Meta is preparing to move its latest AI chip designs into production in September, a step aimed at easing reliance on expensive GPUs and strengthening its in-house computing strategy. The plan reflects the company's push to build more of the hardware behind its AI systems.
According to an internal memo, at least one chip completed testing in roughly six weeks. Meta is developing the chips with Broadcom while manufacturing will be handled by TSMC. The supply chain also includes memory from Samsung, storage from Sandisk, and fiber optic components from Sumitomo Electric.
The new processors belong to Meta's MTIA program, which focuses on training and inference acceleration. The company says its modular chiplet approach is designed to adapt quickly as AI workloads evolve. Some of the new generations are already in use or expected to roll out over the next year.
Meta plans to use the chips for ranking and recommendation systems, broader AI tasks, and inference across its apps. Since launching its own chip efforts in 2023, the company has continued to expand its computing footprint through major infrastructure and power investments.
With capital spending projected between $125 billion and $145 billion this year, Meta is building the foundation for larger-scale AI deployment. Its strategy also includes partnerships and custom hardware deals across the industry, showing how fast the race for efficient AI infrastructure is accelerating. The next phase of this shift could reshape how digital platforms design, train, and deliver intelligent services.
