what the National University of Singapore, data center operator DayOne, and the Australian company Cortical Labs have built. The prototype contains human neurons that were grown from stem cells in a laboratory. According to the parties involved, this form of biological computing could be of interest for AI applications.

The living neurons grow on the CL1, a chip from Cortical Labs equipped with tiny electrodes. These electrodes can send electrical signals to the neurons and then read the cells’ activity. In total, the prototype contains twenty of these units.

Each CL1 contains approximately 800,000 neurons. The entire rack thus contains about 16 million living neurons. That is roughly equal to the estimated number of neurons in a frog’s brain. By comparison, the human brain is estimated to contain about 86 billion neurons.

Computer communicates with living cells

The chips are connected to an operating system developed by Cortical Labs that allows researchers to communicate with the neurons. Information is sent to the cells via electrical stimuli. The neurons’ response can then be read and used to perform a task or simulate a specific environment.

Researchers can also control the neural networks via an API and read their activity. This allows the system to behave like a kind of biological computer.

There is, however, one major limitation to using living cells: they eventually die. The CL1 system provides the neurons with the necessary nutrients and other conditions to keep them alive. According to Cortical Labs, the neurons can function for about six months. After that, they must be replaced.

Energy consumption

The researchers see several potential advantages of biological neural networks for AI. For example, living neurons can adapt to changing conditions and learn through interaction with their environment. According to Cortical Labs, they can also learn from relatively little information.

But one of the main reasons to look into biological computers is energy consumption. The human brain can process complex information using relatively little energy. Cultured neural networks could also be much more energy-efficient than traditional computers.

Cortical Labs states that a CL1 consumes about 25 watts. A complete rack, including the system that keeps the neurons alive, would use about 800 to 1,000 watts. A traditional server rack can easily consume several kilowatts. For AI systems, power consumption can be much higher: specialized AI racks can use more than 100 kilowatts.

No proof yet that it’s faster

However, that does not mean the biological rack is already an alternative to conventional AI servers. It is still unclear how much computational work the system can actually perform.

No benchmarks have yet been published that allow the performance of the biological rack to be compared with that of regular servers or AI systems. It’s also unclear which specific AI tasks the system can perform more efficiently.

For now, the prototype is therefore primarily a demonstration of a new way to use biological neural networks as computers. The coming years will show whether living neurons can indeed become a practical and energy-efficient complement to traditional chips.