Meta Unveils Four New AI Chips With Broadcom
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Meta developed four new MTIA chips—MTIA 300, 400, 450, and 500—with Broadcom, using RISC-V architecture and fabricated by TSMC, to accelerate AI training and inference workloads.
- Meta stated the MTIA 300 is already in production and will be used for training content-ranking algorithms, while the MTIA 400, 450, and 500 will support AI inference and ship between early and late 2027.
- YJ Song, a vice president of engineering at Meta, explained the company is using an iterative, modular chiplet approach to keep pace with rapidly evolving AI models, departing from traditional long chip development cycles.
- Meta claims the MTIA 400 delivers performance competitive with leading commercial products and is being deployed to data centers, while the 450 and 500 will offer increased high-bandwidth memory and low-precision data innovations.
- Meta continues to rely heavily on external AI hardware, recently signing multibillion-dollar deals with Nvidia and AMD and an agreement to rent Google’s chips, despite expanding its custom silicon roadmap.
Why it matters: Meta’s rapid cadence of custom AI chips breaks the norm for social media companies and signals a strategic pivot toward hardware control, but its continued reliance on Nvidia, AMD, and Google shows custom silicon won’t replace external suppliers soon — a costly dual-track strategy that reflects the high stakes of AI infrastructure dominance.
