On August 17, 2026, a 20-unit rack of Cortical Labs CL1 biological computers began operating at the National University of Singapore’s Life Sciences Institute, running 16 million lab-grown human neurons on a total draw of 850–1,000 watts. The partners describe it as a research prototype hosted within DayOne-designed infrastructure.
Each CL1 stays viable for up to six months. Replacement requires sterile lab conditions, trained cell biologists, and newly differentiated stem-cell-derived neurons — so this is long-horizon research, not a product launch.
The rack’s power draw is not the hard part. According to Cortical Labs technical documentation, the machine now installed at the National University of Singapore runs on living human neurons grown from stem cells, and when those cells die, no software patch revives them. A sterile-lab team must grow new ones and install them. That operational constraint — not the efficiency claims in the launch materials — is the economic question the entire wetware thesis now turns on. Singapore is the right place to ask it, because the city-state’s efficiency rules are making conventional silicon expansion expensive. DayOne Data Centers, the operator hosting the prototype, calls it a research commitment rather than a product. Cortical Labs, the Australian biotech behind the CL1, supplies the hardware; the university supplies the stem-cell pipeline. Neither supplies a guarantee. The launch has been described as a shift. The more accurate label is a controlled test.
The power advantage runs on borrowed time
Inside each CL1 unit are roughly 800,000 lab-grown human cortical neurons, reprogrammed from donor skin or blood into induced pluripotent stem cells and grown across a multielectrode array. Electrical pulses stimulate the cells; sensors read the responses.
Each unit draws about 30 watts. That is the figure that makes the comparison to conventional silicon stark.
What has changed is the signal, not the architecture.
According to neuroscience literature and Cortical Labs documentation, a single neuron can route signals through more than 200,000 synaptic channels; a transistor toggles between two states. That biological complexity lets the system learn from far fewer examples. But it does not confer precision or long life. The practical ceiling is the culture, not the chip.
Singapore’s Infocomm Media Development Authority wants all existing data centres to reach a PUE of 1.3 or lower over the next decade.
Efficiency is the easy part; reproducibility is not.
Hon Weng Chong, Cortical Labs’ founder and CEO, positions the system as a complement to silicon. “Biological computing supplements AI in areas where data are sparse,” he said. Professor Rickie Patani, who directs the neurobiology programme at NUS Life Sciences Institute, adds that stem-cell-derived neurons give researchers a disease-relevant platform without animal tissue.
The public launch drew government and industry representatives, according to an NUS Medicine statement.
For an operator trying to win new capacity in Singapore, a rack that sips power is a spreadsheet line, not a science project. The twelve-to-eighteen-month question is whether the replacement economics bend toward a larger rack or retreat into academic labs.
The full chain from nutrient flow to a deployable API is easier to see than to describe.
| Entity | Current rule | New rule | Effective date |
|---|---|---|---|
| Existing Singapore data centres | Regional average PUE of 1.55–1.6 | PUE of 1.3 or lower under IMDA Green Data Center Roadmap | Phased over the next decade from May 2024 |
| New capacity awarded under DC-CFA2 | No uniform green-power mandate for prior capacity | PUE of 1.25 and at least 50 percent green power | December 2025 requirements |
| Chile (neurodata) | No specific constitutional protection for brain activity before 2021 | Brain activity and derived data safeguarded as sensitive under Law No. 21.383 | 2021 constitutional reform |
| Source: IMDA Green Data Center Roadmap; Singapore DC-CFA2; Chile Law No. 21.383 | |||
A city-state with no room for waste
Jamie Khoo, DayOne’s chief executive, frames the prototype as part of a strategy to scale compute while cutting resource intensity. The company has reported a $4.5 billion Series C in June 2026 and is preparing a dual Nasdaq and Singapore listing; the biological rack sits beside its SG1 hyperscale campus as a research bet, not booked capacity.
The wider field is still pre-market. Cortical Labs leads in packaged devices and cloud access; neuromorphic silicon offers a separate, non-biological route to brain-like efficiency.
Karl Friston, the University College London neuroscientist who co-authored the DishBrain study, has suggested that the CL1’s real value lies in enabling experiments on synthetic brains rather than displacing general-purpose computing. In an interview with IEEE Spectrum in June 2025, Friston stated that “the real gift of this technology is not to computer science. Rather, it’s an enabling technology that allows scientists to perform experiments on a little synthetic brain.”
The ethical layer arrives at the same moment. According to the Neurorights Foundation and Chilean legal scholars cited in the UNESCO Courier, brain activity data should receive the same legal status as organ tissue; Chile’s constitutional reform wrote that into law. For Western operators, the risk is not just biological logistics but a novel category of regulated neurodata.
None of this answers the core question the launch exposed.
A rack that must be re-cultured on a fixed schedule is not yet infrastructure. It is a biology experiment with a data-centre address. The next twelve to eighteen months will show whether the economics bend toward replacement or retreat.
Beyond the headline
The Bigger Picture
This prototype is less about replacing GPUs and more about testing whether data-centre growth can decouple from power and cooling limits by introducing a new class of compute tied to life-science infrastructure. Wetware turns neuron cultures into a resource that demands stem-cell pipelines and bio-manufacturing, linking digital expansion to regenerative medicine capabilities rather than chip fabrication alone.
The Science Gap
Public narratives emphasise CL1’s sample efficiency and low power, but laboratory reality still hinges on unresolved issues such as 2D culture scalability, electrode drift, and reproducibility of neural responses across batches. According to a 2025 biocomputing field review, peer-reviewed studies have not yet demonstrated robust performance on complex, real-world tasks beyond Pong and epilepsy models, so claims about wetware’s superiority over deep learning remain promising hypotheses rather than established scientific results.
The Timing
Singapore’s move lands at a moment when AI infrastructure is colliding with hard constraints: aggressive PUE targets, limited grid headroom, and rising concern over water use for cooling. At the same time, neurorights debates in Chile and Europe are maturing, so the first commercial racks of living neurons appear just as regulators start to grapple with mental privacy, making this week’s launch an inflection point rather than just another lab demo.
The next decision for wetware’s early observers
With Singapore treating the prototype as research and regulators already drafting neurorights rules, four groups face distinct decisions.
- Western Data Center Operator in APAC
Evaluate whether wetware can help you meet Singapore’s DC-CFA2 efficiency and green-power rules. Treat the NUS rack as a pilot signal, not a vendor-ready option; ask DayOne or Cortical Labs about replacement logistics and total facility power before proposing a trial.
- Biotech Investor in Neuromorphic Computing
Assess Cortical Labs’ $300-per-week cloud model and the scalability of its replacement cycle. Watch whether DayOne confirms expansion to hundreds of units over the next year; a delay would steer value toward niche research tools rather than data-centre integration.
- AI/ML Researcher Using Cloud Compute
Test Cortical Cloud for sparse-data tasks such as anomaly detection or drug screening, where sample efficiency matters more than throughput. Compare per-task cost and reproducibility against GPU baselines before allocating serious budget. The API is real; the biological variability is also real.
- Bioethicist or Neurolaw Scholar
Track Chile’s neurorights implementation and how EU and US regulators classify neurodata derived from commercial wetware. The NUS deployment gives you a concrete case study for consent, donor records, and mental privacy across jurisdictions. Publish your analysis early; the regulatory gap is still open.
FAQ
How can researchers access a CL1 without running a wet lab?
Cortical Labs offers remote access through a subscription model at US$300 per week per culture slot. Users write Python and interact through APIs while the company manages cell culture and life support. Commercial users still face data-governance and ethics obligations, especially for medical or behavioural models, because the biological substrate creates neurodata records.
Where do the neurons come from, and what donor consent applies?
CL1 neurons are derived from induced pluripotent stem cells reprogrammed from adult skin or blood donations, according to technical materials. That raises questions about informed consent, data retention, and cross-border use of derived cell lines. Prospective users should verify how donor agreements handle commercial applications and whether jurisdictions treat derived neurodata as sensitive health or biometric information.
What operational requirements does a wetware rack impose?
Beyond power and network, a commercial CL1 rack needs cleanroom-adjacent lab space, trained cell biologists, biosafety procedures for nutrient media and waste, and maintenance schedules for gas mixers and pumps. Early deployments will probably depend on university or contract research partnerships rather than standard colocation, which affects where Western firms can realistically host wetware workloads.
Explainer
- Wetware computing
- Wetware computing uses living neurons as the processing substrate instead of silicon transistors. The cells are grown on electrode arrays, stimulated with electrical pulses, and their responses are read as data. The approach trades silicon’s years-long lifespan for biological efficiency that remains experimental.
- PUE
- Power Usage Effectiveness measures total facility energy used for each unit of computing energy. A PUE of 1.3 means about 30 percent overhead beyond IT equipment, while Singapore’s DC-CFA2 rules set 1.25 for new builds. A lower number means less cooling and infrastructure waste.
- Induced pluripotent stem cells
- Induced pluripotent stem cells are adult cells reprogrammed back into a state where they can become many cell types. In Cortical Labs’ CL1, skin or blood cells are turned into human cortical neurons. This avoids embryonic tissue and lets researchers grow disease-relevant cells in controlled conditions.
- Neurorights
- Neurorights are legal protections for brain activity and information derived from it. Chile’s 2021 constitutional reform made it the first country to safeguard mental privacy, identity, and free will as constitutional rights. Commercial wetware raises the question of whether a donor’s stem-cell-derived neural signals count as protected neurodata.
RELATED STORIES
