Meta Develops In-House AI Training Chip to Reduce Reliance on Nvidia

Edited by: Veronika Nazarova

Meta has begun pilot testing its first in-house AI training chip, manufactured by TSMC. This move aims to reduce reliance on third-party suppliers like Nvidia, whose GPUs are widely used in AI.

The chip, a single-use AI accelerator, is designed for power efficiency in AI tasks. Meta plans to initially use it for recommendation algorithms on Facebook and Instagram, with future expansion to generative AI products like Meta AI.

Meta's previous attempt at an in-house inference chip was scrapped, but this new training chip has passed a crucial development milestone.

OpenAI is also developing its own custom AI training chip.

Did you find an error or inaccuracy?

We will consider your comments as soon as possible.