Meta's New AI Chips Reveal a Faster, More Self-Reliant Hardware Strategy

"Meta is building these chips because buying AI hardware at scale is expensive, and relying too heavily on external suppliers leaves less room to shape that hardware to its own needs. Building more in-house could help the company keep AI costs in check."

"By naming MTIA 300, 400, 450, and 500 together, the company is showing how much of its in-house chip effort is now in motion. Each chip marks a different step in that buildout, from recommendation and ranking work through training to generative AI inference."

"It also gives Meta more say over how its systems are built for the work it cares about most. Off-the-shelf hardware can do the job, but custom chips give the company a chance to shape performance around its own products and priorities."

Meta unveiled four custom AI chips—MTIA 300, 400, 450, and 500—developed in less than two years, demonstrating accelerated in-house hardware development. Each chip targets specific workloads: the 300 handles recommendations and training, the 400 supports generative AI while maintaining earlier functions, the 450 focuses on generative AI models, and the 500 advances inference capabilities. Some chips are already deployed while others are scheduled for future rollout. Meta is building custom chips to reduce expenses associated with purchasing AI hardware at scale and to maintain greater control over hardware design tailored to its specific needs and priorities rather than relying on external suppliers.

Read at TechRepublic