Meta, the parent company of Facebook, is currently testing its first proprietary chip designed for training artificial intelligence (AI) systems. This marks a significant step in the company’s efforts to build its own custom silicon and lessen its reliance on third-party suppliers such as Nvidia, according to two sources speaking to Reuters.

The tech giant has initiated a limited rollout of the chip and, if testing proves successful, plans to expand production for broader deployment, the sources revealed.

Developing in-house chips is a key part of Meta’s long-term strategy to reduce infrastructure costs as the company invests heavily in AI-driven technology to fuel growth.

Meta, which also owns Instagram and WhatsApp, has projected its total expenses for 2025 to range between $114 billion and $119 billion. A substantial portion of this, up to $65 billion, is allocated toward AI-related capital expenditures.

One of the sources says Meta’s new training chip is a dedicated accelerator, meaning it is designed to handle only AI-specific tasks. This can make it more power-efficient than the integrated graphics processing units (GPUs) generally used for AI workloads.

Meta has partnered with Taiwan-based semiconductor manufacturer TSMC to make the chip, the source added.

Testing began following the completion of the first “tape-out” phase, a crucial milestone in chip development where the initial design is sent to a factory for production, the second source explained.

A tape-out typically costs tens of millions of dollars and requires three to six months to complete. There is no certainty of success, and if the chip fails, Meta would need to analyze and troubleshoot the issue before repeating the process.

Meta and TSMC declined to comment on the development.

This chip is the latest under Meta’s Meta Training and Inference Accelerator (MTIA) series, which has faced hurdles in the past, including the cancellation of a previous design at a similar development stage.

Nevertheless, in 2023, Meta introduced an MTIA chip for AI inference—running AI models in real-time—within its recommendation systems that personalize content feeds on Facebook and Instagram.

Meta executives have indicated they aim to begin using their own chips for AI training by 2026. AI training involves processing massive amounts of data to improve machine learning models.

Initially, the training chip will be deployed in recommendation systems, with plans to extend its use to generative AI applications, including the Meta AI chatbot, executives stated.

“We’re working on how would we do training for recommender systems and then eventually how do we think about training and inference for gen AI,” Meta’s Chief Product Officer Chris Cox said at the Morgan Stanley technology, media, and telecom conference last week.

Cox described Meta’s chip development efforts as “kind of a walk, crawl, run situation” so far, but said executives considered the first-generation inference chip for recommendations to be a “big success.”

Previously, Meta had scrapped an in-house inference chip after it failed in a small-scale test, leading the company to invest billions in Nvidia GPUs in 2022 instead.

Since then, Meta has become one of Nvidia’s largest customers, amassing GPUs to train and run AI models for recommendation systems, advertising, and the Llama foundation model series. These chips also support inference tasks for over 3 billion users interacting with Meta’s applications daily.

However, the effectiveness of large-scale GPU investments has come under scrutiny as AI researchers question whether further scaling of large language models will yield significant advancements.

This skepticism gained momentum after Chinese startup DeepSeek launched cost-efficient AI models in early 2025 that prioritize inference over traditional training methods.

DeepSeek’s breakthrough caused a temporary market downturn in AI stocks, with Nvidia’s shares dropping by as much as 20% before recovering most of the losses. While investors continue to see Nvidia’s chips as the industry benchmark for AI workloads, recent concerns over trade policies have led to fluctuations in the stock.

Source: Reuters

If you like this post, please share it with others on social media. Follow Anchor Biz IT on LinkedIn.