Meta's Custom MTIA Chip Trims Nvidia Dependence at the Edges, Not the Core

Meta is offloading inference and recommendation workloads onto in-house silicon, but frontier LLM training keeps it firmly in Nvidia's grip—mirroring playbooks already run by Google and Amazon.

ndtv-profit-business · 7/9/2026, 1:03:37 PM · Ask about this →