Meta Developed 4 New Chips to Power Its AI and Recommendation Systems

Meta announced Wednesday that it has developed four new computer chips that will be used to power generative AI features and content ranking systems within the tech giant’s own apps. The hardware will become part of Meta’s existing chip line known as MTIA, or Meta Training and Inference Accelerators.

Meta partnered with Broadcom to develop its latest semiconductors, which are built on top of the open-source RISC-V architecture. They’re being fabricated by Taiwan Semiconductor Manufacturing Corporation, the world’s leading chip producer.

One of the new MTIA chips, the MTIA 300, is already in production, and Meta says the other three—the MTIA 400, 450, and 500—are expected to ship sometime between early and late 2027. Putting out fresh silicon this quickly is unusual by most chip industry standards, but is essentially unheard of for a social media company that historically didn’t produce its own physical computing infrastructure.

YJ Song, a vice president of engineering at Meta, says that AI models are evolving faster than traditional chip development cycles, so AI workloads may change substantially by the time new hardware typically reaches production. “Rather than placing a bet and waiting for a long period of time, we deliberately take an iterative approach. Each MTIA generation builds on the last, using modular chiplets and incorporating the latest AI workload insights and hardware technologies,” Song said in a blog post.

The MTIA 300 will be used primarily for training algorithms that rank and recommend content to the hundreds of millions of people who use apps like Facebook and Instagram each day. The other three chips are designed to support inference, the process of running trained AI models to produce outputs like text or images.

The MTIA 400, which Meta claims delivers performance “competitive with leading commercial products,” has been tested and is expected to arrive at data centers soon. The MTIA 450 will have double the amount of high-bandwidth memory as the MTIA 400, and is supposed to ship in early 2027. Meta says the MTIA 500, which is slated to arrive later next year, will have even more memory than MTIA 450 and include “innovations in low-precision data.”

The MTIA chips are part of Meta’s broader strategy to hoard as much computing power as possible in order to develop cutting-edge artificial intelligence. Meta first shared details about its chip development plans in 2023, when it released its first product under the MTIA banner. As software companies and AI labs continue to train increasingly powerful AI models, they have begun announcing ambitious plans to build custom chips that serve their own specific AI needs. OpenAI, for example, has also said it’s partnering with Broadcom to build custom accelerators, following a path similar to Meta’s.

Earlier this year, Meta was reported to be scaling back some of its in-house efforts to make high-end chips that would compete more directly with leading players like Nvidia. The company now appears eager to dispel that narrative by announcing this new road map for MTIA chips. But making custom silicon remains enormously expensive and technically complex, which means Meta will likely continue purchasing the majority of its AI hardware from other firms, at least in the near future.

That reality is reflected in the company’s recent chip buying spree. Meta unveiled its new MTIA chips shortly after announcing multibillion dollar deals with Nvidia and AMD. The social media giant has also signed an agreement to rent chips made by Google.