Meta’s MTIA Custom AI Silicon at Hot Chips 2026

At Hot Chips 2026, Meta is presenting its MTIA family of AI inference accelerators, revealing an ambitious roadmap to launch four new accelerators within the next few years. Simultaneously, NVIDIA is showcasing Groq 3 LPU accelerators integrated into Vera Rubin clusters. These specialized chips, sourced from Groq, are designed for heterogeneous AI compute to significantly reduce latency during the decode phase of inference, enhancing overall performance for complex AI hardware deployments across the industry.