Meta’s custom AI chips are set to begin production in September, targeting GPU cost relief

Meta Platforms is moving forward with its plan to produce the latest versions of its custom artificial intelligence chips, with manufacturing expected to begin in September, according to an internal memo reviewed by Reuters. The move is part of a broader effort by the social media giant to reduce its heavy reliance on expensive graphics processing units (GPUs) from Nvidia and AMD amid a global component shortage.

Production timeline and partners

The memo indicates that at least one of Meta’s new chip designs successfully passed its testing phase in about six weeks. Meta is working with Broadcom on the chip’s architecture, while Taiwan Semiconductor Manufacturing Co. (TSMC) will handle fabrication. The company is also sourcing RAM from Samsung, storage from Sandisk, and fiber optic equipment from Sumitomo Electric for the related infrastructure.

Meta first detailed the four new chips under its Meta Training and Inference Accelerator (MTIA) program in March. Some of these chips are already in deployment or are scheduled to roll out this year and next. The company is taking a modular approach to chip design, anticipating that its computational needs will shift as AI technology evolves rapidly during the production cycle.

Why this matters for Meta’s AI ambitions

The MTIA chips are designed to handle training for Meta’s ranking and recommendation algorithms, broader AI workloads, and inference tasks for its applications. By producing its own silicon, Meta aims to lower its capital expenditure on third-party GPUs, though the company has made clear it will continue to spend heavily with Nvidia, AMD, and others.

Meta’s infrastructure spending has surged. In April, the company said it expects capital expenditures between $125 billion and $145 billion this year, much of it directed toward AI compute capacity. The company has been signing data center and power deals worldwide, spending tens of billions to secure capacity for training and deploying its Muse and Spark series of AI models. According to the memo, Meta plans to deploy 7 gigawatts of compute this year and double that in 2026.

Industry context: A wave of custom silicon

Meta is not alone in trying to reduce its dependence on Nvidia. OpenAI last month unveiled an inference processor built with Broadcom, and Anthropic is reportedly exploring its own chip designs with Samsung. Amazon and Google have long developed custom chips for AI training and inference, and a growing number of startups are entering the space to meet surging demand.

Meta also signed a deal with ARM last year to secure compute for its recommendation systems, along with a multi-billion-dollar agreement with AMD for its Instinct GPUs and a similar deal with Amazon to use its homegrown CPUs for AI workloads.

Conclusion

Meta’s push into custom AI chip production represents a strategic shift to gain more control over its hardware supply chain and reduce costs. With production set to begin in September, the company is positioning itself to handle the enormous compute demands of next-generation AI models while maintaining flexibility in a rapidly changing technological landscape. Meta declined to comment on the report.

FAQs

Q1: What are Meta’s MTIA chips used for?
MTIA chips are designed for training and running AI models, including Meta’s ranking and recommendation algorithms, as well as broader AI workloads and inference for its apps.

Q2: When will Meta’s new AI chips go into production?
According to an internal memo, production of the latest MTIA chips is scheduled to begin in September.

Q3: How does this affect Meta’s spending on Nvidia GPUs?
While Meta aims to reduce its GPU costs by using its own chips, it still expects to spend heavily with Nvidia, AMD, and other providers for the foreseeable future.