Reports suggest that Meta is testing its first in-house AI training chip and that the chip will be built on TSMC’s 5nm process, with mass production set for 2026. Meta’s MTIA v2 chip will also leverage TSMC’s CoWoS technology, with the design supported by IPs from Broadcom.
For the moment, Meta remains one of NVIDIA’s top customers, using the company’s GPUs to train models for ads, recommendations, and its Llama series, while also handling AI inference for billions of daily users.
Meta’s decision to develop and manufacture its own AI chips, however, comes at a time when DeepSeek’s breakthrough has raised some serious questions around future chip demand as well as the costs associated with chip development.
Meta is not alone in its reliance on Nvidia and Google, Amazon and OpenAI are all said to be looking for alternatives.
Meta is also focused on developing its own chips in a bid to cut infrastructure costs. Not a surprise, when it has allocated $65 billion alone for AI-driven capital spending this year.
The market for AI is surely big enough for Nvidia and the likes of Meta to develop and deploy chips developed in-house, but it does suggest that it's a market that's going to get pretty crowded going forward.
