Singapore's Neuron Rack Has to Be Fed. That Is the Story

The interesting fact about Singapore’s new biological data centre prototype is not that it thinks. It is that laboratory technicians feed it every three days.

On 6 August 2026, NUS Medicine, Singapore headquartered operator DayOne, and Melbourne’s Cortical Labs showed more than 80 guests a 20 unit rack of CL1 biological computers at the NUS Life Sciences Institute. NUS published the partnership on 17 August. The Straits Times reports the stack went live on 16 July. The partners describe the deployment as the first independently operated biologically integrated server rack in the world, and the first of its kind outside Australia. Those qualifiers matter. Cortical Labs already runs a larger biological facility in Melbourne.

Count the cells, then count the staff

Each CL1 grows lab derived human neurons on a microelectrode array. The Straits Times, after a 7 August facility tour, says each unit houses at least 200,000 neurons. That puts the Singapore rack in the low millions of cells, not the 16 million figure some secondary coverage recycled from earlier DishBrain era numbers around 800,000 neurons per dish. If a press release cannot reconcile those estimates, a newsroom should not flatten them into a slogan.

Cortical Labs’ own CL1 page says the onboard life support system is designed to keep cultures viable for up to six months. The Straits Times describes the maintenance plainly: a cocktail of sugar, micronutrients, and pH buffers every three days, plus a gas mix of carbon dioxide, oxygen, and nitrogen. That is not a software patch cycle. It is husbandry.

The energy claim is real. The replacement claim is not.

Founder Hon Weng Chong told the Straits Times each CL1 draws about 30 watts with life support included. Cortical Labs has also put a populated rack in the 800 to 1000 watt range, and The Next Web had to correct earlier copy that treated those rack watts as per unit watts. Even at the conservative end, this is not an H100. Chong compared the CL1 to Nvidia’s H100 SXM at up to 700 watts, and said silicon remains far superior for the fast, precise, repeatable arithmetic that large language models need.

That is the line to keep. Wetware is being sold as a complement for sparse data and changing conditions (Chong names humanoid robots in unstructured spaces, cybersecurity anomaly detection, drug discovery). It is not a ChatGPT moment, and it will not retire a GPU cluster in Jurong.

DayOne’s interest is easier to understand if you remember Singapore’s constraints. Data centres were about 7 percent of national electricity by 2020 after a 2019 pause on new builds. Jamie Khoo, DayOne’s CEO, framed the prototype as a way to pursue scale and lower resource intensity together, aligned with the Green Data Centre Roadmap. A 20 unit research rack does not yet prove that thesis inside a commercial hall. DayOne’s own SG1 is still heading toward 2027 operations. The March partnership framed NUS validation as a step toward a later DayOne commercial deployment. Until a paying operator orders a second rack without a research grant attached, this is a lab with good lighting.

Melbourne is the control group

The Straits Times reports Cortical Labs’ Melbourne site runs 120 CL1s with about 20 paying customers in corporate research and universities, including robotics and gaming work. Access is priced at about US$2,200 a month per CL1, versus about US$4,300 for a high end AI chip rental on major clouds, per Chong. Those are vendor figures, not audited utilisation. They do show the company is selling a research instrument, not vapour.

Plans to grow Singapore to 1000 CL1s sit behind three gates the Straits Times lists in one sentence: regulatory approval, energy efficiency tests, and safety tests. Professor Rickie Patani, who oversees the NUS site, talks about identifying which neurons and support cells should be manufactured at scale for data centre operations. That is a cell process problem, not a model release problem.

Four questions before the next “AI power crisis” take

If you see another headline that a rack of neurons will save the grid, ask four things. Is it in a real rack under independent operation, or a demo cart? Who feeds it, and on what cadence? What is the replacement cycle when a culture dies at month six? Would a data centre operator buy 20 more without a university attached?

Singapore has earned a careful piece because the partners put biology in a rack, published the maintenance, and still admitted silicon wins on LLM arithmetic. The story is energy plus logistics. Anyone selling it as the next ChatGPT is not reading the feeding schedule.

Sources

NUS Medicine newsroom, 17 August 2026, biological data centre prototype with DayOne and Cortical Labs

The Straits Times, 15 August 2026, Singapore newest data centre needs to be fed

The Next Web, 21 August 2026 correction on per unit versus rack wattage

Cortical Labs CL1 product page, culture viability up to six months

DayOne via PR Newswire, 17 August 2026 launch note

Data Center Dynamics on the Cortical Labs and DayOne partnership