en.Wedoany.com Reported - Meta plans to begin mass production of its self-developed artificial intelligence (AI) chips starting in September, aiming to accelerate its self-reliance in AI infrastructure. The company targets doubling its data center computing capacity next year while reducing dependence on NVIDIA by integrating its own semiconductors with cloud infrastructure, thereby enhancing the competitiveness of its AI services.

Reuters reported, citing an internal Meta memo, that Meta plans to start mass production of a data center AI chip codenamed "Iris" in September this year. The chip is part of Meta's fourth-generation "Meta Training and Inference Accelerator (MTIA)" project, with Broadcom providing design support and TSMC handling production. Chip testing was completed within six weeks.

Iris is a custom semiconductor developed by Meta to enhance AI learning and inference performance in its own services such as Facebook and Instagram. This move aims to use self-developed chips to reduce massive AI computing costs while lowering reliance on external graphics processing unit (GPU) suppliers like NVIDIA and AMD.

After unveiling four self-developed AI processors in March this year, Meta plans to release next-generation chips at approximately six-month intervals, aiming to secure AI semiconductor competitiveness at a faster pace than the industry's typical development cycle of over a year.

Investment in AI infrastructure is also expanding. Meta plans to build a total of 7 gigawatts (GW) of computing infrastructure this year, adding an equivalent amount next year to expand total computing capacity to 14 GW. This investment aims to accelerate the expansion of data centers used for training and inference of its self-developed AI models.

The expansion of large-scale computing infrastructure aligns with Meta's recent hints at a cloud strategy. It is understood that while strengthening AI capabilities based on its own data centers, Meta will also provide idle AI infrastructure resources to external customers in the form of cloud services. Some analysts believe that using self-developed chips can reduce GPU procurement costs and improve data center operational efficiency, thereby helping to enhance cloud business profitability.

Meta plans to invest up to $145 billion (approximately 218 trillion Korean won) in AI infrastructure this year, accounting for about one-fifth of the expected $700 billion (approximately 1,057 trillion Korean won) in total AI investment by tech giants this year.

News of supply chain security also emerged. The internal memo shows that Meta has signed long-term supply contracts with Samsung Electronics for memory chips, SanDisk for flash storage, and Sumitomo Electric for fiber optic equipment, to address potential shortages of memory and semiconductors due to the expansion of AI data centers.

On the same day, Meta publicly released its AI coding model "Muse Spark 1.1" as a paid API, moving away from its previous open-source-focused strategy and expanding its self-developed AI model business. Driven by expectations of such investment expansion, Meta's stock price rose 4.7% from the previous trading day to close at $631.5. Although it showed weakness in early trading, the stock turned upward after news of the AI chip mass production plan and AI service expansion emerged.

Forrester Vice President Mike Gualtieri commented that relying on other companies' chips cannot make one an AI giant, and the only practical way to ensure model price competitiveness is to develop self-developed chips like Meta does.