AnswersPublished21 sources
Singapore’s 16 Million-Neuron Biological Data Center Explained
Singapore’s prototype is a 20 unit rack containing roughly 16 million lab grown human cortical neurons. The CL1 units connect living neurons to silicon microelectrode arrays: software sends electrical inputs, records neural spikes, and uses the responses for adaptive computing experiments.
Published byEdited with GPT-5.6 LunaImages generated with GPT Image 1.5
Research answer
Key takeaways
- Singapore’s prototype is a 20 unit rack containing roughly 16 million lab grown human cortical neurons.
- The CL1 units connect living neurons to silicon microelectrode arrays: software sends electrical inputs, records neural spikes, and uses the responses for adaptive computing experiments.
- The system is designed primarily for research into disease, drugs, learning, and specialised AI.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What is Singapore’s biological data center prototype unveiled by the National University of Singapore in July and publicly showcased in Augu. Article summary: Singapore’s prototype is a small “wetware” data-centre rack: 20 Cortical Labs CL1 biological computers at NUS, combining lab-grown human cortical neurons with conventional silicon electronics. It was activated in July 20. Topic tags: general, general web, user generated, government, education. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, wat
Singapore’s new biological data center is best understood as a wetware research rack, not a conventional supercomputer. Hosted at the National University of Singapore’s Life Sciences Institute, it combines 20 Cortical Labs CL1 biological computers with laboratory-grown human cortical neurons and conventional silicon electronics. The system reportedly went live on July 16, 2026, and NUS Medicine publicly showcased it to more than 80 guests on August 6. 4
12
The headline figure is roughly 16 million living neurons across the rack. Each CL1 uses a neuronal culture connected to a silicon microelectrode array, allowing electronics to stimulate the cells and record their electrical activity. Cortical Labs’ software stack, including the biOS interface, is intended to configure experiments, deliver inputs, interpret neural responses, and connect the biological processor to ordinary computing systems. 4
7
How the biological data center works
The system does not replace silicon with a miniature human brain. Instead, it creates a feedback loop between living neurons and digital hardware:
- Software encodes information as electrical stimulation.
- Electrodes deliver that stimulation to selected parts of the neuronal culture.
- The neurons respond through patterns of electrical spiking.
- The microelectrode array records those signals.
- Software converts the activity into an output or uses it to adjust the next input.
This arrangement gives the neurons a role similar to an adaptive biological processor. The silicon hardware handles the interfaces, control, recording, and surrounding infrastructure; the living network supplies its ability to respond and adapt within the task it is given.
Who built and operates it?
The prototype is a collaboration among three organizations:
- NUS Medicine and the NUS Life Sciences Institute cultivate the stem-cell-derived neuronal cultures and lead the scientific research.
- DayOne provides the data-center setting and operational infrastructure.
- Cortical Labs supplies the CL1 biological-computing units, their neuron-support systems, and the associated software stack.
1
4
That division of responsibilities is important. This is not simply a rack of commercial servers placed in a university laboratory. It requires both data-center operations and the biological expertise needed to grow, monitor, feed, and replace living cultures.
How the neurons are kept alive
Unlike a silicon processor, a CL1 requires continuous biological maintenance. Its life-support equipment is designed to keep the neuronal culture viable for up to about six months. Nutrient medium must be replenished regularly; reporting on the Singapore installation says this occurs approximately every three days. Eventually, the culture must be replaced rather than operated indefinitely. 7
12
39
That biological lifespan creates a practical distinction between wetware and conventional computing. Replacing a culture is not equivalent to swapping a failed circuit board. It involves producing or preparing new cells, maintaining sterile conditions, validating the culture, and potentially retraining the new network for the intended task.
Is it more energy-efficient than an Nvidia H100?
Cortical Labs puts the power draw of an individual CL1 at roughly 25 watts, while estimates for the complete 20-unit rack—including supporting equipment—range from 800 to 1,000 watts. 1
For a rough electrical comparison, an Nvidia H100 GPU is commonly described in the available coverage as having a maximum thermal-design-power rating of about 700 watts. On that narrow measure, the entire prototype rack is in the range of roughly one to one-and-a-half H100s. 1
51
But wattage alone does not establish efficiency. The Singapore project has not published an apples-to-apples comparison covering throughput, accuracy, latency, training time, operating cost, or useful work completed against an H100 or a GPU cluster. The rack therefore should not be described as a proven lower-energy replacement for mainstream AI infrastructure. Its potential advantage is more specific: biological networks may be useful for specialised adaptive tasks that do not map neatly onto conventional, highly repetitive GPU workloads. 1
6
The DishBrain experiment behind the project
The Singapore installation builds on Cortical Labs’ earlier DishBrain research. In the 2022 experiment, human induced-pluripotent-stem-cell-derived neurons and rodent cortical neurons were grown on high-density multielectrode arrays and connected to a simplified digital version of Pong. Electrodes supplied information about the game, while the neural activity controlled the paddle. 17
19
20
The experiment used closed-loop feedback: the cultures received different electrical feedback depending on how the system performed. Researchers reported improvements in gameplay over time, while control conditions without the same feedback did not show the same learning effect. 17
18
19
DishBrain did not demonstrate a general-purpose biological computer. It showed that cultured neural networks could be embedded in a constrained digital environment and adapt their activity to a task. The Singapore rack applies that basic pattern—electrical input, living neural response, recorded output—across a more deployable set of CL1 modules.
What could the system be used for?
The most credible near-term applications are research-oriented:
- Disease modelling: studying how neural systems respond under different biological conditions.
- Drug and toxicity testing: observing how neuronal cultures react to candidate compounds.
- Neuroscience: investigating learning, adaptation, and the behavior of living neural networks.
- Specialised computing: exploring adaptive systems for selected AI or robotics problems where limited training data or changing environments matter.
6
These uses are different from running a large language model or rendering conventional AI workloads at GPU scale. Biological computing is being presented as a possible complement to silicon, not as a universal substitute for it.
Could it expand from 20 units to 1,000?
The partners have discussed a possible expansion to as many as 1,000 CL1 units, but that remains conditional on regulatory approval, safety testing, energy-efficiency results, and the ability to operate the system reliably at a much larger scale. 12
A thousand-unit installation would require more than adding 980 machines to a rack. Operators would need reproducible cell production, automated feeding and replacement, sterile handling, continuous quality control, dependable networking, fault tolerance, and procedures for dealing with cultures that age or behave differently from one another.
The unresolved questions
Reliability and reproducibility
Living cultures vary between batches, mature over time, and eventually degrade. Researchers would need consistent methods for measuring performance and determining when a culture is no longer suitable. If a culture is replaced, any task-specific adaptation may not transfer automatically to the new one.
Economics
Electricity is only one part of the cost. Biological computing also requires nutrient media, sterile consumables, laboratory space, skilled technicians, quality control, cell replacement, and potentially downtime. Those costs could offset some of the apparent power advantage.
Scalability
The 20-unit rack is a prototype in a research environment. Operating 1,000 units would test whether the biological supply chain, maintenance process, software controls, and data-center infrastructure can scale together.
Real-world performance
The available evidence shows that cultured neurons can adapt in constrained closed-loop experiments. It does not yet show that the CL1 system beats GPUs on commercially important workloads, or that its total cost of ownership is lower. That performance evidence—not the novelty of putting neurons in a rack—will determine whether biological computing becomes a practical infrastructure category.
The bottom line
Singapore’s prototype is a significant demonstration of biologically integrated computing: 20 CL1 units, roughly 16 million living human cortical neurons, silicon microelectrode arrays, and software working as one experimental system. Its strongest promise lies in neuroscience, drug research, and specialised adaptive-computing experiments.
For now, it is more accurate to call the installation a platform for testing what living neural networks can do than to call it a replacement for GPU data centers. The next milestones are likely to be reproducibility, measurable task performance, operating cost, and whether the system can scale without losing the biological properties that make it interesting in the first place.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Singapore’s 16 Million-Neuron Biological Data Center Explained"?
Singapore’s prototype is a 20 unit rack containing roughly 16 million lab grown human cortical neurons.
What are the key points to validate first?
Singapore’s prototype is a 20 unit rack containing roughly 16 million lab grown human cortical neurons. The CL1 units connect living neurons to silicon microelectrode arrays: software sends electrical inputs, records neural spikes, and uses the responses for adaptive computing experiments.
What should I do next in practice?
The system is designed primarily for research into disease, drugs, learning, and specialised AI.
Sources
straitstimes.comThe biological data centre in Singapore using brain cells
newsbytesapp.comSingapore unveils biological data center prototype with 16 million neurons