Meta's new AI chips will begin production in September

Meta is on track to begin producing the latest versions of its AI-specific chips in September, according to an internal memo cited by Reuters. The move is part of the company's efforts to lower GPU costs amid an unprecedented component shortage. At least one chip sailed through its testing phase in about six weeks, the memo said.

Meta is working with Broadcom on chip design, while Taiwan's TSMC will handle manufacturing. The company is also buying RAM from Samsung, storage from Sandisk, and fiber optic equipment from Sumitomo Electric. Meta detailed four new chips under its Meta Training and Inference Accelerator (MTIA) program in March, some of which are either already deployed or will be deployed this year or next.

The company is taking a modular approach to chip design, anticipating that needs will change as AI evolves rapidly by the time the chips are in production. "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," Meta wrote at the time.

The chips are expected to help the company save on buying GPUs from chipmakers like Nvidia and AMD, although it still expects to spend heavily with those providers as well. Meta intends to use the MTIA chips for training models for its ranking and recommendation algorithms, broader AI workloads, and inference across its applications. The social media company has been producing its own AI chips since 2023.

Meta has been spending massively on securing computing capacity for its AI efforts. In April, the company said it expects capital expenditures between $125 billion and $145 billion this year, much of which goes toward AI. It has struck data center and power deals worldwide, spending tens of billions to secure capacity for training and deploying its new Muse Spark series of AI models. According to the memo, it plans to deploy 7 gigawatts of compute this year and double that next.

Meta also signed a deal with ARM last year for compute for its recommendation systems, along with multi-billion dollar deals with AMD for its Instinct GPUs and with Amazon to use the cloud giant's homegrown CPUs for AI needs.

Meta is not alone in trying to reduce capital flowing to Nvidia. OpenAI last month unveiled an inference processor built with Broadcom, and Anthropic is reportedly considering developing its own chips with Samsung. Amazon and Google already develop their own chips for AI training and inference, and numerous startups are entering the space to meet surging demand. Meta declined to comment.