Meta trials in-house AI chip to reduce reliance on external suppliers: Report
The Menlo Park, California-based firm has initiated a small deployment of the chip and is prepared to escalate production if initial trials meet expectations, Reuters reported, citing two sources familiar with the project.
Meta has reportedly begun testing its first in-house chip for training AI (artificial intelligence) models, as it doubles down its efforts to develop its own specialised silicon and lessen its dependence on third-party suppliers like NVIDIA, Reuters reported.
The Menlo Park, California-based firm has initiated a small deployment of the chip and is prepared to escalate production if initial trials meet expectations, added the report, citing two sources familiar with the project.
Meta is collaborating with a Taiwanese chip manufacturer for this effort, as it focuses on building proprietary chips to manage the hefty infrastructure costs related to AI research.
While the company has projected total 2025 expenses of $114 billion to $119 billion, it has also forecasted an estimated $65 billion in capital expenditures largely directed at AI infrastructure.
It further suggested that Meta’s new hardware is a dedicated accelerator, specialised for running AI-specific tasks more efficiently than the general-purpose graphics processing units (GPUs) typically used for AI workloads.
The testing stage began after the firm completed the first “tape-out” of the chip—a milestone in silicon development work where an initial design is forwarded to a foundry. This process can cost tens of millions of dollars and take between three to six months, without any guarantee of success.
The chip is the latest from the company’s “Meta Training and Inference Accelerator” (MTIA) series. Notably, Meta had previously abandoned an in-house custom inference chip after an initial small-scale rollout proved unsuccessful, opting instead to commit billions of dollars to NVIDIA GPUs in 2022.
Despite this shift, Meta has continued to rank among NVIDIA’s largest customers, deploying its GPUs for various AI-driven tasks, including its Llama series of foundational models.
In a similar move, is said to be working on its own artificial intelligence models to rival longtime partner OpenAI, as per a report by The Information. The tech giant has reportedly started evaluating AI solutions from xAI, Meta, and DeepSeek as potential alternatives for OpenAI’s technology in Microsoft 365 Copilot, the company’s flagship AI product.
The dominance of GPUs in AI has been recently called into question following the launch of low-cost models by Chinese startup DeepSeek, which triggered a market shakeup that erased $592.7 billion from NVIDIA’s valuation.
Edited by Jyoti Narayan