Meta Reportedly Testing AI Training Chip

SEAToday.com, Jakarta - Meta is reportedly testing its own custom-developed chip to train artificial intelligence (AI) models in an effort to reduce infrastructure costs and decrease reliance on Nvidia.

According to a Reuters report cited by Engadget on Wednesday, Meta aims to use the chip for AI training by 2026 if testing proceeds as expected.

The company has already begun deploying a small-scale version of a specialized accelerator chip designed exclusively for AI tasks, making it more power-efficient compared to general-purpose GPUs from Nvidia.

This deployment follows Meta’s completion of its first tape-out phase, a critical step in chip development where the final design is sent for trial production.

The new chip is part of the Meta Training and Inference Accelerator (MTIA) series, a family of in-house silicon designed for generative AI, recommendation systems, and advanced research.

Meta, one of Nvidia's biggest customers, began deploying its MTIA chips last year for inference, the predictive process that powers AI models behind the scenes. These chips are already integrated into Facebook and Instagram content recommendation systems.

Looking ahead, Meta plans to expand the use of its in-house chips beyond content recommendations to power generative AI products such as its Meta AI chatbot.

Writer: Andi Raisa Malaha Thambas

1 SELENGKAPNYA