Meta Is Developing 4 New Chips to Power Its AI and Recommendation Systems (www.wired.com)

Meta has announced the development of four new chips designed to enhance its generative AI features and content ranking systems for its platforms, including Facebook and Instagram. These chips, known as the MTIA series (Meta Training and Inference Accelerators), represent a shift for the company, which is stepping into custom silicon production. Partnering with Broadcom and utilizing the open-source RISC-V architecture, the chips are being manufactured by Taiwan Semiconductor Manufacturing Corporation. The MTIA 300 is already in production, with the MTIA 400, 450, and 500 slated for release between early and late 2027. This announcement is significant for the AI/ML community as it reflects a growing trend among tech companies to create tailored hardware that optimally supports their AI workloads—an area that is evolving rapidly and requires adaptive technological solutions. By employing modular chiplets and leveraging insights from current AI model demands, Meta aims to keep pace with the fast-changing landscape of AI technologies. The MTIA 400 boasts performance comparable to leading commercial chips, while improvements in memory and low-precision data handling are expected in subsequent models. Despite this push for in-house chip development, Meta continues to rely on existing vendors like Nvidia and AMD for a significant portion of its AI hardware, highlighting the complexities of building custom silicon.