Meta aims to reduce dependence on NVIDIA and accelerates its AI chip development: production starts in September

Meta has initiated the final phase of development for its proprietary AI chips and aims to start production in September 2026. According to reports from Reuters, citing an internal memo, the goal is to cut GPU spending at a time of unprecedented component shortages.

The project is advancing rapidly: at least one of the chips has passed testing in just six weeks. Meta is designing the chips in collaboration with Broadcom, but the physical production is entrusted to TSMC. The supply chain is completed by Samsung's RAM, Sandisk's storage, and Sumitomo Electric's fiber optic components, highlighting how complex the supply chain has become behind a single chip.

The MTIA (Meta Training and Inference Accelerator) program was detailed in March 2026, when Meta introduced four new generations of chips intended to be rolled out in phases between this year and next. The architecture is based on modular chiplets, designed to allow upgrades "in months rather than years" as the demands of AI workload change.

Meta's strategy to reduce dependence on Nvidia

The new chips will primarily serve for training ranking and recommendation algorithms, for broader AI workloads, and for the inference of models used in the group’s apps. However, this does not mean a complete goodbye to third-party GPUs: Meta will continue to spend significant amounts with suppliers such as Nvidia and AMD, albeit reducing their relative weight on its accounts. The production of proprietary silicon has been ongoing since 2023.

Investments remain substantial. Meta indicated in April 2026 a capital expenditure between $125 and $145 billion for the current year, a large part of which is aimed specifically at AI infrastructure. The plan includes deploying 7 gigawatts of computing capacity by the end of the year, with the aim of doubling to 14 gigawatts by 2027.

Additionally, there are other agreements signed in recent months: a deal with Arm for computing power dedicated to recommendation systems, a multi-billion-dollar contract with AMD for the Instinct GPUs, and another agreement of the same magnitude with Amazon for the use of the Seattle group's proprietary CPUs in AI workloads.

Meta is not the only one seeking to stem the flow of capital to Nvidia. Last month, OpenAI presented its own inference processor developed together with Broadcom, while Anthropic is reportedly considering the development of its own chips with Samsung. Amazon and Google, for their part, have long been planning dedicated silicon for training and inference of their models in a market crowded with a host of specialized startups.