Meta Unveils Next-Gen In-House AI Chips Amid Strategic Shift from Nvidia and AMD

Meta has officially rolled out its third-generation MTIA (Meta Training Infrastructure Accelerator) chips, designed to power its expanding AI workloads across global data centers. The MTIA-3 processors deliver 40% higher performance per watt compared to the previous generation, enabling more efficient training of large language models at scale. This advancement supports Meta’s plan to deploy 12 new data center facilities by 2027, with the new chips expected to underpin 70% of its AI training infrastructure by year-end. The timing of the launch—just days after Meta executed agreements valued at $6.8 billion with Nvidia and $3.2 billion with AMD—underscores a deliberate strategy to balance in-house innovation with continued reliance on external suppliers. While the company continues to purchase high-performance GPUs for immediate deployment, the MTIA-3 rollout reduces long-term dependency on third-party silicon, particularly Nvidia’s H100 and Blackwell series. Analysts note that Meta’s growing chip autonomy could reshape semiconductor demand dynamics. As in-house AI accelerators capture a larger share of internal workloads, the incremental demand for external GPUs may slow, potentially affecting revenue growth trajectories for Nvidia and AMD. The shift also introduces long-term pricing power considerations, as Meta gains leverage over external suppliers through vertical integration. The market reaction has been mixed: Nvidia (NVDA) shares dipped 2.3% following the announcement, while AMD (AMD) edged up 0.7%. The broader VIX index rose 1.1%, reflecting increased volatility in tech-oriented equities amid shifting AI infrastructure strategies.