DayOne Moves Biological Computing From Lab to Server Rack in Singapore

Inside NUS Medicine’s Life Sciences Institute in Singapore, a server rack runs 20 machines that process information using something conventional computers lack: living human neurons.

The neurons, grown from human stem cells, sit on silicon chips that deliver electrical signals to them and read their responses. Software translates those exchanges into information the computer can use. In effect, the living cells become part of the machine’s information-processing system rather than leaving all the work to silicon.

It is an unusual experiment. It is also moving closer to becoming an infrastructure test.

NUS Medicine, Singapore-headquartered data centre operator DayOne and Melbourne-based Cortical Labs unveiled the 20-unit system on August 17. NUS said the prototype had earlier been showcased on August 6 to more than 80 guests from academia and industry, who viewed a live demonstration of the CL1 units and neural activity. NUS described it as the world’s first independently operated biologically integrated server rack.

The deployment turns a plan announced five months ago into a working research installation. In March, DayOne and Cortical Labs said the NUS prototype would be the first step towards putting biological computers inside a commercial DayOne data centre in Singapore. If the technology clears performance, operational and regulatory hurdles, the partners have outlined a phased expansion that could eventually reach 1,000 units.

That is where the experiment becomes more consequential. Scientists have already shown that cultured neurons can receive information and change their behaviour in response. A data centre operator needs something far less exotic but much harder to deliver—a system that can run consistently, be maintained and eventually scale.

Cortical Labs’ CL1 is built around that challenge. Each unit combines laboratory-grown neurons with conventional electronics. The cells grow on a chip fitted with tiny electrodes that stimulate the neurons and record the electrical signals they produce.

Those signals allow software and the biological network to interact continuously, while an internal life-support system controls temperature, supplies nutrients and removes waste. Cortical says these neurons can remain alive for up to six months.

The technology is commonly called biological computing, or wetware computing. Unlike artificial neural networks, which are mathematical models running on chips, this system uses actual neurons as part of the computing process.

Cortical’s case for doing that rests on an ability biology already possesses: learning from feedback with relatively little energy and data.

Its earlier DishBrain experiment connected human and mouse neurons grown on an electrode array to the video game Pong. Electrical signals told the cells the position of the virtual ball, while neuron activity controlled the paddle. Researchers reported that the neural cultures altered their behaviour as they interacted with the game. The findings were published in Neuron in 2022.

The experiment established that living neural networks could be placed in a closed loop with a digital environment and adapt to what happened inside it. It did not show that biological computers could replace conventional computers.

That distinction matters as Cortical pushes the technology out of the laboratory.

Karl Friston, a theoretical neuroscientist at University College London who has worked with Cortical researchers, told IEEE Spectrum that the CL1’s immediate value may lie less in mainstream computing than in giving scientists a new way to experiment with living neural networks. The platform lets researchers study how neurons respond to stimulation, drugs and changing environments while the cells actively process information.

That makes neuroscience and drug discovery more obvious early markets than running large AI models.

Cortical says researchers can use different cell lines to investigate neurological conditions and measure how treatments change the way neural networks function.

The company is also exploring applications including AI acceleration, robotics and other tasks where rapid adaptation or learning from limited data may eventually offer an advantage. NUS Medicine said the Singapore platform will support research into learning, adaptation, drug discovery and neurological disease.

DayOne’s presence changes the scale of the question.

The company operates conventional digital infrastructure across Asia and Europe. As of February, its portfolio included about 480 MW of data-centre capacity in service or under construction and another 590 MW reserved for future development across Hong Kong, Indonesia, Japan, Malaysia and Singapore, according to Reuters.

It is also preparing to go public. Bloomberg reported on August 11 that DayOne had confidentially filed for a US initial public offering and was considering raising around $5 Bn, with a potential listing as soon as the next quarter. But the details remain subject to change.

For a builder of conventional data centres, interest in neurons stems from the industry’s ultimate constraint: electricity.

Electricity consumption by data centres worldwide is projected to roughly double, from 485 TWh in 2025 to around 950 TWh by 2030, according to the International Energy Agency’s latest 2026 outlook. Electricity use by AI-focussed data centres is expected to triple over the same period. Southeast Asia’s data centre electricity demand is itself expected to more than double by 2030, partly because of the growing cluster around Singapore and southern Malaysia.

Singapore has consequently been selective about adding new data centre capacity. Its second data centre allocation programme, launched in December 2025, made at least 200 MW available, with potentially more capacity tied to new and innovative green-energy pathways.

Biological computing enters that equation with a striking energy proposition.

IEEE Spectrum reported that Cortical’s standard rack configuration, which accommodates 30 CL1 units, consumes about 850 to 1,000 watts. That is dramatically below the power drawn by racks designed for intensive AI computing.

But it is not a like-for-like comparison.

The Singapore prototype contains 20 units, not 30. Moreover, no publicly demonstrated benchmark shows a CL1 rack performing the same commercial AI workload as a GPU rack while using a fraction of the electricity. Lower power draw while performing a different kind of computation does not establish superior computing efficiency.

That gap between promise and proof is precisely what makes the DayOne experiment significant.

The March agreement called for performance and efficiency benchmarking as well as work on governance, biosafety and compliance before larger-scale deployment. But moving the technology into a commercial data centre would raise another set of questions. Can living cells remain stable in an infrastructure environment? Can biological variation be controlled enough for repeatable workloads? And more important, how can a system whose neurons eventually die operate at scale?

Conventional servers are built around predictability. Components can fail, but replacements are manufactured to behave in the same way as before. Biological systems are inherently less uniform, and their behaviour is partly the reason researchers find them interesting in the first place.

Cortical and DayOne, therefore, have to prove more than whether neurons can compute. They have to show whether the properties that make neurons valuable, including their ability to adapt, can coexist with the reliability expected from robust digital infrastructure.

For now, the 20-unit rack at NUS is a research system, not an alternative to the GPU farms powering generative AI. Nor does its relatively low electricity use demonstrate that biological computers can perform today’s AI workloads more efficiently.

But things have changed since the project was announced in March. The idea is no longer confined to a proposal or a single biological computer on a laboratory bench. As many as 20 machines using living neurons are now operating together inside a server rack, and the next question is whether they can make the far harder transition from something scientists experiment with to something a data-centre operator can actually run.