Back to News

Meta Testing In-House AI Training Chips to Reduce Nvidia Dependency

Meta is reportedly testing an in-house chip designed specifically for training AI models, a strategic move aimed at reducing its dependence on external hardware manufacturers like Nvidia. According to a report from Reuters, the chip was developed in collaboration with Taiwan-based semiconductor giant TSMC. Currently, Meta is running a limited pilot deployment to assess the chip’s performance before deciding whether to scale production.

While Meta has previously developed custom AI chips, they have only been used for running AI models rather than training them. Some of the company's past chip development projects were either canceled or scaled back due to underwhelming results.

Despite its ongoing efforts in chip design, Meta anticipates spending around $65 billion on capital expenditures this year, with a large portion allocated to acquiring Nvidia GPUs. If the in-house chip project succeeds, it could significantly reduce costs for the company, providing a major competitive advantage in the AI space.