Meta to Begin Manufacturing Own AI Chips in September 2026 to Reduce Nvidia Dependence

Meta Platforms (recognized as an extremist organization, activity is prohibited in Russia) expects to begin production of its own chip for artificial intelligence systems as early as September 2026 — according to Reuters citing an internal company memo.

Why Meta needs its own chips

The company has been almost entirely dependent on Nvidia's graphics processors for training and running its AI models — as have most major technology players. Developing its own chip should reduce this dependency and give Meta greater control over the cost and availability of computing power, which remains the main bottleneck for the entire AI industry. The company unveiled its first chips in the MTIA (Meta Training and Inference Accelerator) line in 2023, but used them mainly for inference tasks rather than for training truly large models.

• Production of its own chip is planned to begin in September 2026
• Goal — to reduce dependency on Nvidia chips
• The company intends to increase computing power to 14 GW by 2027
• Source of information — an internal Meta memo that Reuters learned about

How much computing power Meta needs

14 gigawatts of computing power that the company expects to have by next year is orders of magnitude more than what most existing data centers consume: for comparison, one gigawatt can supply electricity to hundreds of thousands of homes. Such scale is explained by the fact that Meta continues to invest in training increasingly large models in the Llama family and in infrastructure for its own AI products within Facebook (social network is prohibited in Russia) and Instagram (service is prohibited in Russia). The growing energy consumption of AI company data centers is already noticeable beyond the technology industry: building new capacity requires not only chips but also long-term contracts for electricity supplies, including agreements with nuclear and gas power plants.

Why this is a strategic issue for the entire industry

Demand for Nvidia chips so exceeds supply that major buyers have stood in line for deliveries for years and are forced to accept prices set by Nvidia itself — the company has essentially become the main bottleneck for the entire AI industry. The emergence of proprietary chips from major customers does not completely remove Nvidia from the market: training the largest models still requires top-tier GPUs, and proprietary developments often cover narrower tasks like inference — that is, direct model responses to user queries, which are performed billions of times a day and are cheaper on specialized hardware.

What this means

Proprietary AI chips allow major tech companies to be less dependent on Nvidia and its pricing policy: Google has already taken a similar path with TPU chips and Amazon with Trainium chips, and Meta's entry into this market further intensifies competition for component supplies and data center capacity. For Nvidia, it is a signal that even its largest customers are seeking to diversify computing suppliers, and for the market as a whole — that the race for proprietary AI silicon is becoming the norm for all industry giants, not an exception.