Brain Organoids Emerge as Rival to Silicon Neural Networks

A parallel bet on how to build intelligent machines is running alongside the silicon AI race, and it uses actual human brain tissue. Wired's feature walks through the state of brain organoid computing, where pea-sized spheres of neural tissue grown from stem cells sit on multi-electrode arrays and are treated as a kind of living processor.

The two companies most visibly commercializing this are Australia's Cortical Labs and Switzerland's FinalSpark. Cortical Labs made its name in 2022 when it trained brain cells wired to silicon to play Pong. FinalSpark's Neuroplatform now offers researchers remote access to living organoids, and teams have used it to train organoids on tasks like reading Braille characters. Both companies pitch the same core selling point: brain tissue processes information at a tiny fraction of the energy that GPUs burn. Researchers at Johns Hopkins, where Thomas Hartung coined the term 'organoid intelligence,' have suggested biocomputing could cut AI energy consumption by a million to ten billion times.

The caveat is loud, and it is coming from inside the field. STAT News reported in November that brain organoid pioneers themselves worry inflated claims about biocomputing could backfire, arguing there has not been rigorous work defining what these systems can actually do. The eye-catching benchmarks so far are narrow: a Brainoware team at Indiana University reported 78% accuracy identifying speakers after training on 240 recordings of eight people. That is a proof of concept, not a rival to a transformer.

Nothing in the reporting suggests organoids are about to replace large language models. What they may replace sooner is some of the biological research that runs on animal models or crude cell cultures today, and that is where the near-term revenue sits. The energy story is the bigger long-shot bet: if even a fraction of the million-to-ten-billion-times efficiency claim survives real workloads, the case for pouring ever more power into datacenters looks different a decade out.