Researchers are investigating organoid intelligence — computing with lab-grown clusters of human brain cells — as a possible low-power alternative to conventional AI. Proponents point to the large energy cost of modern models: estimates put training a model like Grok-4 at roughly 310 gigawatt-hours, about $43 million at a $0.14/kWh retail rate.
Organoid intelligence uses brain organoids, small self-organizing clusters of neurons grown in vitro, as living computational substrates. Early demonstrations have shown basic, proof-of-principle capabilities rather than production-grade performance. Teams have taught organoids to adapt firing patterns to improve at the simple videogame Pong. Other demonstrations reportedly used organoids to perform nonlinear signal prediction and basic speech classification tasks.
Separate transplantation studies have placed human-derived organoids into mouse brains, where the tissue matured and formed synaptic connections with host circuitry. Those results show organoids can process biologically meaningful signals in a living system, not only survive in a dish.
Supporters argue biology offers a power advantage: the human brain runs on roughly the power of a dim incandescent bulb, and living tissue adapts through neuroplasticity. Critics caution that organoid computing so far addresses simple problems and must still demonstrate clear performance or scalability advantages over silicon to be competitive and practical.
The field also raises practical and ethical challenges. Labs must feed and maintain living tissue, a reality that researchers sometimes describe wryly as requiring 'pizza and beer' rather than firmware updates. Technical hurdles include reproducible training, interfacing living networks to electronics, and showing sustained, task-relevant gains beyond toy tasks.
In short, organoid intelligence remains an experimental approach that may reduce energy footprints in theory. To become a viable computing platform it must move from demonstrations to repeatable, scalable performance that outperforms or complements existing hardware.
Photo credit: mindmatters.ai
Tags: organoid intelligence, brain organoids, low-power computing, energy consumption, biohybrid computing
Topics: Neurotech industry & startups, Brain–computer interfaces, Neuroscience & neuroplasticity