Human Brain Cells on Computer Chips: What Is Real, and What Isn't

What is actually on sale

Two different things get described as brain-cell computers, and headlines tend to blur them. The first is Cortical Labs' CL1, which the company calls a code-deployable biological computer. Neurons are grown on a silicon chip, kept alive in nutrient solution by an internal life support system and driven through the company's biOS software, with an optional cloud service for remote use. The second is FinalSpark, a Swiss company that offers remote access to neural organoids. An organoid is a small three-dimensional clump of brain-like tissue grown from stem cells, and researchers send it electrical signals from their own code.

CL1 was announced on 12 March 2025. Media reports put the price at about US$35,000 a unit, and the neurons are said to survive for up to six months. That is, in effect, a server with a use-by date. Neuron counts are less clear: Cortical Labs' product page does not state one, and press coverage ranges from roughly 200,000 to 800,000 per unit, so treat any single figure with suspicion.

The most recent milestone is in Singapore. In August 2026 DayOne, Cortical Labs and NUS Medicine announced a rack of 20 CL1 units, which DayOne describes as the first independently operated biologically integrated server rack in the world. It sits in a live research environment at NUS Medicine and is a prototype for neuro-inspired AI, biomedical modelling and drug discovery work. As reported, Cortical Labs' chief executive says the best use cases still need to be identified, which is a candid admission for a product already on sale.

What the research shows, and what it doesn't

The study behind most of the excitement is the 2022 'DishBrain' paper in the journal Neuron. Its authors grew human stem-cell-derived or mouse neurons on a high-density multielectrode array, a grid of tiny electrodes that can both read and stimulate cells. Electrical patterns told the cells where a simulated Pong ball was, and the cells' own activity moved the paddle. The authors report improved performance within about five minutes of play that did not appear in the control conditions, and found that closed-loop feedback, where the cells' output changes what they are subsequently shown, was needed for it.

That is a genuine result, but a narrow one. The authors themselves say their findings are not conclusive and need further work. Some groups were small, many comparisons were exploratory and the improvement was not robustly carried across days. Two of the authors are Cortical Labs staff, so the company has an interest in the work, which is normal in this field but worth knowing. The paper also uses the word sentience in a very low-bar sense, meaning 'responsive to sensory impressions'. That is not the sense most people hear in the headlines.

A 2023 perspective paper from Johns Hopkins researchers in Frontiers in Science gives a useful sense of scale. It describes organoids below 500 micrometres across with fewer than 100,000 cells, set against a human brain of 86 to 100 billion neurons. It also notes that even the best electrode probes record from a small fraction of nearby neurons. The paper is a research programme and not a report of working computers, and its energy comparisons are described by the authors as illustrations rather than rigorous benchmarks.

An overview published in The Conversation in December 2025 by a University of Melbourne researcher, who says he formerly worked for Cortical Labs and has cut financial ties, lists what has been shown so far: playing Pong, basic speech recognition and improved prediction of epilepsy-related activity when paired with electronics. It adds that neural activity in current organoids is primitive, that experts generally agree they are not conscious and that the label 'organoid intelligence' risks implying parity with AI that does not exist.

How to read the energy and speed claims

The pitch is usually that a brain runs on about 20 watts and a data centre needs megawatts to do comparable work. FinalSpark's own site cites an estimate that silicon would need around 10 megawatts to match a brain's efficiency. That comparison is about a whole human brain with tens of billions of neurons. Today's cultures are a tiny fraction of that, so the headline number tells you very little about what a cultured chip can do per watt on a useful task.

Our judgement is that the useful measure is whole-system energy for a defined workload: the neurons plus life support, temperature control, the electronics that read and stimulate them and the conventional computer that interprets the signals. Reported power figures for CL1 vary and come from the manufacturer or press coverage rather than independent measurement. The Singapore announcement gives no measured energy savings or performance figures, and the coverage we read describes the low-power claim as qualitative.

So the honest status is 'plausible and interesting, unproven'. Until someone publishes a matched benchmark, meaning the same task on biological and conventional hardware with energy and cost counted end to end, treat energy and performance claims as marketing hypotheses.

The ethics and UK governance picture

The Nuffield Council on Bioethics published Neural Organoids: Ethical and Governance Considerations in May 2026. It concludes that neural organoids do not currently have the biological complexity associated with sentience. It also finds significant governance gaps, with no shared view on safeguards or where red lines should sit, and it recommends an alliance of tissue banks, funders and regulators to develop best practice, clearer consent for tissue donors and public engagement on what limits people want.

The Council's statement we read treats organoids as research models and does not discuss computing, so it should not be read as a ruling on biocomputers specifically. The Conversation piece makes the related point that most bioethics frameworks treat organoids as biomedical tools and not as components of computing systems, and that leading researchers say commercialisation is outpacing governance. For a business the practical consequence is that rules and expectations here are still being written, which is a reason for caution before attaching your name to any of it.

What it means for your business

Our recommendation, ranked from most to least urgent, is deliberately dull. The technology does not touch your servers, cloud spend, backups or security posture today, and nothing about it justifies changing a procurement plan. If you work in neuroscience, pharmaceuticals or toxicology the picture differs, because these devices are pitched at research and drug testing, including as an alternative to animal models. That is a research decision for specialists and ethics reviewers, not an IT one.

If a vendor, investor or board member does raise it, these questions separate evidence from enthusiasm:

• What task was run, and was the same task run on conventional hardware for comparison?
• Is the energy figure for the whole system including life support, or for the neurons alone?
• Has anyone independent of the vendor reproduced the result, and is it published and peer reviewed or a preprint or press release?
• What happens when the cells die, and can any learned behaviour be carried across to the replacement?
• Who has approved the use of the donor-derived cells, and under what consent and oversight?

What is still uncertain

Most of the important questions are open. Nobody has shown reproducible results across many cultures at useful scale, and individual cultures differ. A six-month lifespan on current devices limits what can be built on them, and it is unknown how reliably anything a culture has learned can be kept or moved. The Conversation overview says the field's current work is about reproducing prototypes, scaling up and finding practical uses, which is an accurate description of a technology that has not yet found its job.

We will revisit this when independent, like-for-like benchmarks appear or when UK governance moves beyond the May 2026 recommendations. Until then it is a fascinating development to follow, not one to plan around.