Meta Deploys MTIA v1-v3 AI Silicon Globally Across Three Continents
Resolved · supply chain · Occurred · Assessed · 1 sources
This assessment is dated 2026-09-17. Check what remains unresolved below before relying on it.
Which investments are exposed, and through which channels?
Internal ASIC deployment lowers Meta's per-token inference capex and reduces sole-source dependency on third-party merchant GPU suppliers.
| Question | What the evidence establishes |
|---|---|
| What was reported? | Reported factMeta Platforms is deploying its proprietary MTIA (Meta Training and Inference Accelerator) silicon across North America, Europe, and Asia Pacific data centers. |
| What else was confirmed? | Reported factThe deployment encompasses MTIA v1 Freya, v2 Artemis, and v3 Iris processor generations. |
| Does the valuation support acting? | Interpretation5 stored valuation assumptions for META; the latest is dated 2026-09-26. |
| What would change the assessment? | UnknownMeta Infra Roadmap Update — Disclosures regarding MTIA v4 taped-out silicon specifications |
What would this event cost your holdings?
Apply your own hypothetical returns to current holding values. The calculator does not infer security effects from this event.
Test this event against my holdingsReported facts
- · Meta Platforms is deploying its proprietary MTIA (Meta Training and Inference Accelerator) silicon across North America, Europe, and Asia Pacific data centers.
- · The deployment encompasses MTIA v1 Freya, v2 Artemis, and v3 Iris processor generations.
- · Meta is planning next-generation v4, v5, and v6 custom ASICs with roadmap milestones scheduled through 2026.
Public research record
Research framework · page evidenceCompleted event assessment
capex efficiency and lower inference cost-per-token
| Type | Claim | Scope |
|---|---|---|
| Assumption | Meta Platforms is deploying its proprietary MTIA (Meta Training and Inference Accelerator) silicon across North America, Europe, and Asia Pacific data centers. | 2026-09-17 |
| Interpretation | capex efficiency and lower inference cost-per-token | META |
Countercase, invalidators, and sources
MTIA v3 handles over 50% of internal recommendation compute, driving notable operating margin improvement in 2027.
- · Meta Infra Roadmap Update — Disclosures regarding MTIA v4 taped-out silicon specifications
- Meta Platforms Global MTIA AI Processor Deployment Analysis Examines Global Deployment of MTIA v1, v2 and v3 ASICs · 2026-09-17
What remains unresolved
- Base path
- Meta runs a hybrid fleet, scaling MTIA for ranking workloads while procuring GPUs for frontier training.
- Adverse path
- Software compilation overhead on proprietary ASICs leads to delayed internal model deployment timelines.
- Supportive path
- MTIA v3 handles over 50% of internal recommendation compute, driving notable operating margin improvement in 2027.
- · Meta Infra Roadmap Update — Disclosures regarding MTIA v4 taped-out silicon specificationsthis week
Evidence
- Meta Platforms Global MTIA AI Processor Deployment Analysis Examines Global Deployment of MTIA v1, v2 and v3 ASICs2026-09-17
“Meta Platforms is expanding its MTIA (Meta Training and Inference Accelerator) AI processor deployment globally across North America, Europe, and Asia Pacific. The analysis covers MTIA v1 Freya, v2 Artemis, and v3 Iris generations, with future v4, v5, and v6 ASICs expected in 2026.”
Supports: Meta is expanding MTIA v1, v2, and v3 deployment globally across NA, Europe, APAC with v4, v5, v6 expected in 2026.