Meta's Next-Gen AI Chips to Begin Production in September, Designed with Broadcom, Manufactured by TSMC

According to TechCrunch, citing an internal memo, Meta is on track to start production of the latest versions of its AI-specific chip in September. At least one chip sailed through its testing phase in about six weeks, the memo said.

Meta is working with Broadcom on chip design and will use Taiwan Semiconductor Manufacturing Company (TSMC) for manufacturing. The company is also purchasing RAM from Samsung, storage from Sandisk, and fiber-optic equipment from Sumitomo Electric, according to the report.

These chips are part of Meta’s Training and Inference Accelerator (MTIA) program. Meta detailed four new chips in March, some of which are currently in deployment or will be deployed this year or next. The company is taking a modular chiplet approach, anticipating that their needs will change as AI evolves rapidly.

Meta stated: “Each MTIA generation builds on the last, using modular chiplets, incorporating the latest AI workload insights and hardware technologies, and deploying on a shorter cadence.”

The custom chips are expected to help Meta save on buying GPUs from Nvidia and AMD, though it still expects to spend heavily with those providers as well. The MTIA chips will be used for training ranking and recommendation algorithms, broader AI workloads, and inference for Meta’s applications. Meta has been producing its own AI chips since 2023.

Meta’s spending on compute capacity is enormous. The company said in April it expects capital expenditures between $125 billion and $145 billion this year, much of which is going toward AI. Meta plans to deploy 7 gigawatts of compute this year and double that next year.

Meta is not alone in trying to reduce dependence on Nvidia. OpenAI last month unveiled an inference processor built with Broadcom, and Anthropic is reportedly considering developing its own chips with Samsung.