By 
Singapore has unveiled a prototype biological data centre that replaces part of the conventional silicon computing stack with something far less conventional: living human neurons. The project, developed by the National University of Singapore (NUS), data-centre operator DayOne and Australian biological-computing company Cortical Labs, brings a 20-unit rack of biological computers into a live research environment.
Together, the system contains roughly 16 million lab-grown human neurons, with each CL1 unit combining living neural cells with silicon-based electronics. The neurons are derived from stem cells and communicate with the computer through microelectrode arrays, allowing researchers to study how biological networks can process information.
The concept sounds like science fiction, but the objective is fairly practical: explore whether biological systems can handle certain computing workloads while using considerably less energy than conventional AI infrastructure.
A Different Approach to AI Computing
The Singapore installation is being positioned as a new type of computing infrastructure rather than a replacement for conventional data centres. Silicon processors remain far better suited to fast, precise and repeatable calculations, including the workloads behind large language models.
Biological computing takes a different approach. Living neurons naturally form adaptive networks and can learn from relatively small amounts of information. Researchers therefore see potential in workloads where adaptability matters more than raw processing speed.
Cortical Labs has suggested applications including robotics, drug discovery, cybersecurity and fraud detection. The company argues that biological systems could be particularly useful in situations where computers need to respond to changing or unpredictable environments.
That could become relevant as AI moves into the physical world. A humanoid robot navigating a changing home or factory, for example, cannot realistically encounter every possible situation during training. Biological neural networks could eventually offer another way of dealing with those uncertainties.
The Energy Question
Energy consumption is one of the biggest reasons biological computing has attracted attention.
AI data centres require enormous amounts of electricity, particularly as companies deploy increasingly powerful accelerators. By comparison, Cortical Labs says each CL1 consumes around 25 to 30 watts, with a complete 20-unit rack using roughly 800 to 1,000 watts.
That does not mean biological computers are ready to replace GPUs. The neurons require carefully controlled conditions, including nutrients, gases and temperature management. Their biological nature also creates maintenance challenges that silicon hardware does not face.
The cells have a limited operating lifetime, making long-term reliability another major question for the technology.
Singapore’s experiment is therefore less about replacing Nvidia-style computing hardware today and more about exploring what the next generation of computing might look like.
From Laboratory Experiment to Data Centre
What makes the project notable is its move beyond a laboratory demonstration. NUS says the 20-unit CL1 deployment is the world’s first independently operated biologically integrated server rack. The facility gives researchers a controlled environment to investigate biological computing alongside conventional digital infrastructure.
There are still major hurdles before this approach can become commercially competitive. Researchers need meaningful benchmarks, reliable biological maintenance systems and evidence showing where living neurons outperform or complement conventional processors.
That is perhaps the most important point to keep in mind. Having 16 million neurons inside a server rack is an impressive demonstration, but the real test will be what those neurons can accomplish at scale.
If the research succeeds, future data centres may not be built exclusively around silicon. They could combine conventional processors with biological systems, using each where it performs best. Singapore’s prototype is an early experiment in that possibility — and a sign that the search for more efficient AI computing is beginning to look beyond traditional hardware.
Comments