Meta says its new AI chips will start production in September

In short: Meta plans to begin production of new in-house AI chips in September, according to a report citing an internal memo.

What happened

Meta is on track to start manufacturing the latest versions of its AI-specific chips in September, Reuters reported, citing an internal memo. The chips are part of Meta’s Meta Training and Inference Accelerator program, often called MTIA.

A chip is a small piece of hardware that does math very fast, like a specialized engine. Meta wants these chips to handle both “training” and “inference.” Training is the long practice stage where an AI model learns from lots of examples, and inference is the everyday stage where it gives answers and makes predictions.

The memo said at least one chip completed testing in about six weeks. Meta is working with Broadcom on the chip design, and it plans to use TSMC in Taiwan to manufacture the chips. Reuters also reported Meta is sourcing other parts from several suppliers, including Samsung for memory (RAM) and Sandisk for storage.

Meta has said it is using a modular approach, meaning the chip is built from smaller blocks that can be swapped or updated (like using Lego pieces instead of a single molded part). The idea is to adapt faster as AI needs change.

Why it matters

Big tech companies are spending huge amounts on the computers needed to run AI, and many rely heavily on GPUs from companies like Nvidia and AMD. If Meta’s MTIA chips work well at scale, Meta could reduce some of those purchases and have more control over its AI costs and supply, especially during hardware shortages.

Source: TechCrunch AI