Doubling Down: Meta Launches Mass Production of Own AI Chips in September

Meta is officially announcing a transition to proprietary hardware. On July 9, 2026, the company confirmed that mass production of the MTIA (Meta Training and Inference Accelerator) chip will begin this September.

The company’s goal is to double its computing capacity by the end of the year. As with OpenAI and their Jalapeño chip, Meta’s strategy is aimed at minimizing dependence on NVIDIA and optimizing operating expenses. Using its own silicon allows Meta not just to reduce the cost of inference but also to ensure unique architectural adaptation for the specifics of its social networks and Llama models. The move of major hyperscalers toward proprietary ASIC solutions is the main trend for the second half of 2026, signifying the end of the era of universal GPU omnipotence in the B2B infrastructure market.

Source: Meta / Reuters