Melbourne, September 2025.
In an Australian data centre, server racks consume less power than a hairdryer.
This is not because they use more efficient chips, but because in place of the chips there are human neurons. Living.
Somewhere in the world, right now, someone is renting human brain cells over the internet.
The price is around 300 dollars a week, and no equipped laboratory or expertise in biology is required.
All it takes is an account, a connection and some Python code.
Cortical Labs, an Australian start-up founded in 2019, has built the world’s first commercial biological computer: the CL1. Each unit contains around 200,000 human neurons grown from stem cells, interfaced with a silicon chip. The system communicates with the neurons through electrical impulses, watches them respond and adapt, and uses them to compute. Since March 2026 anyone can access these neurons through Cortical Cloud, the company’s cloud platform.
Take note. We are not talking about a laboratory experiment. This is a commercial product. Available to buy. Real.
The complete rack consumes between 850 and 1,000 watts: a fraction of the tens of kilowatts required by a standard AI system of equivalent computing capacity.
We are witnessing the birth of a new category: Artificial Biological Intelligence.
What is meant by Artificial Biological Intelligence
Artificial Biological Intelligence, or ABI, is not yet an established term in the scientific literature. It describes the convergence of artificial intelligence and the life sciences in two different directions that are advancing in parallel.
The first is biocomputing: using living neurons as computational components. Biology not as a metaphor for AI, then (as in artificial neural networks), but as a real physical substrate. Living neurons that learn, adapt and process information.
The second direction is the opposite: using AI to design and generate biological systems. Language models applied to DNA, algorithms that synthesise proteins never seen in nature, systems that read the code of life as if it were text and write new code.
The two directions converge on the same implication: biology ceases to be a purely observational science and becomes an engineering discipline. We no longer describe how life works: we design it.
The neurons that learn to play Doom
To understand it, though, let’s start with what ABI can do.
Cortical Labs’ DishBrain system, documented in the journal Neuron in 2022, showed human neurons able to learn to play Pong within minutes. In 2025 the CL1, the commercial version of the system, ran a version of Doom. Imperfectly, yes. With latency, certainly. But it did it.
The connection between a digital gaming environment and a living biological substrate worked.
Neurons generate small electrical impulses to communicate with the network. We do something similar: we send impulses that represent bits of information and read their responses.
Brett Kagan, Chief Scientific Officer at Cortical Labs (IEEE Spectrum, 2025)
Obviously, the point here is not that human neurons play Doom better than a latest-generation processor. We are not there yet. They already do so, however, with much, much lower energy consumption, using far smaller quantities of data to learn, and in an adaptive way that does not require the full training cycle typical of traditional machine learning.
Johns Hopkins coordinates an international research community on organoid intelligence, and in 2023 published the Baltimore Declaration, an ethical framework for the responsible development of biocomputing. The position shared by most of the scientific community is that wetware is conceived as complementary to traditional silicon, not a replacement for it.
The AI that reads and writes the code of life
The second trajectory of ABI starts from a different observation. DNA is a sequence. Proteins are structures that can be determined from sequences. The molecular language of biology has formal properties similar to the natural language that AI models have learnt to process with surprising effectiveness.
A study published in Nature Biotechnology in 2026 describes AI models designed to process the language of life simultaneously: mapping and predicting the flows of information through DNA, RNA and proteins. The entire system of gene expression as a single object of analysis and prediction.
The most immediate practical implications concern drug discovery. Today, developing a new drug takes on average ten to fifteen years and more than a billion dollars. A huge share of that cost lies in identifying and validating molecular targets: understanding which proteins are involved in a disease and how a chemical compound might interact with them.
Generative AI systems for biology can simulate these interactions in billions of variants in parallel, identifying promising candidates in weeks rather than years. Admittedly, they do not eliminate clinical trials (and that is actually a good thing). But they move the starting point forward, drastically reducing the research space that has to be explored empirically.
Why ABI matters
The two strands we have described are not parallel lines but converging ones, and their convergence generates value that multiplies.
Biocomputing brings the properties of biological learning to computation: adaptability, energy efficiency, the ability to operate in complex physical environments. Biological AI brings the design and simulation capabilities of artificial intelligence to biology: speed, scale, the ability to explore vast spaces of possibility.
The result is a shift in the relationship between technology and nature. For most of the history of technology, engineering has imitated nature: birds’ wings inspired aeroplanes, the brain inspired artificial neural networks. ABI reverses this relationship: technology no longer imitates biology but generates it, and uses it as a component.
The most immediate consequences can already be seen in three areas.
Medicine and drug discovery
AlphaFold, the DeepMind system that predicted the three-dimensional structure of almost every known protein, irreversibly changed structural biology. That was 2020. In 2026, more advanced biological generation systems are producing proteins with specific functions designed from scratch: drug candidates for rare diseases that would never have found room in the traditional commercial models of pharmaceutical research.
Sustainable computing
Global data centres currently consume around 2% of the world’s electricity. As AI models grow, that percentage is set to rise. A CL1 rack consumes less than one kilowatt to handle computational workloads that would require tens of kilowatts in silicon. It is not yet scalable to the size of large data centres, but it points towards hybrid biological computing that could change the energy equation of AI in the long term.
Robotics and physical systems
The third area (perhaps less obvious) is biocomputing. A drone or a robot operating in complex physical environments needs to process sensory information in real time, adapt to unforeseen conditions and consume little energy. These are exactly the conditions in which biological neurons outperform silicon.
ABI, Ethics and Safety
ABI raises questions that the scientific community is beginning to address without yet having reached shared answers.
The most urgent concerns ethics. Consciousness.
If human neurons can learn, adapt and respond contextually to complex stimuli, at what point does this biological substrate acquire some form of subjective experience? The question is not merely philosophical: it has direct ethical implications for research, experimentation and the commercial use of these systems.
In other words: when will these (life-based) systems be able to call themselves “alive”?
Johns Hopkins’ 2023 Baltimore Declaration sought to establish an ethical framework, but consensus remains fragile. Today, more than ever, the discussion is becoming urgent.
The second question concerns biosecurity. Systems that generate new genetic sequences or new proteins from prompts are powerful tools in the hands of researchers with legitimate intentions. Yet they are also potentially dangerous in the hands of those without them. Synthetic biology has a history of debates on dual use, the use of the same technologies for both civilian and military purposes, which biological AI takes to a new level of complexity.
The Governance of Artificial Biological Intelligence
There is something dizzying about what is happening in this field, and it would be wrong to ignore it.
Are human neurons that learn to play a game the same thing as human neurons that think? Probably not. But the fact that the question is topical again, in a context where the neurons are not inside a head but on a chip that can be bought for 35,000 dollars or rented for 300 a week, is already an unprecedented shift in the relationship between human and machine.
Of course, a great many questions arise that can and must be addressed now, before it is too late. What is certain is that the distinction between the technology sector and the scientific and healthcare sector is becoming increasingly blurred (as we have also seen with Claude Science and Midjourney Medical). For precisely this reason, however, and precisely because a regulatory framework for biological AI does not yet exist, now is the time to build real Governance.
Biology has always been the territory of emergence, of irreducible complexity, of living matter that escapes formalisation. ABI does not resolve this irreducibility, but begins to use it as a resource rather than a limitation.
Perhaps this is the most profound transformation: not that machines are becoming biological, but that we are beginning to treat biology as a material from which to build machines. And that this distinction, on which we have built centuries of philosophy, medicine and technology, is becoming less sharp than, until just a few months ago, we would have been accustomed to thinking.
New Connections (FAQ)
Is renting human neurons grown from stem cells for 300 dollars a week already ethically problematic, or does the regulatory framework still have time to take shape?
Time has already run out. Not in the sense that the situation is beyond recovery, but in the sense that the product is commercial, purchasable and accessible now, while the regulatory framework does not yet exist. Johns Hopkins’ 2023 Baltimore Declaration laid some ethical foundations shared by the research community, but it has no force of law in any jurisdiction. The central question, the threshold at which a biological substrate acquires some form of subjective experience, remains open. Opening it now, before commercial scale makes the answer politically too costly, is exactly the kind of conversation that technology governance almost never manages to have in time.
Is the CL1 genuinely competitive with traditional AI chips, or is it still too early to compare them?
It depends on what you measure. In terms of pure computational performance, traditional silicon has no rivals today. But comparing the CL1 with a GPU on standard benchmarks is like comparing a lung with an industrial compressor: they measure different things. The advantage of wetware is not speed but adaptivity and energy efficiency in open-ended tasks. A biological system learns from far smaller quantities of data, consumes a fraction of the energy and handles environmental uncertainty in a way that silicon replicates only with much more complex architectures. The relevant comparison is not CL1 versus GPU. It is CL1 versus the edge systems that currently operate in uncontrolled physical environments.
What actually changes for a company that relies on AI to decide, produce or communicate, knowing that biology is becoming a computational material?
The underlying question about the governance of an organisation’s cognitive infrastructure is changing. As long as AI was silicon, the risks were technical: data bias, model reliability, access security. When the substrate becomes biological, the risks take on a new dimension: ethical, regulatory, and one of responsibility towards the very material being used. Organisations that treat AI as a commodity to be integrated without governance oversight are building dependencies on systems that will change in nature before they have had time to understand how to govern them. This is not tomorrow’s emergency. It is a structure being built today, while there is still room to do so.
Sources and further reading
Cortical Labs, CL1 and Cortical Cloud (corticallabs.com). The first commercial biological computer and the wetware-as-a-service platform.
IEEE Spectrum, This $35,000 Computer Is Powered by Human Brain Cells (spectrum.ieee.org). Interview with CSO Brett Kagan on how the CL1 works.
E+T Magazine, Wetware: Is your next computer a living one? (eandt.theiet.org). Overview of Cortical Labs, FinalSpark and Johns Hopkins. March/April 2026.
Information Age, This Melbourne data centre runs on human brain cells (ia.acs.org.au). Cortical Labs’ Bio Data Centre, operational since September 2025.
Nature Biotechnology, Generalist biological AI (nature.com). AI models for processing DNA, RNA and proteins simultaneously. 2026.
FinalSpark, Open and remotely accessible Neuroplatform (ncbi.nlm.nih.gov). Peer-reviewed academic research on wetware computing and organoid intelligence.