Meta · MTIA · Layer 2
Meta MTIA v2
Nobody outside Meta. It is in this table because leaving it out would misrepresent the market — hyperscaler custom silicon collectively outweighs AMD as a threat to NVIDIA.
internal Not available to anyone outside the company that built it.
Specifications
- Memory
- Not published
- Bandwidth
- Not published
- Compute
- Not published
- Power
- Not published
- Interconnect
- Not published
- Software
- Internal, PyTorch-based
- Form factor
- accelerator
- Workload
- inference
Verified 2026-09-06. Fields reading “not published” are exactly that — we do not estimate a figure a vendor withholds.
The catch
You cannot rent, buy or benchmark it. Its market effect is entirely in the GPU orders Meta does not place.
The economics
Relevant to you only as demand it removes from the market you buy in.
Cost per million tokens is the metric that decides most real purchases, and there is no neutral benchmark for it — every published figure comes from a company selling one side of the comparison. Read all of them, ours included, as claims rather than measurements.
The power question underneath this
A rack of frontier accelerators draws 120–200 kW against a 2026 average of about 27 kW, and the US grid interconnection queue exceeds 2,600 GW with waits approaching five years. Whether you can energise this part is now a harder question than whether you can buy it.
Layer 1 — Energy · The interconnection queue · Rack power density · Tokens per watt
Compared against
- Microsoft Maia 200 vs Meta MTIA v2 — Two chips nobody outside their owners can touch. They matter as the GPU orders that never get placed.Compare with Microsoft Maia 200
- Meta MTIA v2 vs NVIDIA B200 — The custom-silicon question in one line: every MTIA Meta deploys is a B200 NVIDIA does not sell.Compare with NVIDIA B200
Related silicon
- AWS Inferentia 2 — Not published in directly comparable terms, cloud only
- Cerebras WSE-3 — 44 GB on-chip SRAM, rent
- Google TPU v7 (Ironwood) — 192 GB HBM3E per chip, cloud only
- Groq LPU — 230 MB SRAM per chip, no HBM, rent
- NVIDIA DGX Spark — 128 GB unified LPDDR5X, buy
- Qualcomm Cloud AI 100 Ultra — 128 GB LPDDR, buy