Meta Tests Its Own Custom AI Chips to Reduce Nvidia Dependence
Meta has initiated trials of its own custom chips designed for training AI systems, aiming to reduce its reliance on hardware manufacturers like Nvidia. TechCrunch reported the development.
According to a Reuters report, Meta’s chip is specifically designed to handle AI workloads and has been developed in collaboration with Taiwan-based semiconductor company TSMC. The company is currently conducting limited trials, and if the results prove satisfactory, it plans to scale up production.
Previously, Meta had developed custom AI chips, but those were used solely for model inference, not for training. Several of the company’s past chip design projects failed to meet internal expectations and were either scrapped or limited in scope.
This year, Meta has allocated $65 billion for capital expenditures, with a significant portion set to be spent on purchasing Nvidia GPUs. However, if the company successfully implements its in-house chips, it could achieve substantial cost savings—marking a major milestone for the social media giant.
