Extropic is not a traditional startup, but is developing a novel chip that harnesses the thermodynamic fluctuations that occur within electronic circuits that are usually an annoyance for engineers.
These fluctuations can be used to perform highly efficient probability calculations that could be useful for AI applications where artificial reasoning is required.
The company has shared more information with Wired, which shows it is on track to develop a chip three to four orders of magnitude more efficient than current offerings, thus helping to reduce AI’s high energy demands.
It would be suitable for use in datacentres, and is especially good for Monte Carlo simulations, used in areas such as finance, biology and AI.
Extropic’s CEO, Guillaume Verdon, admits that taking on established chipmakers such as Nvidia seems insane, but argues that given the incredible demand for AI computing power, now is the time to try to revolutionise computing.
