BofA models $8.5B in 2027 MTIA savings; the next two chip generations get names for the first time

2026-09-15 · part of Meta Capex

What happened

Bank of America estimates Meta could save roughly $8.5 billion in 2027 by running AI workloads on its own custom MTIA silicon instead of third-party chips — a modeled analyst projection, not company guidance — built on an assumed 5-6GW of owned capacity in 2027 at ~$200B total spend, chips at 60% of that, and custom silicon priced ~40% below third-party equivalents. First time this thread has concrete names/timing for the next two MTIA generations: MTIA 450 ("Arke") in H1 2027, MTIA 500 ("Astrid") later in 2027, both co-developed with Broadcom for inference workloads — Broadcom CEO Hock Tan confirmed delivering "three generations of MTIA accelerators to Meta" through end of 2027. Corroborated across Bloomberg, Investing.com, and 24/7 Wall St. (Bloomberg, 24/7 Wall St.)

Sources 2

Bloomberg bloomberg.com high 0.84 disputed / split
24/7 Wall St. 247wallst.com high 0.871

Credibility ratings combine two public datasets: Lin et al. 2023, PNAS Nexus 2(9):pgad286 (CC BY 4.0) for the domain-quality score, and Wikipedia's perennial-sources list (CC BY-SA 4.0) for the reliability tier. A rating describes an outlet's general record, never this particular article. Unrated means absent from those datasets — not a judgement.

The wider story

This event sits inside Meta Capex, which carries how it developed before and since.

Story 10 of 13 in this thread's timeline, oldest to newest

Adjacent threads (shared entities):

· Where the Capex Lands

· Nvidia's Order Book