Singapore's neuron rack has an operator, a landlord and a 20-kilowatt power bill

The number that makes this legible sits on Cortical Labs' own specification. A single CL1 is rated at 850 to 1,000 watts ^[11], so twenty of them in one frame is 17 to 20 kilowatts ^[12], and The Next Web points out that a conventional silicon rack sits in the same band ^[13]. Held at the top of that range for a year, the rack accounts for roughly 175 megawatt hours ^[17]. That is a facility line item, attached to a machine whose throughput has not been published ^[21].

The reason the draw looks so ordinary is that most of it is not computation. TNW's reading is that the power goes into keeping the cells alive ^[13], and that the efficiency, for now, belongs to the neurons rather than to the box around them ^[26]. FinalSpark, which runs 16 brain organoids on a remote platform from Vevey, claims its biological processors use about a million times less energy than digital chips ^[14]. That claim describes cells doing the computing; the incubators, pumps and temperature control are the part that reaches the electricity bill ^[15]. Any energy case for wetware has to survive its own life support.

Which is why the staffing is the more interesting disclosure. The CL1's life support keeps neurons viable for up to six months ^[16], so a twenty-unit rack implies on the order of 40 culture replacements a year ^[23], and NUS Medicine has committed its own people to growing and maintaining the cells under Rickie Patani's supervision ^[7]. The cells themselves are grown from stem cells and wired to the hardware through microelectrode arrays ^[6]. A demonstration unit does not need a husbandry rota. This one does, and someone has signed for it.

Cortical Labs is not coy about what the rack is for. Founder Hon Weng Chong says the prototype "shifts the conversation from research to commercial application," naming drug discovery, humanoid robotics, cybersecurity and fraud detection ^[19], and the company's technical pitch is tasks where data is scarce, on the argument that living neural systems learn from limited information and adapt as conditions change ^[20]. DayOne's chief executive Jamie Khoo frames the project as helping bring biological computing to market ^[22]. On 6 August, more than 80 guests watched the arrays and live neural activity in operation ^[5].

The pitch is aimed at a real problem: the IEA has data centre electricity demand up 17 percent in 2025 ^[9], and NUS researchers believe biological computing could run on a fraction of conventional power ^[8]. But the figure that would let a buyer act on that is watts per useful task, and TNW says nobody has published it ^[21], with the announcement itself carrying no numbers behind the efficiency claim ^[10]. Meanwhile the Netherlands is assembling a neuromorphic hub around chips that imitate neurons without needing to be fed ^[18], which is the same promise with no incubator attached.

The install still earns attention, for an unglamorous reason. Until now the field's claims came from platforms and papers; this one comes with a named operator, a data centre partner and a metered enclosure, which means the missing number is now measurable rather than merely absent.