In an effort to reduce its spending on graphics processing units (GPUs) during a severe component shortage, Meta is on track to begin mass-producing the latest version of its custom AI chip this September, according to an internal memo cited by Reuters. The memo noted that at least one chip passed its testing phase in roughly six weeks. Meta is collaborating with Broadcom on the chip's design, while Taiwan's TSMC will handle manufacturing. Additionally, the company is procuring RAM from Samsung, storage from Sandisk, and fiber optic equipment from Sumitomo Electric, the report stated.

In March, Meta detailed four new chips developed under its Meta Training and Inference Accelerator (MTIA) program. Some of these chips are already being deployed, while others are scheduled for rollout this year or next. The company is adopting a modular design strategy, anticipating that its needs will shift as AI technology evolves rapidly by the time the chips enter 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.

These custom chips are expected to help Meta reduce its reliance on purchasing GPUs from companies like Nvidia and AMD, though the company still plans to spend heavily with those suppliers as well, according to Reuters. Meta intends to use the MTIA chips for training models that power its ranking and recommendation algorithms, broader AI workloads, and inference tasks for its applications. The social media giant has been producing its own AI chips since 2023.

Meta has been investing heavily to secure the computing capacity needed for its various AI initiatives. In April, the company projected capital expenditures between $125 billion and $145 billion this year, with a significant portion allocated to AI efforts. It has been striking data center and power deals worldwide, spending tens of billions to secure computing capacity for training and deploying its new Muse Spark series of AI models. According to the memo cited by Reuters, Meta plans to deploy 7 gigawatts of compute this year and double that amount next year.

Last year, Meta also signed a deal with ARM to secure computing power for its recommendation systems, in addition to a multi-billion dollar agreement with AMD for its Instinct GPUs and a multi-billion dollar deal with Amazon to use the cloud giant's homegrown CPUs for AI-related needs.

Meta is not alone in trying to curb the flow of capital to Nvidia. Last month, OpenAI unveiled an inference processor it is building with Broadcom, and Anthropic is reportedly considering developing its own chips with Samsung. Amazon and Google both design their own chips for AI training and inference, while a host of startups are entering the space to meet surging demand. Meta declined to comment.