Wetware: the strange, serious rise of computers made of neurons.

The most energy-efficient learning hardware ever characterized is not a chip. It is a dish of a few hundred thousand cultured neurons, and in 2022 it learned to play Pong. The DishBrain experiments, published in Neuron, showed that in vitro neurons embodied in a simulated game world reorganize within minutes to keep a paddle rallying — no reward chemistry, no genetic engineering, just structured electrical feedback exploiting the neurons' native drive to make their world predictable.

The field acquired its research agenda a year later, when a Johns Hopkins-led group coined the term organoid intelligence in Frontiers in Science: scale brain organoids past the point where they can support memory and learning, interface them with sensors and machine-learning harnesses, and study biological computation directly — with an embedded ethics program rather than a bolted-on one, since the substrate under study is living human neural tissue.

Then it became a product. In March 2025, Melbourne's Cortical Labs — the DishBrain team — launched the CL1, which it bills as the world's first code-deployable biological computer: cultures of human neurons grown from stem cells made from volunteers' blood on silicon electrode arrays, kept alive by an onboard life-support system and programmable through software. Customers can buy a unit outright or buy time on the chips remotely, a model the company calls Wetware-as-a-Service, through its Cortical Cloud. According to Cortical Labs' launch release, as reported by New Atlas, an entire rack of CL1 units uses around 850 to 1,000 watts — under a kilowatt, the kind of number that makes a silicon engineer look twice.

We follow this field closely. The scientific draw is obvious: living neurons are the ground truth that neuromorphic silicon abstracts, and dish-scale cultures let researchers test theories of plasticity and sparse coding that no simulator fully captures. The caution is just as obvious. Systems built from human neurons that learn raise questions — about moral status, donor consent, and where sentience could conceivably begin — that deserve answers ahead of capability, not after it. The right rule for wetware mirrors the right rule for models: capability should never outrun understanding.

Correction, November 9, 2026: this article originally called the CL1 "the first commercial biological computer" and cited only the CL1 product page for its human neurons, remote access and rack power. The product page calls it "the world's first code deployable biological computer" and does not mention those details, so we changed the wording, added sources for each and replaced "around a kilowatt" with the reported 850 to 1,000 watts.