Singapore’s CL1 Biological Data Center: How Living Neurons Could Change AI Infrastructure

Cortical Labs says the CL1 integrates recording, application execution and life-support functions in the device, without requiring external compute for those functions. Its controlled environment is designed to keep the neuronal cultures viable for up to six months.

Why the neurons matter

Neural networks can change their activity in response to experience and feedback. That plasticity is the central reason biological computing attracts interest: for some tasks, living networks may adapt to new inputs without the same volume of data or conventional retraining used by digital systems.

That does not mean the CL1 is generally intelligent or conscious. The platform demonstrates a way to interface living neurons with software; it does not establish human-like understanding, sentience or broad intelligence. Those claims would require evidence beyond the system’s current research demonstrations.

How CL1 builds on DishBrain

The Singapore prototype follows Cortical Labs’ 2022 DishBrain research. In that experiment, researchers grew human stem-cell-derived and rodent neurons on high-density microelectrode arrays, then connected the cultures to a simplified version of the video game Pong. Electrical signals represented information about the game, while feedback helped the neural cultures modify their activity. The study reported apparent learning during real-time gameplay.

DishBrain was a laboratory demonstration. CL1 packages the same broad idea into a more complete, remotely operable system with integrated life support and software tools. That shift—from an experimental dish to a commercially offered computing unit—is the significance of the CL1 platform, even though the practical uses are still being evaluated.

Energy use versus conventional AI hardware

Cortical Labs has described a CL1 as using roughly 30 watts. A 30-unit rack has been reported at approximately 850 to 1,000 watts.

Those figures make the system’s potential energy advantage attractive, particularly as AI infrastructure increases electricity and cooling demand. But power consumption alone does not establish better efficiency. A meaningful comparison would need to measure the same task, quality level, throughput, software overhead and operating conditions against GPUs or other data-center accelerators.

The available reporting does not provide a public, independently benchmarked task-for-task result showing that CL1 outperforms modern GPUs. For now, the lower-power claim should be treated as a proposition to test, not as proof that biological computers can replace conventional AI infrastructure.

The trade-off: living systems have a limited operating life

The CL1’s biological component is also its main operational constraint. The neuronal cultures are designed to remain viable for up to six months, while conventional silicon servers are typically deployed over much longer hardware lifecycles.

A biological installation therefore needs culture management, environmental control and replacement procedures. It is not a maintenance-free chip that can simply be installed and left to run for years. Any large-scale deployment would need to show that the benefits of biological adaptation and lower power use justify those additional biological and operational requirements.

What the CL1 cost when it launched

Cortical Labs’ reported 2025 launch-era pricing included:

• Single CL1 unit: about US$35,000
• 30-unit rack purchase: about US$20,000 per unit
• Wetware-as-a-Service: about US$300 per unit per week for remote access to in-house cultures

These figures describe launch-era pricing and should not be assumed to be current contractual rates. The cloud model is significant because it lets researchers access living neural cultures remotely without operating the biological hardware themselves.

Could Singapore scale to 1,000 units?

The 20-unit NUS rack is a validation stage. Reporting says a future Singapore facility could eventually house up to 1,000 CL1 units, but that expansion is conditional on regulatory approval and energy-efficiency and safety testing.

DayOne and Cortical Labs have also said that site design and operational planning will come before a larger build-out, with attention to performance benchmarking, governance, biosafety and compliance.

That means the 1,000-unit figure is a possible future capacity, not commissioned infrastructure. The prototype must first establish how the systems perform under realistic workloads, how reliably the cultures can be maintained, how the installation fits within data-center operations and what biological-material controls are required.

The available sources do not establish that a 1,000-unit biological deployment will be operating by any particular date. A planned data-center milestone should therefore not be treated as a guarantee that the biological expansion will proceed on the same schedule.

Potential applications—and what remains unproven

The partners and related reporting point to several possible uses for biological computing:

• Adaptive AI: systems that respond to changing inputs and potentially learn from less data.
• Drug discovery and neurological research: living neuronal cultures as experimental models.
• Robotics: research into adaptive control and machine behavior.
• Cybersecurity: exploration of systems that can respond to changing or anomalous patterns.
• Neuroscience: experiments into learning, plasticity and biological adaptation.

These are prospective applications, not established performance claims. The CL1 has not been shown by the provided evidence to outperform digital systems across general AI, robotics, drug discovery or cybersecurity.

Why the project matters for Singapore’s data-center strategy

Singapore is testing the CL1 in the context of tighter expectations around data-center efficiency and sustainability. The project is consistent with the direction of Singapore’s Green Data Center Roadmap, which addresses how the country can expand digital infrastructure while managing energy and environmental constraints.

Its immediate value is therefore exploratory. The prototype gives researchers, infrastructure operators and regulators a way to evaluate whether wetware computing can complement conventional silicon systems for narrow workloads. It does not signal the end of GPU data centers; it is an early test of whether living neural networks could become one part of a more diverse and energy-conscious computing stack.