Meta Platforms has announced an accelerated roadmap to deploy four in‑house AI chips over the next two years as it scrambles to meet surging compute demand. The company said it is advancing its MTIA (Meta Training and Inference Accelerator) family — MTIA 300, 400, 450 and 500 — with MTIA 300 already in production to support core ranking and recommendation training workloads.
Meta says the later designs (400, 450 and 500) will be capable of handling the full range of its internal workloads, and that the firm intends to rely on these chips primarily for inference tasks through 2027 to shore up capacity for generative AI. Engineering leadership has framed the programme as iterative and responsive to rapidly changing model requirements, with MTIA 450 targeted for early 2027 and MTIA 500 expected roughly six months after that.
The initiative is part of a broader strategy to diversify hardware sources and reduce dependence on external suppliers such as Nvidia and AMD, which have supplied the bulk of datacentre GPUs driving recent AI advances. Meta will, however, maintain a hybrid approach: it continues to buy third‑party accelerators and has committed billions of dollars in recent purchases of Nvidia and AMD hardware even as it expands its internal silicon effort.
Meta’s chip push follows earlier recruitment and acquisition moves aimed at building a bespoke silicon capability. After an unsuccessful bid for South Korea’s FuriosaAI, the company acquired California‑based Rivos Inc. and absorbed more than 400 employees, reflecting a “dual‑track” strategy of buying conventional hardware while investing in customised designs tailored to Meta’s platform needs.
The firm’s stated design philosophy emphasises specialization rather than generality: by removing features unnecessary for Meta’s workloads, the company hopes to drive down costs per unit of useful compute. That approach mirrors a wider industry trend in which the largest cloud and tech firms build vertically integrated stacks — from models to chips — to extract efficiency gains and control supply chains.