According to reports, Meta is testing an internal chip to train AI systems, a part of a strategy to reduce their dependence on hardware manufacturers such as NVIDIA.
According to ReutersThe finish line, which is designed to handle specific AI workloads, was manufactured in association with the TSMC firm based in Taiwan. The company is piloting a “small implementation” of the chip and plans to expand production if the test is successful.
Goal has implemented custom AI chips before, but only to execute models, not train them. As Reuters points out, several of the company’s chips design efforts have been canceled or reduced after not fulfilling internal expectations.
Meta expects to spend $ 65 billion on capital expenses this year, much of which will be allocated to the NVIDIA GPUs. If the company manages to even reduce a fraction of that cost by changing to internal chips, it would be a great victory for the giant of social networks.