The brain runs on twenty watts. A data center runs on megawatts.
In laboratories across the country, scientists are cultivating clusters of living neurons — miniature echoes of the human brain — and asking whether biology might succeed where silicon is beginning to falter. The brain, running on the power of a dim lightbulb yet housing a quadrillion connections, has long been the silent benchmark against which all computing is measured; now, researchers at institutions like Johns Hopkins are proposing we build with it directly. The field, called organoid intelligence, is years from practical use, but it raises questions that reach well beyond engineering — about consciousness, about what we owe to living tissue we create, and about the nature of intelligence itself.
- Silicon chips are approaching a hard physical ceiling — transistors can only be packed so densely — while the energy cost of training a single AI system already rivals a decade of human caloric consumption.
- Lab-grown brain organoids, three-dimensional clusters of living neurons, have already been trained to play Pong, proving that biological tissue can learn and respond — not just simulate the process.
- The gap between proof-of-concept and practical application is vast: current organoids hold 50,000 cells, but enterprise-scale use demands 10 million, a scaling challenge that has no clear solution yet.
- Ethical fault lines are opening alongside the technical ones — questions about whether increasingly complex organoids might develop awareness, and what moral weight attaches to neural tissue grown, trained, and discarded.
- Researchers see a nearer-term path through medicine: patient-specific organoids could model Alzheimer's or test drug effects on living, genetically matched brain tissue before general biocomputing ever arrives.
- The field is calling for the same sustained investment that propelled artificial intelligence — with the understanding that organoid intelligence may take decades to travel from laboratory curiosity to transformative tool.
The human brain runs on roughly twenty watts — the glow of a dim lightbulb — yet contains 100 billion neurons linked by more than one quadrillion connections. Silicon chips, for all their speed, are approaching the physical limits of how densely transistors can be packed, and the energy cost of training advanced AI systems has grown staggering. Training AlphaGo, for instance, consumed more electricity than sustaining an active adult for ten years. Scientists are now asking a pointed question: if artificial intelligence was modeled on the brain to begin with, why not build computers from actual neural tissue?
The answer they are pursuing is called organoid intelligence — three-dimensional clusters of lab-grown neurons, sometimes called brain organoids, that behave in miniature like thinking tissue. Thomas Hartung of Johns Hopkins University has become one of the field's leading voices, arguing that these biological structures could eventually outpace traditional computers in speed, energy efficiency, and the capacity to learn. The proof of concept arrived in late 2022, when Dr. Brett Kagan and colleagues trained a flat sheet of brain cells to play Pong — a modest demonstration, but one that confirmed the underlying principle.
The road from that demonstration to practical application is long. Current organoids contain around 50,000 cells; researchers estimate they would need to reach 10 million before enterprise use becomes feasible. And beyond the engineering challenges lie ethical ones that cut deeper than those surrounding conventional AI — questions about whether increasingly complex organoids might develop some form of awareness, and what responsibilities attach to neural tissue that is grown, trained, and potentially discarded.
Hartung points to a more immediate application: organoids grown from a patient's own skin cells could allow scientists to study neurological diseases like Alzheimer's in living tissue, or to observe how drugs affect learning and memory in a system genetically matched to that individual. This kind of personalized research may arrive well before general-purpose biocomputing. For now, organoid intelligence remains a distant but serious promise — a reminder that the most powerful computer we have ever encountered still lives inside the human skull.
The human brain operates on a scale of efficiency that silicon chips cannot match. Scientists working in labs across the country are now trying to harness that biological advantage by growing brain tissue in dishes—three-dimensional clusters of neurons and supporting cells that behave, in miniature, like thinking tissue. These lab-grown structures, called brain organoids, represent a new frontier in computing that researchers believe could outpace traditional computers in speed, power consumption, and the ability to learn and adapt.
The concept, known as organoid intelligence, emerges from a straightforward observation: artificial intelligence was modeled on how brains work in the first place. Why not, researchers ask, skip the silicon middleman and build computers from actual neural tissue? A paper published in the journal Frontiers in Science lays out the case. Thomas Hartung, a professor at Johns Hopkins University's Bloomberg School of Public Health, has become one of the field's most vocal advocates. He points out that silicon-based computers excel at raw calculation—an AI system called AlphaGo, trained on data from 160,000 games of Go, would require a human player to spend five hours a day for more than 175 years to accumulate the same experience. But there's a ceiling approaching. Transistors cannot be packed more densely into chips without hitting fundamental physical limits. The brain, by contrast, contains roughly 100 billion neurons connected through more than one quadrillion connection points. It is, in Hartung's framing, an architecture of staggering complexity and efficiency.
The energy difference is perhaps the most striking advantage. Training AlphaGo consumed more electricity than it takes to sustain an active adult for ten years. A human brain, meanwhile, runs on about twenty watts—the power of a dim lightbulb. This disparity suggests that if organoid intelligence can be made to work at scale, it could revolutionize computing in fields where power consumption and processing speed both matter: data centers, artificial intelligence training, real-time analysis of complex systems.
But the technology remains in its earliest stages. Current brain organoids contain roughly 50,000 cells. To become practically useful, they would need to scale up to 10 million cells—a twentyfold increase that presents significant technical challenges. Researchers have demonstrated proof of concept: in late 2022, Dr. Brett Kagan and colleagues trained a flat structure of brain cells to play the video game Pong, showing that organoid tissue can learn and respond to stimuli. It was a small victory, but it proved the basic principle works.
Beyond the engineering hurdles lie ethical questions that cut deeper than those surrounding artificial intelligence. There is the question of consciousness—whether organoids, as they grow more complex, might develop some form of awareness. There is also the question of what it means to grow human neural tissue in a laboratory, to train it, to use it, potentially to discard it. These are not abstract concerns; they will shape how the field develops.
Hartung sees a more immediate practical application: personalized brain organoids grown from a patient's own skin cells could allow researchers to study neurological diseases like Alzheimer's in living tissue, or to test how drugs affect learning and memory in a system that is genetically unique to that individual. This kind of research could happen sooner than general-purpose biocomputing.
What the field needs now, Hartung argues, is a scientific community willing to invest in the work. The success of artificial intelligence suggests this is possible. But organoid intelligence, if it is to move from laboratory curiosity to practical tool, will require years—perhaps decades—of sustained research, funding, and problem-solving. For now, it remains a distant promise: a reminder that the most powerful computer we know of is still the one inside our skulls.
Notable Quotes
Silicon-based computers are certainly better with numbers, but we're reaching the physical limits of silicon because we cannot pack more transistors into a tiny chip. The brain is wired completely differently.— Thomas Hartung, Johns Hopkins University
Brains are more energy efficient than the computers powering AI models. The amount of energy spent training AlphaGo is more than is needed to sustain an active adult for a decade.— Thomas Hartung, Johns Hopkins University
The Hearth Conversation Another angle on the story
Why would we want to build computers from brain cells when silicon chips are already so fast at calculation?
Because speed at calculation isn't the only thing that matters. A brain can learn, adapt, recognize patterns in messy real-world data, and do all of it while using a fraction of the energy a data center needs. Silicon hits a wall—you can't make transistors much smaller. Brains don't have that problem.
But you said a brain would take 175 years to learn what AlphaGo learned in weeks. Doesn't that prove computers are better?
At one specific task, yes. But AlphaGo needed a decade's worth of electricity to get there. A human brain runs on twenty watts. If you could scale up organoids to millions of cells, you'd have something that learns faster than a human and uses almost no power. That's the real advantage.
How close are we to that actually working?
Not very. Right now organoids have about 50,000 cells. You'd need 10 million to do anything practical. And we've only proven they can learn Pong. There's also the ethical question—at what point does growing brain tissue become something we need to worry about morally?
So this is decades away?
Probably. But the science works in principle. The question now is whether enough researchers and funding will move toward it. That's always the harder part.