Meta Platforms is testing its first in-house chip designed to train artificial intelligence (AI) systems, aiming to reduce its dependency on hardware suppliers like Nvidia. According to Reuters, the custom AI chip was developed in collaboration with Taiwan-based TSMC and is tailored to handle AI-specific workloads.

The chip is currently undergoing a small-scale deployment, with Meta planning to scale up production if the tests prove successful. This marks a shift for Meta, as previous custom AI chips were only used to run models, not train them.

Reuters noted that some of Meta's earlier AI chip design efforts had been canceled or scaled back due to failing to meet internal expectations. However, this latest development signals a renewed push to strengthen Meta's AI infrastructure.

Meta is projected to spend $65 billion on capital expenditures this year, much of which will go toward purchasing Nvidia GPUs. If the company can transition even a portion of its AI training processes to in-house chips, it could significantly cut costs.

The current testing phase will be crucial in determining whether Meta can successfully integrate its own AI hardware into its broader AI strategy. This move highlights Meta’s ongoing efforts to enhance its AI capabilities while aiming for greater control over its technological resources.

PHOTO: REUTERS

This article was created with AI assistance.