By Dr. Julie Kim | Staff Writer Published February 7, 2026 | HarvardScienceReview.org
The modern supercomputer is a marvel of engineering, capable of exascale calculations that drive our climate models and generative AI. But it has a fatal flaw: it is stupidly inefficient.
To match the processing power of a single human brain—which operates on about 20 watts of power, roughly the same as a dim lightbulb—a silicon-based supercomputer requires upwards of 10 megawatts. That is an energy differential of six orders of magnitude.
As Moore’s Law slows and the energy demands of AI skyrocket (a crisis we have documented extensively), physicists are hitting the “heat wall.” We can only shrink transistors so much before they become unstable.
The solution, according to a growing coalition of biophysicists and engineers at institutions like Johns Hopkins and Cortical Labs, is not better silicon. It is “wetware.”
Welcome to the field of Organoid Intelligence (OI).
The Hardware is Alive
Organoid Intelligence is the intersection of tissue engineering and computer science. It involves growing three-dimensional clusters of human neurons—called brain organoids—derived from stem cells.
These are not just flat cultures in a petri dish. They are complex, 3D structures that mimic the architecture of the developing human cortex, complete with synaptic connections and the capacity for “plasticity” (learning).
But a brain in a jar does nothing without input. To turn these organoids into computers, scientists place them on Microelectrode Arrays (MEAs). These chips act as the interface, sending electrical signals (input) to the organoid and recording the neuronal firing patterns (output).
The “Pong” Moment
The proof of concept arrived with a splash when Cortical Labs demonstrated that their system, dubbed “DishBrain,” could learn to play the video game Pong.
This wasn’t simple programming. The neurons were not “told” the rules. Instead, they were fed electrical feedback: a predictable, organized signal when they hit the ball, and a chaotic, random signal when they missed.
Driven by the Free Energy Principle—the biological imperative to minimize surprise and unpredictability—the neurons self-organized. They physically rewired their synaptic connections to master the game in five minutes. It took DeepMind’s reinforcement learning AI 90 minutes to do the same thing.
The “Green” Supercomputer
Why does this matter? Because biology is the ultimate efficiency engine.
While silicon chips process information sequentially (the von Neumann bottleneck), biological brains process information in parallel, with memory and computation happening in the same location (the synapse).
If we can scale OI, we could theoretically build “biocomputers” that run complex AI models using a fraction of the energy of today’s data centers. Imagine a server farm that runs on glucose instead of a coal plant.
The Ethical Frontier
Of course, this technology sits on a razor’s edge of bioethics. We are not just building chips; we are growing neural tissue.
At what point does an organoid become “sentient”? Currently, these clusters have about 50,000 cells—roughly the brainpower of a cockroach. They feel nothing. But as we scale to millions of neurons, the question of “suffering” becomes non-trivial. The Baltimore Declaration toward OI has already begun to map out the ethical guardrails.
The Future is Really in the Future
We are likely decades away from a commercial “bio-processor” in your laptop. But for specific tasks—like drug discovery or low-power pattern recognition—the age of silicon supremacy may be ending.
The computer of the future might not be built in a foundry. It might be grown in an incubator.
