Singapore’s biological data center prototype is a hybrid computing system: living, stem-cell-derived human cortical neurons sit on electrode-equipped silicon chips, while conventional electronics and software send signals to and read activity from the cells. It is not a facility made entirely from biological material, nor has it been shown to replace modern GPU infrastructure. 3428

Who developed and operates it?

The project brings together three partners:

• NUS Medicine, the Yong Loo Lin School of Medicine at the National University of Singapore, provides the research setting at the NUS Life Sciences Institute.
• Cortical Labs, a Melbourne-based biological-computing company, supplies the CL1 biological computers and neuron-on-chip technology.
• DayOne, a Singapore-headquartered data-center developer and operator, provides the infrastructure and operates the deployment. 3611

The partners describe the installation as the world’s first independently operated biologically integrated server rack, although that “world-first” characterization is a claim made by the project partners and repeated in secondary coverage. 31623

When and where was it launched?

The prototype is housed at the NUS Life Sciences Institute in Singapore. NUS Medicine, DayOne and Cortical Labs showcased the working system to more than 80 guests from academia, industry and the technology sector on August 6, 2026. The public announcement of the launch followed on August 17, 2026. 346

Some reporting gives a different operational date, saying the setup went live on July 16. That date is not consistent with the partners’ August showcase and launch announcements, so the clearest documented milestones are the August 6 demonstration and August 17 announcement. 13

What is inside the 20-unit rack?

The rack contains 20 Cortical Labs CL1 biological computing units. Reports estimate approximately 16 million living human neurons across the system, or about 800,000 neurons per unit if distributed evenly. 257

Each CL1 combines:

1. A culture of human neurons derived from stem cells.
2. A silicon chip fitted with a microelectrode array.
3. Electronics that stimulate the neurons and record their electrical activity.
4. Software that translates between digital inputs, electrical signals and measured neural responses. 21226

The silicon hardware does not simply imitate the neurons. Instead, it provides the interface and control system while the living network supplies the biological component that can form and adapt connections. This makes the CL1 a hybrid platform rather than a purely biological computer.

How are the neurons kept alive?

Living-cell computing requires laboratory support that conventional servers do not. The Straits Times reports that technicians feed the cultures a mixture of sugar, micronutrients and pH buffers every three days, while a gas mixer supplies carbon dioxide, oxygen and nitrogen. 13

Those details illustrate the central trade-off: the neurons may offer biological adaptability, but they require sterile handling, environmental control and ongoing monitoring. The supplied reporting does not establish the full maintenance schedule, operating cost or replacement procedure for every unit in the Singapore rack.

How much power does it use?

Cortical Labs puts the power draw of an individual CL1 at approximately 25 watts. The complete 20-unit rack has been reported at roughly 800 to 1,000 watts. 2

That figure may be attractive for specific adaptive or neuroscience workloads, but it is not a complete energy-efficiency comparison. A fair comparison with GPUs would need to measure the same task, model, throughput, latency, accuracy, cooling requirements, availability and total facility energy use. The available sources do not provide that like-for-like benchmark, so the prototype should not be described as generally more efficient than GPUs.

How long can the biological system operate?

The evidence is less settled on lifespan. Some reporting says CL1 neuron cultures can remain viable for up to roughly six months, but that figure should be treated as a reported platform limit rather than a demonstrated service-life result for the entire Singapore rack. 1017

A GPU and a living-neuron unit also fail in fundamentally different ways. GPUs can typically be maintained or replaced as electronic components, while biological units require continued culture support and eventual replenishment or replacement of the cell substrate. The supplied evidence does not establish a precise replacement interval, uptime record or total cost of ownership for the Singapore installation.

What does it have to do with DishBrain?

The Singapore rack follows the same broad idea associated with Cortical Labs’ earlier DishBrain work: connect cultured neurons to an electrode array, place them in a closed digital environment, and provide feedback based on their activity. Background reporting identifies a 2021 demonstration in which cultured neurons interacted with a simplified Pong task. 2021

The important distinction is scale and setting. DishBrain was a laboratory experiment involving a neural culture and a game-like feedback loop. Singapore’s installation places the neuron–silicon interface into a 20-unit rack designed for continued research and infrastructure testing. The supplied launch sources do not provide a new Pong result or a direct performance comparison between DishBrain and the Singapore system.

What could researchers use it for?

The immediate purpose is research rather than general-purpose cloud computing. The partners are testing whether biological computing can operate in a data-center-style environment and support neuroscience-related investigation. 36

Potential research directions include:

• Studying how human neural networks learn and adapt.
• Modelling disease-related changes in neural systems.
• Testing how drugs affect human-derived neural cultures.
• Exploring biological components for low-power, adaptive AI systems.
• Connecting biological computation to conventional software and infrastructure.

These are possible or proposed applications, not established production workloads. The sources do not show that the Singapore rack is already delivering commercial AI services or outperforming silicon accelerators on a defined task.

Are larger deployments planned?

The Singapore prototype is part of a broader effort by Cortical Labs and DayOne to move biological computing from laboratory equipment toward data-center infrastructure. Reporting also identifies a biological data center in Melbourne and the Singapore collaboration as parallel developments. 1

However, the supplied evidence does not confirm a specific expansion quantity, construction timetable or committed commercial rollout. One report describes the Singapore deployment as a first step toward putting biological computers inside a commercial DayOne data center, but that remains a stated direction rather than evidence of a completed commercial deployment. 24

What must be solved before biological computing scales?

Repeatable performance

Living cultures can vary between batches and change over time. Operators would need standardized cell production, predictable behavior, stable programming methods and reproducible results across units.

Reliability and maintenance

A commercial service would need continuous sterile media management, temperature and gas control, health monitoring, fault isolation and replacement procedures. It would also need to maintain service when individual biological units degrade or fail.

Credible benchmarking

Power consumption alone cannot establish an advantage over GPUs. Biological systems would need to demonstrate measurable benefits on defined workloads across accuracy, latency, throughput, energy, cooling, uptime and total cost.

Manufacturing and integration

Scaling would require reliable production of large numbers of neuron-electrode units, along with integration into conventional networks, storage, scheduling systems and security controls.

Economics

The business case must account for stem-cell sourcing, culture media, laboratory automation, technicians, quality control, environmental systems and biological replacement. A lower chip-level wattage would not necessarily mean a lower total operating cost.

Governance and ethics

Because the platform uses human-derived neural cells, commercial deployment would also require clear rules for cell provenance, consent, biosafety, ethics oversight and the management of increasingly complex neural cultures.

The bottom line

Singapore’s installation is best understood as a 20-unit biological computing research rack containing roughly 16 million living human neurons, not as a conventional data center powered solely by brain cells. NUS Medicine hosts the prototype, Cortical Labs provides the CL1 technology and DayOne supplies and operates the data-center infrastructure. 236

Its significance is that biological computing has been placed into a live server-rack environment. Its limitations are equally important: no supplied evidence demonstrates superiority over GPUs, and the practical questions of neuron lifespan, maintenance, repeatability, uptime and cost remain open. For now, the prototype is a test of whether living neural networks can become a useful complement to silicon—not proof that they are ready to replace it. 234

Singapore’s biological data center prototype is a hybrid computing system: living, stem-cell-derived human cortical neurons sit on electrode-equipped silicon chips, while conventional electronics and software send signals to and read activity from the cells. It is not a facility made entirely from biological material, nor has it been shown to replace modern GPU infrastructure. 3428

Who developed and operates it?

The project brings together three partners:

• NUS Medicine, the Yong Loo Lin School of Medicine at the National University of Singapore, provides the research setting at the NUS Life Sciences Institute.
• Cortical Labs, a Melbourne-based biological-computing company, supplies the CL1 biological computers and neuron-on-chip technology.
• DayOne, a Singapore-headquartered data-center developer and operator, provides the infrastructure and operates the deployment. 3611

The partners describe the installation as the world’s first independently operated biologically integrated server rack, although that “world-first” characterization is a claim made by the project partners and repeated in secondary coverage. 31623

When and where was it launched?

The prototype is housed at the NUS Life Sciences Institute in Singapore. NUS Medicine, DayOne and Cortical Labs showcased the working system to more than 80 guests from academia, industry and the technology sector on August 6, 2026. The public announcement of the launch followed on August 17, 2026. 346

Some reporting gives a different operational date, saying the setup went live on July 16. That date is not consistent with the partners’ August showcase and launch announcements, so the clearest documented milestones are the August 6 demonstration and August 17 announcement. 13

What is inside the 20-unit rack?

The rack contains 20 Cortical Labs CL1 biological computing units. Reports estimate approximately 16 million living human neurons across the system, or about 800,000 neurons per unit if distributed evenly. 257

Each CL1 combines:

1. A culture of human neurons derived from stem cells.
2. A silicon chip fitted with a microelectrode array.
3. Electronics that stimulate the neurons and record their electrical activity.
4. Software that translates between digital inputs, electrical signals and measured neural responses. 21226

The silicon hardware does not simply imitate the neurons. Instead, it provides the interface and control system while the living network supplies the biological component that can form and adapt connections. This makes the CL1 a hybrid platform rather than a purely biological computer.

How are the neurons kept alive?

Living-cell computing requires laboratory support that conventional servers do not. The Straits Times reports that technicians feed the cultures a mixture of sugar, micronutrients and pH buffers every three days, while a gas mixer supplies carbon dioxide, oxygen and nitrogen. 13

Those details illustrate the central trade-off: the neurons may offer biological adaptability, but they require sterile handling, environmental control and ongoing monitoring. The supplied reporting does not establish the full maintenance schedule, operating cost or replacement procedure for every unit in the Singapore rack.

How much power does it use?

Cortical Labs puts the power draw of an individual CL1 at approximately 25 watts. The complete 20-unit rack has been reported at roughly 800 to 1,000 watts. 2

That figure may be attractive for specific adaptive or neuroscience workloads, but it is not a complete energy-efficiency comparison. A fair comparison with GPUs would need to measure the same task, model, throughput, latency, accuracy, cooling requirements, availability and total facility energy use. The available sources do not provide that like-for-like benchmark, so the prototype should not be described as generally more efficient than GPUs.

How long can the biological system operate?

The evidence is less settled on lifespan. Some reporting says CL1 neuron cultures can remain viable for up to roughly six months, but that figure should be treated as a reported platform limit rather than a demonstrated service-life result for the entire Singapore rack. 1017

A GPU and a living-neuron unit also fail in fundamentally different ways. GPUs can typically be maintained or replaced as electronic components, while biological units require continued culture support and eventual replenishment or replacement of the cell substrate. The supplied evidence does not establish a precise replacement interval, uptime record or total cost of ownership for the Singapore installation.

What does it have to do with DishBrain?

The Singapore rack follows the same broad idea associated with Cortical Labs’ earlier DishBrain work: connect cultured neurons to an electrode array, place them in a closed digital environment, and provide feedback based on their activity. Background reporting identifies a 2021 demonstration in which cultured neurons interacted with a simplified Pong task. 2021

The important distinction is scale and setting. DishBrain was a laboratory experiment involving a neural culture and a game-like feedback loop. Singapore’s installation places the neuron–silicon interface into a 20-unit rack designed for continued research and infrastructure testing. The supplied launch sources do not provide a new Pong result or a direct performance comparison between DishBrain and the Singapore system.

What could researchers use it for?

The immediate purpose is research rather than general-purpose cloud computing. The partners are testing whether biological computing can operate in a data-center-style environment and support neuroscience-related investigation. 36

Potential research directions include:

• Studying how human neural networks learn and adapt.
• Modelling disease-related changes in neural systems.
• Testing how drugs affect human-derived neural cultures.
• Exploring biological components for low-power, adaptive AI systems.
• Connecting biological computation to conventional software and infrastructure.

These are possible or proposed applications, not established production workloads. The sources do not show that the Singapore rack is already delivering commercial AI services or outperforming silicon accelerators on a defined task.

Are larger deployments planned?

The Singapore prototype is part of a broader effort by Cortical Labs and DayOne to move biological computing from laboratory equipment toward data-center infrastructure. Reporting also identifies a biological data center in Melbourne and the Singapore collaboration as parallel developments. 1

However, the supplied evidence does not confirm a specific expansion quantity, construction timetable or committed commercial rollout. One report describes the Singapore deployment as a first step toward putting biological computers inside a commercial DayOne data center, but that remains a stated direction rather than evidence of a completed commercial deployment. 24

What must be solved before biological computing scales?

Repeatable performance

Living cultures can vary between batches and change over time. Operators would need standardized cell production, predictable behavior, stable programming methods and reproducible results across units.

Reliability and maintenance

A commercial service would need continuous sterile media management, temperature and gas control, health monitoring, fault isolation and replacement procedures. It would also need to maintain service when individual biological units degrade or fail.

Credible benchmarking

Power consumption alone cannot establish an advantage over GPUs. Biological systems would need to demonstrate measurable benefits on defined workloads across accuracy, latency, throughput, energy, cooling, uptime and total cost.

Manufacturing and integration

Scaling would require reliable production of large numbers of neuron-electrode units, along with integration into conventional networks, storage, scheduling systems and security controls.

Economics

The business case must account for stem-cell sourcing, culture media, laboratory automation, technicians, quality control, environmental systems and biological replacement. A lower chip-level wattage would not necessarily mean a lower total operating cost.

Governance and ethics

Because the platform uses human-derived neural cells, commercial deployment would also require clear rules for cell provenance, consent, biosafety, ethics oversight and the management of increasingly complex neural cultures.

The bottom line

Singapore’s installation is best understood as a 20-unit biological computing research rack containing roughly 16 million living human neurons, not as a conventional data center powered solely by brain cells. NUS Medicine hosts the prototype, Cortical Labs provides the CL1 technology and DayOne supplies and operates the data-center infrastructure. 236

Its significance is that biological computing has been placed into a live server-rack environment. Its limitations are equally important: no supplied evidence demonstrates superiority over GPUs, and the practical questions of neuron lifespan, maintenance, repeatability, uptime and cost remain open. For now, the prototype is a test of whether living neural networks can become a useful complement to silicon—not proof that they are ready to replace it. 234