Meta's coverage in our briefs centres on three interacting positions: the MTIA custom-silicon ramp, the AMD MI450 training commitment, and the implied role as a CoreWeave material customer.
The MTIA / Broadcom partnership. Per the May 17 brief: Broadcom's 3-year partnership with Meta for custom AI chips was confirmed May 15. MTIA is Meta's in-house ASIC line, designed in collaboration with Broadcom for inference (and increasingly training) workloads. Per the May 13 brief: Meta is one of four hyperscalers named in the Broadcom $20bn → $100bn AI revenue trajectory. Meta's MTIA volumes are a meaningful share of that growth curve.
The AMD MI450 / Helios commitment. Per the May 17 brief: "AMD MI450/Helios with Meta + OpenAI as customers for 6 GW of training capacity." The Meta + OpenAI combination is a substantial check on Nvidia's training share. If MI450 pricing on launch undercuts B200, Meta is one of the two largest customers driving that competitive pressure. The 6 GW figure is large enough to be material at the cluster-economics level for both AMD and the Nvidia counterfactual.
The CoreWeave customer concentration. Like Microsoft, Meta is widely believed to be Customer A or B in CoreWeave's Q1 2026 10-Q disclosure of "significant customer concentration risk." Meta's simultaneous MTIA buildout (custom-silicon insourcing) is therefore a direct headwind for CoreWeave's revenue retention — Meta is one of the firms most likely to reduce H100 rental commitments as MTIA capacity comes online.
The Llama / inference scale context. Meta's Llama family is among the largest open-weight model lines, generating massive internal inference demand. Llama serving on MTIA (vs Nvidia GPUs) determines a material share of Meta's AI infrastructure economics over the next 18 months. The internal Meta workloads MTIA is built to serve are precisely the kind of high-volume inference where ASIC TCO advantages are largest.
The composite picture: a name that is simultaneously a meaningful substitution-arc beneficiary (MTIA + AMD MI450) and a non-trivial source of substitution risk for downstream names (CoreWeave). Coverage in our briefs has been substantive but spread across multiple second-order interpretations rather than concentrated on a single deal.