Meta is reportedly exploring the development of its own chip for training AI systems as part of a strategy to lessen its dependence on hardware manufacturers like Nvidia. According to Reuters, this new chip, engineered to handle AI-specific tasks, was developed in collaboration with TSMC, a Taiwan-based company. The social media giant is currently running a “small deployment” of the chip and intends to increase production if the initial tests yield positive results. While Meta has previously rolled out custom AI chips, they were only used for running models rather than for their training purposes.

As noted by Reuters, many of the company’s chip design initiatives have been shelved or significantly scaled back after not meeting internal benchmarks. Meta anticipates spending $65 billion on capital expenditures this year, with a substantial portion allocated for Nvidia GPUs. If the company can cut even a small part of that expense by transitioning to in-house chips, it would represent a significant achievement for the tech giant.