As energy demands soar in data centers and the need for chips intensifies, could biological cells offer a solution? Australian startup Cortical Labs is pioneering this concept by establishing two innovative biological data centers in Melbourne and Singapore. These facilities will utilize chips populated with reproducible neurons for data processing.

Cortical Labs stands out as a leader in the emerging field of biological computing, using nerve cells linked to microelectrode arrays to both stimulate and record cellular responses during data input. Recently, the company showcased its flagship computer, the CL1, demonstrating its ability to learn to play games like Doom within a week.

The Melbourne data center is set to feature approximately 120 CL1 units, while a collaboration with the National University of Singapore will launch with 20 units, aiming for a total of 1,000 CL1s, pending regulatory approval. This ambitious expansion is designed to enhance their cloud-based brain computing services.

Michael Barros from the University of Essex remarks, “Biological computers like CL1 have been developed by multiple research teams globally but pose construction challenges for widespread adoption.” He continues, “Cortical Labs is making biocomputers more accessible, set to be the first company to do this at scale.”

These biological systems can be trained for tasks like playing Doom, although understanding the optimal training methods for neurons remains a complex issue. Reinhold Scherer, also from the University of Essex, notes, “Having access can facilitate explorations in learning and programming, yet neurons cannot be programmed as traditional computers.”

Moreover, Cortical Labs asserts that its biological data centers are significantly more energy-efficient than conventional computing systems, with each CL1 unit consuming just 30 watts compared to thousands of watts used by state-of-the-art AI chips.

Paul Roach from Loughborough University highlights that scaling up these systems to function like traditional data servers could lead to remarkable energy savings, even if they require nutrients to sustain the neuron chips. However, the cooling requirements are expected to be much lower than in traditional setups, indicating considerable power conservation according to Cortical Labs’ estimates.

Yet, the technology is still nascent. Tjeerd Olde Scheper, who has collaborated with a competitor, FinalSpark, poses questions about efficacy, stating, “We’re still in early development stages.” He emphasizes that transitioning from a small network managing simple tasks to a larger-scale language model is a substantial leap.

A primary challenge remains: the capacity to save training outcomes and utilize these neurons for computational algorithms beyond specific tasks like gaming. Retraining these neurons after their life cycle is another hurdle, as Scherer points out, “If retraining is needed every month, longevity of use becomes an issue.”