The Rise of Biological Computing: When Living Neurons Meet Silicon

When Living Neurons Meet Silicon — and Why a Server Rack Full of Human Brain Cells Could Change Everything.

Somewhere inside a server rack at the National University of Singapore, 16 million living human neurons are firing. Not metaphorically. Not in simulation. They are alive — bathed in a carefully maintained solution of salts and proteins, fed a steady diet of glucose, kept at body temperature, and threaded through with electrodes that whisper data to them and listen to what they whisper back. They are, in the most literal sense imaginable, computing.

This is not science fiction. It is not even particularly distant science. It is August 2026, and the world’s first biological data center prototype is operational — a modest but genuinely historic assembly of twenty device units, each one housing roughly 200,000 lab-grown neurons interfaced with silicon hardware. Together, they represent a new answer to a question the computing industry has been quietly dreading: what comes after silicon reaches its limits? The question sounds urgent and contemporary. The pursuit of an answer, it turns out, is over two centuries old.

The answer, it turns out, may have been growing in our skulls all along.

A History 235 Years in the Making

The origin story of biological computing does not begin in a Melbourne laboratory or a Singapore server room. It begins in Bologna, in 1791, when an Italian physician named Luigi Galvani noticed something strange: that an electrical spark applied to the severed leg of a frog caused the muscles to convulse as though the animal were still alive. Galvani had discovered bioelectricity — the foundational insight that living nerve tissue carries and responds to electrical signals. The implications were seismic, even if the world would take another two centuries to fully reckon with them. Every neuron on every chip in Singapore today is, in a direct and traceable sense, a consequence of that frog’s leg twitching on a dissection table.

A half-century later, in 1850, the German physicist Hermann von Helmholtz did something that seemed mundane but proved transformative: he measured how fast a nerve signal actually travels. The answer — roughly 27 meters per second — was shockingly slow by the standards of electricity, which moves near the speed of light. That finding demolished the prevailing assumption that the nervous system was a simple electrical circuit, something that could be understood by the principles governing telegraph wires. Instead, it revealed the nervous system as something deeply, irreducibly complex — a biological system operating by its own logic, at its own pace, for its own purposes. The brain, scientists began to suspect, was not merely an organ. It was a computational apparatus unlike anything they had imagined.

The theoretical birth of biological computing arrived in 1943, in a paper of almost unsettling prescience. Warren McCulloch, a neurophysiologist, and Walter Pitts, a largely self-taught mathematician, published “A Logical Calculus of the Ideas Immanent in Nervous Activity” — a work that proved, mathematically, that networks of biological neurons could in principle compute any logical function that a machine could. They were not building anything. They were describing what the brain already was: a universal computing device assembled from living tissue. The paper directly inspired the logic circuit architectures that would underpin every digital computer built in the decades that followed. The silicon age, in other words, was conceived inside a model of the brain.

The engineers and theorists who came after McCulloch and Pitts moved quickly. In 1958, psychologist Frank Rosenblatt built the Perceptron — a physical machine that learned from experience by adjusting the weights of its connections in response to feedback, directly modeling the way biologists believed synapses strengthened and weakened in the brain. It was the first working demonstration of machine learning rooted explicitly in biological principles, and its descendants, scaled beyond anything Rosenblatt could have foreseen, are now the engines of modern artificial intelligence. Meanwhile, mathematician Norbert Wiener was weaving these threads into a broader theory. His concept of Cybernetics — coined in the late 1940s and developed through the 1960s — framed feedback, control, and communication as universal principles shared by animals and machines alike. In doing so, it laid the conceptual groundwork for the idea that biology and hardware were not categorically separate things, but two expressions of the same underlying logic.

The leap from theory to living demonstration came in two electrifying moments. In 1963, neuroscientist José Delgado implanted a wireless radio receiver directly into the brain of a bull and, in a Spanish bullring, pressed a button on a handheld transmitter to stop a charging animal in its tracks. The bull’s own nervous system had been commandeered by a radio signal. Six years later, in 1969, Eberhard Fetz showed that monkeys could be trained to voluntarily control the firing of a single identified neuron in their motor cortex — and that the firing of that neuron could be wired to move a needle on an external meter. A biological neuron, operating inside a living brain, had been made to act as a switch for a machine. The boundary between nervous system and device had been crossed, in both directions, for the first time.

It was Jacques Vidal, a computer scientist at UCLA, who in 1973 formalized what these experiments had been circling toward. He coined the term Brain-Computer Interface — BCI — and defined the research agenda that would drive the field for the next five decades: the systematic study of how biological neural signals could be read, interpreted, and used to control external devices. That agenda bore its most dramatic fruit in 2004, when the BrainGate consortium implanted a 96-electrode silicon array into the motor cortex of Matthew Nagle, a 25-year-old man paralyzed from the neck down following a stabbing. Within weeks, Nagle could move a cursor on a computer screen using thought alone. He could mentally play Pong. He could open and close a robotic hand. For the first time, silicon and living neurons were working in genuine, real-time partnership inside a human body — not in theory, not in an animal model, but in a person watching a cursor move because his neurons told it to.

Against this backdrop, the founding of Cortical Labs in Melbourne, Australia in 2019 reads not as a sudden departure but as an arrival. When Hon Weng Chong, Brett Kagan, and Andy Kitchen set out to grow living neurons on silicon chips as a computing substrate in their own right, they were drawing on 228 years of accumulated insight — from Galvani’s frog to Nagle’s cursor — and asking the next logical question: what if the neurons were not just a controller, but the computer itself? What if, instead of reading biology to drive silicon, we let biology do the driving?

The answer to that question began to take shape in a Melbourne laboratory, in an experiment involving a paddle, a ball, and a dish of neurons that had never heard of Pong.

From Pong to a First-Person Shooter

The moment that announced biological computing to the world arrived not with fanfare, but with a blip on a screen. Researchers at Melbourne-based Cortical Labs had placed roughly 800,000 human and mouse neurons onto a CMOS silicon chip — a device they called DishBrain — and connected it to a simple feedback loop. When the neurons moved a paddle correctly in the Pong arcade game, they received a mild electrical stimulus. When they missed, the stimulus was different, more chaotic. Within minutes, the neurons began to adapt. They learned to play Pong. The results were published in the journal Neuron, and the scientific community’s collective jaw dropped.

What DishBrain demonstrated was something almost philosophically disorienting: that neurons, given nothing more than structured feedback and electrical nudges, will spontaneously organize themselves toward a goal. No code. No reward function in the conventional sense. Just biology doing what biology does — learning from the environment in real time.

“Given nothing more than structured feedback and electrical nudges, neurons will spontaneously organize themselves toward a goal. No code. Just biology doing what biology does.”

But DishBrain was a prototype. The follow-up was more astonishing still. Independent researcher Sean Cole, working with Cortical Labs’ next-generation CL1 platform, used a Python API to hook approximately 200,000 cortical cells up to Freedoom, a 1993 first-person shooter. In under a week, the neural culture had learned to navigate the game’s environments. A dish of human cells, smaller than a thumbnail, was playing a video game that children of the nineties remember as genuinely challenging.

The milestone matters not because gaming is the point — it isn’t — but because it established a proof of concept that is now impossible to dismiss: living neurons can be taught to process complex, real-time information, and they can do it fast.

Inside the Machine That Breathes

The CL1, Cortical Labs’ commercial product, is a remarkable object. From the outside, it resembles a compact laboratory incubator — sleek, unobtrusive, roughly the size of a large shoebox. Inside, it is something else entirely. Approximately 200,000 lab-grown human neurons are deposited onto a silicon chip fitted with microelectrode arrays. Those electrodes perform two simultaneous functions: they stimulate the neurons with precisely calibrated electrical pulses, and they record the neurons’ responses in real time. The chip can both speak to the neural network and listen to it — a two-way conversation between silicon and biology measured in milliseconds.

The system connects to Cortical Labs’ biOS — a biological intelligence operating system — which serves as the translator between the living hardware and the software layer above it. Developers interact with the neurons through a Python API, issuing instructions and receiving responses as though querying any other computing substrate. The abstraction is remarkable: the programmer need not think about ion channels or synaptic potentials. They think about inputs and outputs, tasks and feedback. The biology handles the rest.

And handle it, it does. The neural network inside the CL1 is not static. It is continuously rewiring itself through a process researchers call self-wiring learning — the connections between neurons strengthen or weaken based on the patterns of activity they experience. No one programs those rewirings. They emerge from the same electrochemical processes that allow a child to learn to ride a bicycle. The system adapts. It grows, in a manner of speaking, more capable.

The CL1’s internal life-support system — temperature regulation, nutrient flow, atmospheric control — keeps the neural cultures alive for up to six months. It is, in every meaningful sense, a machine that breathes.

A Data Center That’s Alive

In August 2026, a partnership between NUS Medicine — the Yong Loo Lin School of Medicine — Singapore data center operator DayOne, and Cortical Labs announced something unprecedented: the world’s first biological data center prototype. Twenty CL1 units, arrayed in a standard server rack at the National University of Singapore, housing a combined estimated 16 million living human neurons. A server rack. Full of human brain cells. Sitting in a university building, connected to the internet.

The sheer fact of it demands a moment of stillness. We have built data centers before. Enormous ones, sprawling across desert landscapes, consuming the electricity of small cities, cooled by rivers of chilled water. This one fits in a rack and runs on roughly 1,000 watts total — about what it takes to run a kitchen toaster and a hair dryer simultaneously. And it is alive.

The Singapore installation is explicitly described as a prototype — a proof of concept rather than a production system. No one is routing the world’s search queries through neurons. Not yet. But what it demonstrates, beyond reasonable doubt, is that the infrastructure paradigm of biological computing is achievable. That neurons can be integrated into the physical architecture of modern data centers. That the question has shifted from whether this is possible to how far it can go.

“The question has shifted from whether this is possible to how far it can go.”

The Power Paradox

To understand why this matters, consider the electricity bill of artificial intelligence. Training a large language model can consume millions of watts over weeks. Running inference — generating responses in real time — requires data centers of staggering scale, cooled by industrial infrastructure, powered by grids that increasingly struggle to keep pace. AI’s energy appetite is, by now, a recognized crisis. It is one of the central constraints on how far the current generation of silicon-based AI can realistically expand.

Biological computing does not share this problem. Each CL1 unit runs on approximately 20 watts. Twenty. The Singapore rack of twenty units — 16 million living neurons — runs at roughly 1,000 watts total. The reason is fundamental: living neural networks operate on the same low-energy electrochemical principles as the human brain itself, which runs on roughly 20 watts of metabolic power while simultaneously managing perception, memory, language, emotion, movement, and a ceaseless background hum of unconscious computation that no silicon system has come close to replicating.

The power paradox of our AI moment is this: the most sophisticated information-processing system known to science runs on the energy of a dim light bulb, yet we have built our artificial approximation of it using hardware that consumes power at industrial scale. Biological computing, in one sense, is simply the observation that we had the answer in front of us all along — and are only now learning to wire it into a server rack.

The Hard Limits

None of which is to suggest that biological computing is ready to replace your GPU cluster. It is not. The field’s proponents are admirably candid about this, and intellectual honesty demands that the limitations receive the same column inches as the breakthroughs.

Scale is the core problem. The human brain contains an estimated 100 trillion synaptic connections. The Singapore prototype holds 16 million neurons — a remarkable achievement, but one that barely registers against that figure. Matching even a fraction of the brain’s connectivity is not a near-term engineering challenge. It is a long-term scientific one, and it is far beyond what current technology can address.

The electrode problem is equally humbling. Current microelectrode arrays observe only a tiny fraction of the total electrical activity occurring within a neural culture. Imagine pressing a stethoscope against the outside wall of a packed stadium and trying to understand the conversation from six seats near one exit. You would catch murmurs. Fragments. You would miss almost everything. That is roughly the epistemic position of a researcher reading from today’s electrode arrays — the signals they capture are real, but they are a sliver of the full picture.

“Imagine pressing a stethoscope against the outside wall of a packed stadium and trying to understand a conversation from six seats near one exit. That is the epistemic position of today’s electrode arrays.”

Steve Furber, professor of computer engineering at the University of Manchester and one of the field’s most respected voices, has noted significant limitations around both scale and programmability. The neurons learn, yes — but directing what they learn, with precision, at scale, remains a profound challenge. And the cultures themselves are not permanent. A lifespan of months, not years, means biological computing systems require ongoing maintenance, periodic replacement, and careful biological stewardship. These are not insurmountable constraints, but they are real ones.

What Comes Next

Even so, the trajectory is unmistakably upward. In May 2026, researchers at Princeton University published results from a platform called 3D-MIND — a flexible three-dimensional electronic mesh embedded directly inside lab-grown neural networks of approximately 70,000 neurons. Unlike flat electrode arrays that can only listen to the surface of a neural culture, the 3D-MIND mesh distributes electrodes throughout the entire tissue volume, achieving stable interaction tracking over six months. It is, in electrode terms, the difference between pressing that stethoscope against the stadium wall and threading microphones through every row of seats.

The implications are significant. Better electrode coverage means better understanding of what the neurons are actually doing — which means better ability to direct their computation toward specific tasks. The 3D-MIND approach, combined with improving life-support systems and richer software interfaces, points toward a future in which biological computing substrates become genuinely programmable in a meaningful sense.

The applications on the near horizon are not gaming. They are brain-disease modeling — using living neural networks to test how diseases like Alzheimer’s develop, and how potential drugs might interrupt them. They are hybrid AI research, combining the pattern-recognition efficiency of silicon neural networks with the genuine learning plasticity of biological ones. They are energy-efficient AI infrastructure, offering a path forward as the power demands of conventional AI begin to collide with the physical limits of electrical grids. And they are, perhaps most profoundly, a new lens through which to study the brain itself — because to interface with neurons in real time, at scale, is to finally begin reading a language that life spent hundreds of millions of years writing.

The Deepest Question

Step back far enough and the story of biological computing resolves into something larger than a technology forecast. It is a story about what computation actually is. For decades, we have treated it as a property of silicon — of transistors and logic gates and the rigid choreography of binary arithmetic. We built our digital civilization on that premise. Now we are discovering, in Melbourne laboratories and Singapore server rooms and Princeton clean rooms, that computation is something older and stranger than that. It is a property of organized matter responding to its environment. It is what neurons have been doing for hundreds of millions of years, long before any human thought to build a chip.

The living server rack in Singapore is not just an engineering curiosity. It is a signal. It suggests that the boundary between the biological and the digital — a boundary we have long assumed was fixed, even sacred — is more permeable than we imagined. That the future of computing may not be built. That it may, in some meaningful part, be grown.

Sixteen million neurons are firing in a rack in Singapore, right now, as you read this. They do not know they are making history. But they are.

Sources & References

1. McCulloch, W. & Pitts, W. (1943). “A Logical Calculus of the Ideas Immanent in Nervous Activity.” Bulletin of Mathematical Biophysics. — Wikipedia overview.
2. Eberhard Fetz — voluntary neuron control demonstrated, 1969. Wikipedia.
3. Matthew Nagle — first human BCI implant, BrainGate 2004. Wikipedia.
4. Kagan, B.J. et al. (2022). “In vitro neurons learn and exhibit sentience when embodied in a simulated game-world.” Neuron, 110(23), 3952–3969. PubMed PMID: 36228614.
5. Kagan, B.J. et al. (2022). DishBrain paper. Neuron, Cell Press (full text, open access).
6. Cole, S. / Cortical Labs. “A living bio-computer made of human neurons just learned to play Doom.” Futura Sciences, March 2026.
7. Cortical Labs. CL1 — The world’s first code-deployable biological computer. Official product page. corticallabs.com.
8. Kagan, B.J. et al. (2025). “The CL1 as a platform technology to leverage biological neural system functions.” Nature Reviews Bioengineering. DOI: 10.1038/s41551-025-01348-z.
9. NUS Yong Loo Lin School of Medicine. “NUS Medicine, DayOne and Cortical Labs Unveil Biological Data Center Prototype in Singapore.” Published 17 Aug 2026.
10. DayOne Data Centers Limited. “DayOne Launches Singapore’s First Biological Data Center Prototype with Cortical Labs and NUS Medicine.” PR Newswire, 17 Aug 2026.
11. Princeton Engineering. “New 3D device harnesses living brain cells for computing.” Office of Engineering Communications, 27 Apr 2026.
12. Mritunjay, K., Sturm, J.C. & Fu, T.M. (2026). “A three-dimensional micro-instrumented neural network device.” Nature Electronics. DOI: 10.1038/s41928-026-01608-1.

This article reflects developments as of August 2026. Research in biological computing is advancing rapidly; findings and commercial availability are subject to change. All facts and figures are drawn from peer-reviewed publications, institutional announcements, and manufacturer disclosures current to the date of publication.