Meta MTIA
Meta's Training and Inference Accelerator for recommendation and GenAI
Meta designs the MTIA (Meta Training and Inference Accelerator) line to run its ranking/recommendation and increasingly its GenAI (Llama) workloads more cost-efficiently than merchant GPUs. MTIA v2 shipped in 2025 with next generations aimed partly at training, developed with Broadcom/Marvell support and fabbed at TSMC. Meta remains one of the largest buyers of Nvidia GPUs, so MTIA is a complement that reduces marginal GPU spend rather than a full replacement.
Chip
MTIA v2 (2025), training-focused next-gen
Primary use
Ranking/recsys, expanding to GenAI
Partners
Broadcom/Marvell; TSMC-fabbed
How it fits the stack
Meta MTIA with what it depends on (above) and what it feeds (below). The figure renders as a crawlable diagram and upgrades to an interactive 3D graph as it scrolls into view.
Meta MTIA in the AI stack. Meta MTIA with its immediate upstream dependencies (top) and downstream dependents (bottom) in the AI value chain. Hover a node in 3D, or read the full relationships below.
Graph data (text) — 4 entities, 4 relationships
- Meta MTIA —depends on→ TSMC (Taiwan Semiconductor Manufacturing Company)
- Meta MTIA —partners with→ Broadcom
- Meta MTIA —competes with→ Nvidia
- Meta MTIA —designs→ Broadcom