Singapore switched on a data center of 16 million living human neurons. It needs feeding every 3 days.
At the National University of Singapore, a rack of 20 Cortical Labs CL1 units is now running what NUS calls the world's first independently operated biologically integrated server...
Read the original caption
At the National University of Singapore, a rack of 20 Cortical Labs CL1 units is now running what NUS calls the world's first independently operated biologically integrated server rack: living human neurons, grown from stem cells on electrode-fitted silicon chips, taking part in the computation. The catch that makes it news is that the rack is alive. Technicians feed the culture a sugar, micronutrient and pH cocktail every three days, a gas mixer supplies CO2, oxygen and nitrogen as life support, and the neurons stay viable for up to about six months before they are replaced. Cortical Labs puts a CL1 at roughly 25-30 watts and a full 20-unit rack at 800-1,000 watts, against on the order of 10,200 watts for a single eight-GPU AI server. The headline figure of about 16 million neurons comes from 20 units x ~800,000; NUS also cites at least 200,000 per unit, so treat the total as reported rather than lab-measured. It remains a prototype. Proposed uses include drug discovery, neurological research, robotics, cybersecurity and fraud detection, and the partners have discussed scaling toward 1,000 units subject to regulatory and energy approval.
The real news
The story, straight
At the National University of Singapore, a rack of 20 Cortical Labs CL1 units is now running what NUS calls the world's first independently operated biologically integrated server rack: living human neurons, grown from stem cells on electrode-fitted silicon chips, taking part in the computation.
The rack is alive: technicians feed the culture a sugar, micronutrient and pH cocktail every three days, a gas mixer supplies CO2, oxygen and nitrogen, and the neurons stay viable for up to about six months before they are replaced.
A CL1 draws roughly 25-30 watts and the full rack 800-1,000 watts, against the order of 10,200 watts for a single eight-GPU AI server.
The headline figure of about 16 million neurons is a calculation of 20 units times about 800,000; NUS also cites at least 200,000 per unit, so the total is reported, not lab-measured. It remains a prototype.
The absurd
Why this is ridiculous
Singapore switched on a data center made of 16 million living human neurons - and it needs to be fed every three days. The first server room in history that can starve: sugar, micromacronutrients, pH, three times a week, or the rack goes down.
IT support has a new ticket: 'the culture is hungry.'
The breakdown
How it breaks down
The interesting number is the watts: 800-1,000 for the whole living rack, against about 10,200 for one eight-GPU server.
If brain-on-chip compute can keep that ratio while getting smarter, the economics of intelligent inference flip - the expensive part of AI was never the idea, it was the electricity and the heat. The neurons are the thesis; the power bill is the proof.
The upside
Why it could matter
The upside is the menu: drug discovery, neurological research, robotics, cybersecurity, fraud detection - fields where biological pattern, not string matching, is the actual object. And the honest constraint is the clock: six months of viability before a replacement.
Biology as compute is not a server farm; it is a greenhouse. The teams that win will be the ones that learn to treat inference like a crop, with a feeding schedule.


