Biological Data Centers: Human Neurons, AI and Ethics (2026)

Research Review | Updated October 27, 2026

Review type: Narrative evidence review

Scientific focus: Biological computing, cultured human neuronal networks, hybrid silicon-neuron systems, energy efficiency and neuroethics.

Abstract

Background: Biological computing integrates living neuronal cultures with electronic interfaces to investigate whether biological networks can perform selected computational tasks. In August 2026, the National University of Singapore (NUS) Medicine, DayOne and Cortical Labs publicly unveiled a 20-unit biological computing prototype in Singapore.

Objective: To evaluate the scientific basis, demonstrated capabilities, commercialization claims, energy-efficiency proposition and ethical implications of neuron-based computing.

Methods: This narrative review examines institutional announcements, manufacturer information, peer-reviewed research and published ethical guidance available through October 27, 2026. It is not a systematic review, a new experimental study or an independent technical audit.

Findings: Research has demonstrated that cultured neuronal networks can interact with software through electrical stimulation and recording. A 2022 study demonstrated changes in performance in a closed-loop simulated Pong environment. The newer CL1 platform packages related biological-computing concepts into a product intended for research and experimental applications. However, the public evidence reviewed here does not establish general-purpose AI superiority, independently verified commercial-scale energy savings or consciousness in these systems. The reported total number of neurons in the Singapore installation also requires clearer primary-source documentation.

Conclusion: Biological computing is a legitimate emerging research field, but its potential should be separated from demonstrated performance. Transparent benchmarking, independent replication, donor-consent safeguards and proportionate ethical oversight are needed as the technology develops.

Keywords: biological computing; biological data center; cultured human neurons; Cortical Labs CL1; synthetic biological intelligence; AI infrastructure; neuroethics.

1. Introduction

Computing systems have traditionally used semiconductor devices to encode, process and transmit information. Biological computing explores an alternative: using living cells and their electrical activity as part of a computational system.

The concept is not purely speculative. Researchers have already integrated cultured neurons with electrode arrays and software environments, allowing electrical activity to be recorded and feedback to be delivered to the cells. The resulting systems are neither conventional computers nor biological brains in the ordinary sense. They are hybrid arrangements in which living neural networks interact with electronic hardware and software.

The topic received renewed attention when NUS Medicine announced a 20-unit biological data center prototype in Singapore on August 17, 2026. The project involves NUS Medicine, data center operator DayOne and Australian biological-computing company Cortical Labs.

The announcement raises several questions. What does a biological data center actually do? How much of its computing depends on conventional electronics? Are its energy-efficiency claims independently established? Does the presence of living human neurons create evidence of consciousness, or does it raise a different set of ethical questions?

This review addresses those questions using an evidence-first approach.

2. Review methods and evidence appraisal

This is a narrative review of selected primary and secondary sources, including the NUS institutional announcement, the manufacturer's CL1 product information, peer-reviewed neuronal-computing research and published literature on neural-system ethics.

Each major claim is classified according to the kind of evidence available. Institutional confirmation establishes that a deployment was announced; it does not independently validate every performance claim made about it. A peer-reviewed laboratory experiment may establish a narrow capability without demonstrating commercial usefulness or general-purpose intelligence.

No new biological experiments, direct hardware measurements, independent energy tests or systematic meta-analysis were conducted for this review. Conclusions are therefore limited to the sources identified in the references.

3. What is a biological data center?

A biological data center is a computing environment that uses living biological cells as part of its information-processing architecture. In the Singapore project, lab-grown human neurons are integrated with silicon chips, microelectrode arrays and software.

The distinction is important: the neurons do not simply replace every silicon component. Conventional electronics remain essential for delivering electrical stimulation, recording neural responses, running software and connecting the biological system to external equipment.

3.1 The Singapore prototype

According to NUS Medicine's August 2026 announcement, the prototype comprises 20 Cortical Labs CL1 biological computing units deployed as a server rack in a live research environment at the NUS Life Sciences Institute.

The collaboration is intended to investigate possible applications in areas such as neurological research, drug discovery and computing. The announcement describes a prototype and a pathway toward potential expansion, not proof that biological computing has replaced conventional commercial data centers.

3.2 How the system works

The basic architecture can be described in five stages:

Cell preparation: Neurons are cultivated from stem-cell-derived material and maintained under controlled laboratory conditions.
Electrical interface: The neuronal culture is placed on a silicon platform containing electrodes.
Input: Electronics deliver structured electrical stimulation to the culture.
Neural response: The neuronal network generates electrical activity that can be recorded by the electrodes.
Feedback and interpretation: Software interprets the recorded activity and can use it to update the simulated environment or guide subsequent stimulation.

This closed-loop architecture allows researchers to investigate whether neural networks can adapt their activity in response to structured inputs and feedback.

Cortical Labs describes its CL1 biological computer as a programmable system combining real neurons, silicon electronics and a biological-computing software environment. The company's stated capabilities and design objectives should be distinguished from independently replicated performance results.

3.3 How many neurons are in Singapore's installation?

Some media reports have estimated that the 20-unit rack contains approximately 16 million neurons, based on a reported figure of around 800,000 neurons per unit. Other reporting has described individual CL1 units as containing at least 200,000 neurons.

The public NUS announcement confirms the 20-unit configuration but does not publish a measured total neuronal count for the installation. Consequently, the 16-million figure should be presented as a reported estimate rather than a verified census of the cells in the Singapore rack.

The distinction is more than editorial. The number of cells alone does not establish the number of functional connections, computational capacity, learning performance or equivalence to a human brain.

4. What has biological computing demonstrated?

4.1 The DishBrain experiment

A foundational study published in Neuron in 2022 investigated cultured neuronal networks interacting with a simulated version of the arcade game Pong.

In the study, human or mouse cortical neurons were connected to a high-density multielectrode array. Electrical activity from the cells was used to interact with the simulated environment, while feedback was returned to the culture through electrical stimulation.

The researchers reported changes in gameplay performance over time under closed-loop feedback conditions, with effects observed within minutes. Their findings supported the proposition that neuronal cultures can alter their activity in response to a structured input-and-feedback environment.

Primary reference: Kagan BJ, Kitchen AC, Tran NT, et al. In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron. 2022;110(23):3952–3969.e8. doi:10.1016/j.neuron.2022.09.001.

Several qualifications are necessary:

The experiment evaluated a specific biological system and a specific simulated task.
Changes in measurable task performance do not establish human-like understanding, general intelligence or the ability to perform arbitrary computational workloads.
Evidence of adaptive electrical activity does not, by itself, establish subjective experience or consciousness.
The paper included researchers affiliated with Cortical Labs, making independent replication and transparent performance comparisons particularly valuable.

The study provides a scientific basis for further investigation. It should not be interpreted as proof that a small neuronal culture possesses the full cognitive capacities associated with a human brain.

4.2 Biological computing is not the same as artificial general intelligence

The terms biological intelligence, synthetic biological intelligence and artificial intelligence refer to different concepts and should not be treated as interchangeable.

A biological system may exhibit adaptive responses without demonstrating broad competence across unrelated tasks. Similarly, the ability of a system to learn within a controlled environment does not show that it can reason, plan, communicate or generalize with the reliability expected of a general-purpose AI system.

Claims about intelligence should therefore specify the exact task, input conditions, performance metric, comparison baseline and degree of independent replication.

5. Commercialization: what is available and what remains experimental?

Cortical Labs publicly offers information about purchasing its CL1 hardware and accessing biological computing through its cloud platform. This represents a commercial route for researchers and developers to experiment with the technology.

However, the commercial availability of a research platform is not equivalent to evidence of broad industrial adoption or established economic value.

The Singapore installation is a research prototype. According to NUS Medicine, its purpose includes investigating how biological computing might be developed into practical applications. More extensive deployment would require additional technical validation and evidence that the system can deliver useful performance under real-world operating conditions.

The distinction between current capability and proposed application can be summarized as follows:

Research into neural adaptation

Evidence status: Demonstrated in defined laboratory experiments.

Cultured neural networks have been interfaced with software and studied under controlled feedback conditions. Reproducibility and the range of applications remain important questions.

Drug discovery and neuroscience

Evidence status: Plausible research application.

Biological systems may offer useful ways to investigate neuronal responses to compounds and study aspects of neurological function. Their ability to improve drug-development outcomes requires application-specific validation.

Robotics and adaptive control

Evidence status: Experimental potential.

Biological networks may be worth evaluating in environments where feedback, adaptation and limited training data matter. Claims of superiority over conventional machine-learning systems need controlled head-to-head testing.

Cybersecurity and fraud detection

Evidence status: Proposed use case.

These are possible applications discussed by the companies involved. Independent evidence of better detection rates, lower false-positive rates, improved robustness or lower total operating costs is needed before superiority can be established.

6. Energy efficiency: the central claim that needs independent testing

Biological computing has attracted attention partly because living neuronal networks may perform some forms of adaptive information processing with relatively low energy requirements.

NUS Medicine and its project partners have presented the Singapore prototype as a possible lower-power alternative for selected computing applications. This is an important research hypothesis, but it is not yet equivalent to a demonstrated reduction in energy consumption for a defined commercial workload.

A fair comparison requires more than measuring the electrical power drawn by the neuron-containing device.

6.1 What should a credible benchmark measure?

Total system energy: Include electronic interfaces, control computers, environmental regulation, temperature management and other essential support systems.
Energy per useful task: Measure the electricity required to complete a defined task at a specified quality level.
Task performance: Compare the biological system with a suitable conventional CPU, GPU or other baseline performing the same task.
Reliability: Account for variability between neuronal cultures, failed runs, maintenance requirements and changes in performance over time.
Lifetime and replacement: Include the resources needed to prepare, maintain and replace cultures, rather than treating biological material as cost-free infrastructure.
Scale: Test whether any efficiency advantage persists as the system grows beyond a small prototype.

A low-power device could still be inefficient for a particular workload if it completes fewer useful tasks, requires substantial external processing or needs frequent replacement. Conversely, a biological system could prove attractive for a narrow class of adaptive tasks even if it is not competitive for conventional data processing.

The appropriate question is therefore not whether neurons are inherently more efficient than silicon in every context. It is whether a biological-computing system can perform a defined task with better overall energy, performance, reliability or cost characteristics than the best practical alternative.

Future publications should report independent, reproducible measurements that allow researchers to test this proposition quantitatively.

7. Consciousness and the meaning of living neuronal computing

The presence of living human neurons in a computer can understandably provoke questions about consciousness. Those questions deserve careful scientific and ethical treatment, not assumptions in either direction.

7.1 Living neurons do not automatically imply consciousness

The CL1 system described by Cortical Labs contains cultured neurons interfaced with electronics. It is not an intact human brain, and its use does not involve connecting a living person to a computer or creating a complete human nervous system.

Neurons can generate electrical signals, form functional networks and change their activity in response to stimulation. These properties are scientifically significant but do not establish that a culture experiences sensations, has a sense of self or possesses subjective awareness.

The DishBrain paper's title uses the term "sentience," but a measured behavioural or electrophysiological response should not be treated as decisive evidence of subjective experience. A separate question is whether more complex future biological systems might raise different concerns as their architecture and functions evolve.

Consensus literature on nervous-system organoids and assembloids has urged scientists to define terms such as learning, intelligence and sentience carefully and avoid inferring complex mental states solely from cellular activity. Although not every neuronal culture is an organoid, the underlying caution about scientific language is relevant.

See A nomenclature consensus for nervous system organoids and assembloids, published in Nature in 2022.

7.2 Why consciousness research needs proportionate safeguards

The lack of demonstrated consciousness should not be used as a reason to ignore future ethical questions. It should also not be used as a justification for claiming that current devices are conscious.

As biological computing develops, researchers should establish evidence-based oversight procedures that are proportionate to the complexity of the system. Such procedures could include predefined review criteria, transparent reporting of neural architecture and activity, reassessment when capabilities change, and clear responsibility for ethical decisions.

Ethical guidance for human brain organoid research has highlighted issues including cell sourcing, informed consent, commercialization, uncertainty, moral status and responsible innovation. These concerns are relevant to the broader field, although the precise safeguards appropriate for a particular neuronal-computing platform should be determined by its actual design and capabilities.

For additional context, see The ethical landscape of human brain organoids and a mindful innovation framework, published in Nature Reviews Bioengineering in 2024, and the Nuffield Council on Bioethics report on neural organoids, published in May 2026.

8. Human-cell sourcing, consent and research governance

When human-derived cells are used in computational systems, ethical governance must extend beyond the question of consciousness.

Important issues include the provenance of the cell lines, informed consent for the original donation and subsequent uses, appropriate review of research protocols, privacy and handling of associated donor information, commercialization arrangements, and the allocation of responsibility for long-term cell maintenance and disposal.

Where cells originate from reprogrammed donor material, the fact that the material is used to develop a computing platform makes clear documentation of permitted uses particularly important. The applicable consent and oversight requirements depend on the source of the cells, the research design, institutional rules and relevant law.

Responsible development should include the following safeguards:

Traceable provenance: Document the source of the cell lines and the scope of consent for their use.
Independent oversight: Ensure that appropriate institutional ethics and biosafety reviews are in place.
Transparent methods: Publish sufficient information about culture preparation, system design, testing conditions and adverse events for independent assessment.
Proportionate monitoring: Reassess ethical risks if the complexity or capabilities of a system materially change.
Clear accountability: Define who is responsible for cell maintenance, research records, disposal and responses to unexpected findings.

The International Society for Stem Cell Research guidelines provide an important international reference for responsible stem-cell research. Specific requirements should be interpreted in the context of the relevant cell system rather than applied indiscriminately.

9. Limitations of the current evidence

Several limitations prevent strong conclusions about the future of biological data centers.

First, independent performance data are limited. Institutional announcements and product information establish the existence of a system and describe its intended capabilities. They do not, on their own, demonstrate that it outperforms conventional computing.

Second, the application evidence is narrow. Experimental adaptation in a controlled task cannot establish superior performance in drug discovery, robotics, cybersecurity or other complex applications without separate validation.

Third, energy comparisons require common benchmarks. Device-level power claims cannot be compared fairly with GPU or server energy consumption unless the useful work performed, task quality and supporting infrastructure are also measured.

Fourth, biological variability introduces operational challenges. Cultured neuronal systems must remain viable and function within controlled environmental conditions. Reproducibility, longevity, maintenance burden and scaling remain important evaluation criteria.

Fifth, the neuronal count is not a performance metric. The exact total reported for the Singapore prototype is not established in the institutional announcement. Even a confirmed cell count would not, on its own, demonstrate a particular level of intelligence, consciousness or computational capability.

Finally, ethical conclusions must remain proportionate to the evidence. The presence of living cells justifies clear governance, but it does not establish that a system possesses subjective awareness or should automatically be treated as equivalent to a human being.

10. Priorities for future research

The next phase of the field should prioritize verifiable results over headline claims. Several research priorities would make the evidence more useful to scientists, investors, regulators and the public.

Independent replication: Replicate core findings in laboratories not directly affiliated with the platform manufacturer.
Standardized benchmarks: Establish comparable tasks, performance metrics, baselines and reporting standards for biological-computing systems.
Full energy accounting: Measure total system electricity per useful task and report uncertainty, reliability and maintenance requirements.
Reproducibility: Quantify differences between cultures, units, experimental runs and operating periods.
Application-specific evaluation: Demonstrate a measurable benefit in a real research or industrial workflow before claiming commercial superiority.
Proportionate ethical governance: Develop transparent processes for consent, cell sourcing, oversight and reassessment as biological systems become more complex.

These priorities would help identify where biological computing offers a meaningful advantage and where conventional approaches remain more practical.

11. Conclusion

Biological computing represents a genuine and scientifically interesting development in the integration of living neuronal networks with electronic systems. The Singapore prototype established by NUS Medicine, DayOne and Cortical Labs demonstrates that this approach is progressing beyond isolated laboratory demonstrations toward an organized research infrastructure.

Yet the significance of the achievement should not be confused with claims that remain unproven. The existence of living neuronal networks does not demonstrate consciousness. Commercial availability does not establish broad market adoption. And a claim of improved energy efficiency requires independently verified measurements against suitable conventional systems.

The most credible assessment is therefore balanced: biological computing may become valuable for selected tasks involving adaptation and biological investigation, but its advantages will have to be established experimentally. The field should advance through transparent methods, independent benchmarking and proportionate ethical oversight rather than sensationalism or technological hype.

For now, the appropriate scientific position is to recognize the milestone, test its claims and remain precise about what the evidence actually demonstrates.

Frequently asked questions

What is a biological data center?

A biological data center uses living biological cells as part of a computing system. The Singapore prototype integrates cultured human neurons with silicon electronics, electrode arrays and software.

Does Singapore's biological data center contain a living human brain?

No. The system uses cultured neurons integrated with electronic hardware. It is not an intact human brain, and the presence of living neurons does not establish human-like consciousness.

Does the Singapore installation definitely contain 16 million neurons?

The figure has appeared in media reports, but the NUS institutional announcement confirms the 20-unit installation without publishing a verified total neuronal count. The 16-million figure should therefore be treated as a reported estimate, not a confirmed measurement.

Can biological computers replace GPUs?

There is insufficient independent evidence to conclude that biological computers can generally replace GPUs. Their potential usefulness should be evaluated for specific tasks using common performance, reliability, cost and energy benchmarks.

Are biological computers conscious?

There is no established evidence that the system described here is conscious. Adaptive electrical activity and task-related changes are not, by themselves, proof of subjective experience.

Are biological data centers more energy-efficient?

They are being investigated partly for their potential energy advantages, but a fair conclusion requires measurements of total system energy per useful task, including the equipment and environmental controls needed to maintain the living cells.

What are the main ethical concerns?

Important issues include donor consent, cell-line provenance, appropriate research oversight, commercialization, transparency, responsible handling of biological material and proportionate reassessment of future neural systems as their complexity changes.

References

National University of Singapore, Yong Loo Lin School of Medicine. NUS Medicine, DayOne and Cortical Labs Unveil Biological Data Center Prototype in Singapore. August 17, 2026. Institutional announcement.
Kagan BJ, Kitchen AC, Tran NT, et al. In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron. 2022;110(23):3952–3969.e8. https://doi.org/10.1016/j.neuron.2022.09.001.
Cortical Labs. CL1: Biological Computer. Manufacturer information and stated product capabilities. Official product page.
Lee LY. Forget silicon chip servers, Singapore's newest data centre needs to be fed. The Straits Times. August 15, 2026; updated August 16, 2026. News report.
Pașca SP, et al. A nomenclature consensus for nervous system organoids and assembloids. Nature. 2022. Nature.
Shlobin NA, Savulescu J, Baum ML. The ethical landscape of human brain organoids and a mindful innovation framework. Nature Reviews Bioengineering. 2024;2:785–796. Nature Reviews Bioengineering.
Nuffield Council on Bioethics. Neural organoids: Ethical and governance considerations. May 12, 2026. Report and recommendations.
International Society for Stem Cell Research. Guidelines for Stem Cell Research and Clinical Translation. 2025 update. ISSCR guidelines.
Focal Points. Big Tech Has Officially Turned Data Centers Into Human–Silicon Hybrid Monstrosities. 2026.

Disclosure: This article is an editorial narrative review of publicly available sources. It is not an independent laboratory validation, a formal systematic review or a determination of the consciousness status of any biological-computing system. Manufacturer statements are identified as such and should not be interpreted as independently verified results.