Why it matters
Meta's MTIA lets it avoid Nvidia's markup while relying on TSMC to make the chips. Like Amazon with Trainium, Meta has its own AI chip. Meta uses it only for itself.
Tap a figure for the line in its source.
Meta gives Broadcom
- of Meta's own AI chips, the first phase of a multi-gigawatt rollout; no price disclosed
Broadcom gives Meta
- co-design of the chips and the technology to deploy them
What happened
Meta is deploying MTIA, its own custom AI chip that it co-designed with Broadcom. The deployment starts at more than a gigawatt. Like Amazon with Trainium, Meta avoids Nvidia's markup. Meta uses MTIA only for itself, while Amazon makes Trainium available to others. TSMC still makes all of the chips.
Sources 2 documents — 1 company release · 1 press report
- States more than 1 GW initial commitment
- States foundry
The record's own summary of its sourcing: Broadcom PR + Meta Q1'26 call · as of Apr 2026
The evidence 2 sourced figures
The record's headline MTIA custom silicon with Broadcom: initial commitment >1
- Size
- >1 GW initial
- When
- Apr 2026
- Status
- announced
- Structure
- custom-ASIC co-development + deployment
- Programme
- Meta buildout
- Parties
- Meta designer/deployer (routes around Nvidia, not TSMC) · Broadcom ASIC + deployment partner
- Record
- disclosed / primary source · as of Apr 2026
Resolved figures 2 each with the verbatim line from its document
- initial-commitmentmore than 1 GW>1 GW initial MTIA commitment, first phase of a multi-gigawatt rollout — corroborated by Meta's own Q1'26 call ('more than 1GW of our own custom silicon... with Broadcom'); no contract value disclosed by either party
“The initial commitment exceeds 1GW and is the first phase of a sustained, multi-gigawatt rollout.”
Broadcom PR 4/14/2026 · globenewswire.com - foundryTSMC makes the MTIA chips — 'The MTIA chips are fabricated on TSMC's advanced nodes', MTIA 100 on 7nm and MTIA 200 on 5nm, the 300-series onward 'reportedly moving to 3nm' (Tom's Hardware, May 2026); Broadcom's release does not name the foundry
“The MTIA chips are fabricated on TSMC's advanced nodes: MTIA 100 on 7nm, MTIA 200 on 5nm, and the 300-series onward reportedly moving to 3nm with CoWoS packaging.”
Tom's Hardware, 'The custom AI ASIC state of play (May 2026)' · tomshardware.com