Australia-based start-up Cortical Labs has announced plans to construct two data centres using neuron-filled chips. The facilities in Melbourne and Singapore will house its CL1 biological computers, which have demonstrated the ability to play video games like Doom. The initiative aims to scale up cloud-based brain-computing services while reducing energy consumption.
Cortical Labs, an Australian company developing biological computers, is advancing its technology by building dedicated data centres. These centres will incorporate chips filled with neuronal cells connected to microelectrode arrays, enabling the systems to process data through cellular responses. The firm recently showed that its CL1 unit could learn to play the game Doom within a week, building on earlier demonstrations with Pong.
The first data centre, located in Melbourne, will accommodate about 120 CL1 units. A second facility in Singapore, developed in partnership with the National University of Singapore, will start with 20 units and expand to 1,000 pending regulatory approval. This expansion supports Cortical Labs' cloud service for brain computing, making the technology more accessible.
Experts highlight the potential benefits. Michael Barros from the University of Essex noted, “What [Cortical Labs] is doing is essentially allowing its biocomputer to be accessible at a large scale,” adding that they will be the first to achieve this. Reinhold Scherer, also at Essex, explained, “You don’t program neurons like standard computers,” emphasizing the need to explore new training methods.
Power efficiency is a key advantage, with each CL1 requiring only about 30 watts compared to thousands for advanced AI chips. Paul Roach from Loughborough University suggested that scaling to room-sized setups could yield significant energy savings, though biological systems may need nutrients and less cooling.
Challenges remain, as the technology is early-stage. Tjeerd olde Scheper from Oxford Brookes University cautioned, “We’re still in the early days of this development.” Steve Furber from the University of Manchester pointed out difficulties in scaling to complex tasks like large language models and issues with memory storage and retraining, given the neurons' roughly 30-day lifespan.
While biological computers differ from silicon-based ones and cannot yet perform standard calculations, these data centres represent a step toward broader application in research and computing.