A Swiss startup has built a machine that thinks with living human brain tissue instead of silicon chips.
FinalSpark launched its Neuroplatform with 16 lab-grown brain organoids running computational tasks. The company says the system consumes a million times less power than conventional processors doing the same work.
The platform is not science fiction. It is online now. Research institutions around the world can run experiments on living human neural tissue through a web browser.
It raises a simple question. If your phone charger draws more power than a biological processor, did we build computation the wrong way?
What Is a Brain Organoid?
A brain organoid is a three-dimensional cluster of human brain cells grown from stem cells. The cells self-organize into structures that resemble parts of a developing brain. They form neurons. The neurons fire electrical signals. Networks take shape.
Scientists have been growing organoids for over a decade. They started as crude clumps of cells. Today they reach a few millimetres across and contain hundreds of thousands of neurons. They are about the size of a lentil.
The cells do not form a complete brain. An organoid has no sensory input. It does not see or hear. It also lacks blood vessels and immune cells, which limits how much it can grow. But the neural networks inside are real. They fire, connect and adapt.
FinalSpark’s platform places these organoids on Multi-Electrode Arrays. The electrodes record electrical activity around the clock at 30,000 samples per second. Researchers send electrical pulses into the array and watch how the organoids respond. It is a closed loop of stimulation and feedback. Input goes in. Biological processing happens. Neural output comes back out.
That loop is the basis of biocomputing.
The Energy Problem Silicon Cannot Solve
Training a single large AI model can burn through enough electricity to power hundreds of homes for a year. The human brain does not come close. It runs on roughly 12 to 20 watts. That is less than most LED light bulbs.
The gap is staggering. A supercomputer simulating even a fraction of human brain activity can draw megawatts. The brain itself achieves far more complex pattern recognition and learning at a fraction of that cost.
FinalSpark claims its biological processor uses a million times less energy than the silicon equivalent. Even if the headline number oversimplifies the comparison, the underlying point holds. Biological neural networks compute using chemistry and ion channels. Silicon chips force electrons through billions of transistors. The energy profiles belong to different universes.
Researchers at the University of Sydney have pointed out that the brain does not fire all neurons at once. It uses sparse coding, activating only the pathways needed for a given task. That selectivity is where most of the efficiency comes from. Silicon chips, by contrast, tend to run full parallel operations for every calculation.
Thomas Hartung and the Case for Organoid Intelligence
Thomas Hartung, a professor at Johns Hopkins University, has been pushing the idea of organoid intelligence for years. In February 2023, he and his team published a roadmap in Frontiers in Science laying out a path toward biocomputers powered by living brain cells.
“Computing and artificial intelligence have been driving the technology revolution, but they are reaching a ceiling,” Hartung said at the time.
His proposal is not about replacing silicon tomorrow. It is about building a parallel track. Train organoids to perform simple computational tasks. Scale them up with more cells, more electrodes and more sophisticated input-output systems. Pair biological hardware with machine learning algorithms that interpret the neural signals.
Hartung described the goal as a future biocomputer that could outperform artificial systems on certain types of learning and pattern recognition. The human brain excels at tasks like voice recognition and visual pattern detection precisely because those tasks are what brains evolved to do. Silicon systems bolted together those capabilities artificially. Biology comes with them built in.
The Johns Hopkins OI roadmap envisions organoids with more cell types, richer genetic programming and better vascularisation. It is an enormous engineering challenge. The alternative is watching AI energy consumption climb faster than anyone can build power plants.
Brain Cells Playing Pong
Before FinalSpark, an Australian startup proved that living neurons can learn.
Cortical Labs built a system called DishBrain that placed roughly 800,000 human brain cells on a silicon chip. The cells received electrical signals representing the position of a virtual ball and the paddle in the classic video game Pong. When the cells moved the paddle to hit the ball, they received a predictable reward signal. When they missed, the input was chaotic.
The neurons learned the pattern. They started hitting the ball more reliably over time. The experiment demonstrated something important. Living cells can adapt their firing behaviour based on feedback, and they can do it without being programmed to do so.
The DishBrain results were published in Neuron and confirmed by independent researchers. The cells were not conscious. They were not playing a game in any human sense. But their adaptive response provided concrete evidence that biological neural tissue can process information and modify its own behaviour based on feedback loops.
FinalSpark’s approach is different. Instead of one flat cell culture, it uses three-dimensional organoids. That matters. A 3D organoid has layered architecture closer to real brain tissue. More connections mean more computational potential. The trade-off is that organoids are harder to keep alive and harder to read from.
Living Machines Have Deadlines
Silicon chips last for years. Brain organoids do not.
FinalSpark’s system has to keep the 16 organoids alive by feeding them nutrient medium, maintaining temperature and monitoring waste buildup. Each organoid has a lifespan. The platform must replace dead tissue, recalibrate the electrodes and keep experiments reproducible across biological turnover.
That is not a bug. It is a feature of working with living matter. Biological systems adapt, degrade and regenerate in ways inert materials never will. The challenge for biocomputing is building infrastructure that embraces this cycle rather than fighting it.
FinalSpark’s Air-Liquid-Interface approach helps. The organoids sit on a permeable membrane with nutrients flowing from below. This reduces the volume of medium needed and allows easier exchange of gases.
The result is organoids that survive past 100 days with continuous recording and stimulation.
Ethics Before Scale
This is where the conversation shifts from engineering to ethics. If living human brain tissue processes information, adapts and learns, what responsibilities do researchers have toward it?
Hartung’s team convened an ethics consortium before scaling up the technology. The question is not whether today’s organoids are conscious. Current evidence says they are not. The question is what happens if the systems grow large enough to cross a threshold nobody defined yet.
The consortium recommends ongoing welfare monitoring, clear boundaries on experimental scope and a pre-planned shutdown protocol if organoids show unexpected signs of complexity. The parallel with AI safety debates is unavoidable, but with biology the stakes feel different. These are living human cells.
FinalSpark uses chemical reward signals with dopamine. That mirrors how reinforcement learning works in animals. It is effective. It also demands ethical care.
What Comes Next
For now, the biocomputer is a research tool. Researchers use it to study neural plasticity, drug toxicity and disease models. The computing applications are secondary, but they are real.
The most likely near-term use is pharmaceutical testing. Drug companies need living human neural tissue to screen neurological compounds. A remotely accessible platform with 16 organoids that can survive months and record activity around the clock offers something animal testing never could.
Further out, biocomputation could complement silicon systems in energy-intensive tasks like pattern recognition and signal processing. Nobody expects your next laptop to run on brain cells. But a data center could offload specific learning tasks to biological processors and save enormous amounts of electricity.
FinalSpark is betting on that future. The platform is open now. The organoids are firing. The energy numbers are hard to ignore.
References
Hartung, T., Smirnova, L., & Maertens, A. (2023). Organoid intelligence (OI): The new frontier in biocomputing and intelligence-in-a-dish. Frontiers in Science, 1, 1017235. https://doi.org/10.3389/fsci.2023.1017235
Jordan, F., Kurtys, E., & Schuster, M. (2024). Open and remotely accessible Neuroplatform for research in wetware computing. Frontiers in Artificial Intelligence, 7, 1376042. https://doi.org/10.3389/frai.2024.1376042
Kagan, B. J., Kitchen, A. C., Topping, N. T., et al. (2022). In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron, 110(23), 3953-3969. https://doi.org/10.1016/j.neuron.2022.09.033
