Meta is set to commence production of its latest AI-specific chips in September, aiming to reduce reliance on external GPU suppliers amid ongoing component shortages. This development follows the successful testing of at least one chip, which completed its evaluation phase in approximately six weeks.
Collaborating with Broadcom on chip design, Meta will utilize Taiwan Semiconductor Manufacturing Company (TSMC) for manufacturing. Additionally, the company is sourcing RAM from Samsung, storage solutions from SanDisk, and fiber-optic equipment from Sumitomo Electric to support this initiative.
In March, Meta unveiled four new chips under its Meta Training and Inference Accelerator (MTIA) program. These chips are designed to support ranking and recommendation systems, as well as broader AI workloads. The company has adopted a modular approach to chip design, allowing for rapid iteration and adaptation to the fast-evolving AI landscape. Each MTIA generation builds upon the previous one, incorporating the latest insights and technologies to meet the dynamic demands of AI applications.
By developing in-house AI chips, Meta aims to reduce expenditures on GPUs from companies like Nvidia and AMD. However, the company anticipates continued significant investments with these providers. The MTIA chips are intended for training models that power Meta’s ranking and recommendation algorithms, as well as for broader AI workloads and inference tasks across its suite of applications.
Since 2023, Meta has been producing its own AI chips to bolster its AI infrastructure. The company has been making substantial investments to secure the necessary computing capacity for its AI initiatives. In April, Meta projected capital expenditures between $125 billion and $145 billion for the year, with a significant portion allocated to AI development.
To support its AI efforts, Meta has been entering into data center and power agreements worldwide, investing tens of billions to ensure adequate computing capacity for training and deploying its new Muse Spark series of AI models. The company plans to deploy 7 gigawatts of compute capacity this year, with plans to double that in the following year.
In addition to its in-house chip development, Meta has secured agreements with ARM for its recommendation systems and has entered into multibillion-dollar deals with AMD for its Instinct GPUs and with Amazon to utilize the cloud giant’s homegrown CPUs for AI-related needs.
Meta’s strategic move to develop custom AI chips reflects a broader industry trend among tech giants to create proprietary hardware tailored to their specific AI workloads. This approach not only aims to optimize performance and efficiency but also to reduce dependency on external suppliers, thereby enhancing control over their AI infrastructure.
