Key Takeaways
- Meta AI chips will launch in 2026 as part of the company’s strategy to develop in-house processors for AI systems.
- The chips, developed under the Meta Training and Inference Accelerator program, include models MTIA 300, MTIA 400, MTIA 450, and MTIA 500.
- MTIA 300 focuses on recommendation algorithms, while MTIA 400 is optimized for generative AI and has passed internal testing.
- Future chips MTIA 450 and MTIA 500 will enhance memory capacity and processing performance for inference operations.
- Meta partners with Broadcom and Taiwan Semiconductor Manufacturing Company for manufacturing, using open-source RISC-V architecture.
Meta AI chips will begin large-scale deployment in 2026. The company announced plans to produce batches of its own artificial intelligence processors. The initiative is part of Meta’s strategy to build internal computing hardware. These Meta AI chips are designed to support artificial intelligence systems across the company’s platforms. The effort also aims to reduce reliance on external chip providers.
Meta AI Chips Developed Under the MTIA Program
The Meta AI chips belong to the Meta Training and Inference Accelerator program. The company introduced four chip models under this system. The models are MTIA 300, MTIA 400, MTIA 450, and MTIA 500. Each processor is built to handle specific AI workloads. These include training and inference tasks used by artificial intelligence models.
MTIA 300 has already reached the production stage. This processor focuses on recommendation algorithms. These algorithms rank and deliver content on services such as Facebook and Instagram.
Meta AI Chips Designed for Generative AI and Data Centers
The MTIA 400 chip is designed for generative artificial intelligence workloads. The processor has already undergone internal testing. Meta plans to deploy it in company data centers. The chip supports AI models used in content generation and other advanced applications.
Two additional Meta AI chips are planned for future release. MTIA 450 will include larger high-bandwidth memory capacity. MTIA 500 will add further improvements in memory size and processing performance. These designs will mainly support inference operations.
Meta AI Chips Manufacturing and Industry Partnerships
Meta developed the AI processors with assistance from Broadcom. Taiwan Semiconductor Manufacturing Company handles fabrication. The chips use the open-source RISC-V architecture.
Custom AI chips perform specialized computing tasks. They differ from general-purpose graphics processing units commonly used for artificial intelligence.
Meta continues to invest heavily in AI infrastructure. The company also signed agreements with Nvidia and Advanced Micro Devices. These deals provide access to additional processors and computing capacity.
Meta first revealed its custom chip program in 2023. The expanded Meta AI chips roadmap now includes multiple generations of processors scheduled for future deployment.
Source: https://www.reuters.com/world/asia-pacific/meta-unveils-plans-batch-in-house-ai-chips-2026-03-11/