Meta's Custom AI Chips Enter Production in September as GPU Costs Surge

Meta is on track to begin manufacturing the latest versions of its custom AI chips in September, according to an internal memo cited by Reuters. The move comes as the company seeks to lower its GPU costs during an unprecedented component shortage that has strained the broader tech industry.

At least one of the new chips completed its testing phase in approximately six weeks, the memo noted. Meta has partnered with Broadcom on the chip design, while Taiwan's TSMC will handle manufacturing. The company is also sourcing RAM from Samsung, storage from Sandisk, and fiber optic equipment from Sumitomo Electric, according to the report.

MTIA Program: A Modular Approach to AI Hardware

The four new chips were developed under Meta's Meta Training and Inference Accelerator (MTIA) program, which the company first detailed in March. Some of these chips are already in deployment, while others are expected to roll out later this year or next.

Meta has taken a modular approach to designing the chips, anticipating that its needs will shift as AI technology evolves rapidly by the time the chips reach production. The company emphasized that each generation of MTIA chips is built to adapt to changing workloads and hardware advancements.

"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 in March.

The MTIA chips are intended for training models that power Meta's ranking and recommendation algorithms, as well as broader AI workloads and inference tasks across its applications. Meta has been producing its own AI silicon since 2023, marking a growing push toward self-reliance in critical hardware.

Cutting GPU Costs Amid Massive Spending

The new chips are expected to help Meta reduce its purchases of GPUs from chipmakers such as Nvidia and AMD, though the company still anticipates spending significantly with those providers. The strategy reflects an effort to balance proprietary silicon with third-party hardware as AI computing demands continue to escalate.

Meta's spending on computing infrastructure has reached extraordinary levels. In April, the company projected capital expenditures between $125 billion and $145 billion for this year alone, with a substantial portion directed toward its AI initiatives.

The company has been securing data center and power agreements across the globe, investing tens of billions to obtain enough computing capacity to train and deploy its new Muse Spark series of AI models. Meta plans to deploy 7 gigawatts of compute this year and intends to double that figure next year, according to the memo cited by Reuters.

Beyond its custom chips, Meta has pursued multiple hardware partnerships to bolster its AI infrastructure. Last year, the company signed a deal with ARM to secure compute for its recommendation systems. It also reached a multi-billion dollar agreement with AMD for Instinct GPUs and a separate multi-billion dollar deal with Amazon to use the cloud provider's homegrown CPUs for AI-related workloads.

A Broader Industry Push Toward Custom Silicon

Meta is not alone in its drive to develop custom AI silicon. OpenAI last month unveiled an inference processor that it is building in collaboration with Broadcom. Anthropic is reportedly considering developing its own chips with Samsung as a partner.

Amazon and Google both already produce their own chips for AI training and inference, and a growing number of startups are building specialized hardware to meet skyrocketing demand across the industry.

The trend highlights a broader effort among major technology companies to reduce their dependence on Nvidia, which has dominated the AI chip market. By developing proprietary silicon, these firms aim to gain greater control over costs, performance, and supply chains at a time when computing capacity has become a critical competitive advantage.

Meta declined to comment on the reported plans.

As the AI hardware race intensifies, Meta's September production milestone represents a significant step in the company's strategy to build a more self-sufficient AI infrastructure. Whether custom chips can meaningfully offset the staggering costs of AI development remains an open question. Share this article with your network and let us know your thoughts on the future of AI hardware.

Source: TechCrunch AI