To take the initiative in the costly global artificial intelligence competition, social media giant Meta is accelerating its efforts to reduce reliance on external suppliers for its underlying hardware. According to reports, Meta plans to complete the deployment of four generations of self-developed AI chips by the end of 2027. This ambitious roadmap aims to provide computing power support for its rapidly expanding AI business through customized hardware supply, while effectively mitigating long-term dependence on external vendors such as NVIDIA.

Currently, Meta's self-developed chip system has formed a clear iteration hierarchy. The MTIA 300, which focuses on content sorting and recommendation model training, has already entered mass production; the MTIA 400, codenamed "Iris," has passed laboratory testing and entered the deployment phase. More advanced MTIA 450 (codenamed "Alce") and MTIA 500 (codenamed "Astrid") are planned to be launched in the first and second half of 2027, respectively. This high-frequency R&D rhythm reflects Meta's determination to keep hardware evolution aligned with AI algorithm iterations.

Although Meta is investing billions of dollars to build its self-developed chip team and has acquired startups like Rivos to expand its talent pool, its strategic logic is not entirely "closed off." Meta's senior leadership has clearly stated that the company currently follows a dual-track strategy: on one hand, continuing to be one of the world's largest GPU buyers, signing large-scale agreements with NVIDIA and AMD to secure basic computing power; on the other hand, using self-developed chips to eliminate redundant functions in general scenarios, thereby achieving higher efficiency in specialized tasks such as Instagram feed sorting and generative AI inference.

This deep integration of software and hardware is becoming a new moat for top tech companies. Although the chip development cycle usually takes two years and faces significant engineering challenges, Meta firmly believes that custom chips can effectively reduce long-term operating costs by eliminating non-essential features. In the current context of unexpectedly high computing demand, Meta is reviewing and optimizing the technical roadmaps of each generation of MTIA chips, striving to achieve a dynamic balance between self-developed capabilities and external procurement, ensuring its competitive advantage in the field of generative artificial intelligence.