Meta, the owner of Facebook, has started testing its first in-house AI training chip. The initiative began on March 11 in New York and aims to lessen Meta’s need for outside suppliers like Nvidia and decrease high infrastructure costs. The chip is designed specifically for AI uses like recommendation systems and generative AI.
Meta’s new AI chip is an important step toward self-reliance.
Created with Taiwan-based TSMC, the chip is a special accelerator built for AI training jobs. It went through a successful tape-out process, a crucial step in making chips that usually takes three to six months and costs a lot of money. This project shows Meta’s plan to innovate its own silicon and stop paying so much to use parts from other companies like Nvidia.
In the past, efforts with custom chips for inference came with problems, causing Meta to rely on Nvidia GPUs at first. The success of its inference chip has encouraged Meta to focus more on making good training accelerators.
Recap: Meta has started testing its new in-house AI training chip with help from TSMC to lower costs and its reliance on Nvidia.
