From Singapore: Australian biotechnology is not often the story that moves markets in this region, but Cortical Labs this week gave the global tech industry a reason to pay close attention to Melbourne. The company published a video showing its CL1 biological computer running the 1993 first-person shooter Doom, powered by what it describes as approximately 200,000 living human neurons sitting on a microchip. The demo is deliberately playful, but the commercial architecture behind it is anything but.
The CL1 was released in March by Melbourne-based startup Cortical Labs, marketed as the world's first code-deployable biological computer. The device is a self-contained unit in which lab-grown human neurons, derived from skin or blood samples of adult donors and reprogrammed into induced pluripotent stem cells before being differentiated into brain cells, sit on a planar electrode array with 59 electrodes on metal and glass. Inside that compact housing, an internal life-support system handles temperature control, waste filtration, gas mixing, and circulation, keeping neurons alive for up to six months.
Cortical Labs calls this computational model "Synthetic Biological Intelligence," a term meant to distinguish living computation from traditional artificial intelligence. The distinction matters commercially. In a global AI sector dominated by American hyperscalers and Chinese state-backed research programmes, a Melbourne startup positioning itself in an entirely separate category is a calculated strategic choice, not just a branding exercise.
From Pong to Doom: A Measurable Step Up
Cortical Labs first gained attention in 2022 after teaching approximately 800,000 human and mouse neurons on a chip to play the classic Atari game Pong. When those earlier DishBrain neurons learned Pong, they picked it up in about five minutes, a time that a standard deep reinforcement learning algorithm would take roughly 90 minutes to match. The jump to Doom is a significant technical escalation. While Pong only requires moving a bar up and down, Doom requires moving forward, backward, left and right in 3D space and defeating enemies, making it considerably more complex.
CTO David Hogan explains that independent researcher Sean Cole managed to pipe the video feed from the game into patterns of electrical stimulation so that the neurons could, in effect, feel the action. If the neurons fire in a specific learned pattern, the Doom character shoots; another pattern prompts movement. In this way, the brain cells can find enemies, shoot them, and progress through the level.
The company was candid about the current limits. Dr. Alon Loeffler acknowledged in the demo that the 200,000 human neurons currently play "a lot like a beginner" and die frequently, while emphasising that the team has "solved the interface problem." Cortical Labs' position is that the hard engineering work, connecting living tissue to a software environment reliably and at speed, is now done. The learning quality is the next challenge.
The Sceptic's View Deserves an Airing
Not everyone is persuaded the neurons are doing what the headlines suggest. Independent researcher Sean Cole, who built the demo himself, documented in the project's GitHub repository that his decoder software "tends to start becoming a policy head," meaning the conventional software may be learning to route around the neurons entirely. The docstring in his own CL1-side code states plainly that "the CL1 device performs NO computation," with the PyTorch models and game logic living elsewhere.
This is a serious scientific question, not a dismissal. What remains unproven is whether 200,000 human neurons can ever carry the policy, the actual decision-making, instead of just riding along. Cortical Labs has not claimed otherwise; the company frames the Doom demo as proof that the input-output interface works, and invites the research community to develop better learning strategies from there. That is an honest position, but investors and institutions evaluating the CL1 should understand the distinction between a working interface and a working biological processor.
The Commercial Case Is Stronger Than the Gaming Demo Implies
Each unit sells for USD $35,000, with volume pricing dropping to $20,000 per unit in 30-unit server rack configurations, and a full rack draws only 850 to 1,000 watts. For context, a rack of CL1 units consumes 850 to 1,000 watts, notably lower than the tens of kilowatts required by a data centre setup running comparable AI workloads. The company began shipping the first 115 commercial systems in 2025 and maintains cloud connectivity for live monitoring and remote code deployment.
The energy efficiency argument is where Cortical Labs has its sharpest commercial edge in the Asia-Pacific market. Data centre power consumption is a genuine constraint on AI adoption across the region, from Singapore's strict limits on new data centre construction to the rolling power shortages affecting parts of Southeast Asia. A computing substrate that draws a fraction of conventional GPU-server power, and that learns from smaller datasets, is a proposition worth taking seriously regardless of one's views on the Doom demo.
The CL1 is positioned as the first biological computer enabling medical and research labs to test how real neurons process information, offering what the company describes as an ethically superior alternative to animal testing while delivering more relevant human data and insights. In a recently accepted paper using an in vitro epilepsy model, the CL1 restored function in impaired neural cultures. Cortical Labs' chief scientific officer Brett Kagan noted that epileptic cells cannot learn games very well, but applying antiepileptics to the cell culture allows them to learn better, alongside a range of other previously inaccessible metrics. That is a tangible pharmaceutical research application, with commercial partners in Japan, South Korea, and Singapore already active in the neurological drug discovery space.
Ethics Cannot Be Deferred
The ethical considerations surrounding the use of human brain cells in computing remain an area of active discussion, particularly regarding questions of consciousness and sentience in artificially assembled neural systems. Cortical Labs has engaged bioethicists directly. Chief scientific officer Brett Kagan has addressed these implications openly, emphasising that the neuron networks are not conscious. He noted that the company's very first published paper was an ethics paper, and that since then it has collaborated with numerous independent international bioethicists, philosophers, and regulatory experts.
Those assurances are reasonable, but they are also early-stage. As biocomputing scales, the regulatory frameworks governing it will need to keep pace. Australia's Department of Health and the National Health and Medical Research Council will inevitably be drawn into questions about the sourcing, use, and disposal of human-derived neural tissue in commercial computing products. How that oversight develops will shape whether Australian biocomputing retains a competitive advantage or gets bogged down in regulatory uncertainty.
For Australian exporters and research institutions, the signal from this week's demo is that Cortical Labs is serious about commercialisation, not just publication counts. Theoretical neuroscientist Karl Friston of University College London has described the CL1 as potentially "the first commercially available biomimetic computer, the ultimate in neuromorphic computing that uses real neurons." Whether that assessment proves correct will depend on whether the neurons can be shown to contribute meaningfully to computation, not just serve as a biological relay. The Doom demo, taken on its own terms, is a proof of platform. The proof of substance is still being written.