Meta Platforms is aggressively pursuing hardware autonomy to fuel its AI ambitions. The company is reportedly in talks for a $6.5 billion agreement with Samsung Foundry to manufacture the third generation of its proprietary MTIA (Meta Training and Inference Accelerator) chips. This strategic pivot involves adopting a cutting-edge 2nm process, marking a significant shift away from its long-term reliance on TSMC.
Breaking the NVIDIA Dependency
The move is designed to decouple Meta's growth from the supply constraints and pricing power of NVIDIA GPUs. By designing its own silicon, Meta can optimize performance specifically for its AI workloads while securing a more resilient supply chain. This hardware push is critical to achieving Meta's massive infrastructure goal: reaching 5 gigawatts of computing capacity by 2030 to support its evolving AI and cloud ecosystem.
Samsung's Ascent in the Custom Silicon Market
Meta is not alone in turning to South Korea. Samsung Foundry is rapidly becoming a powerhouse for custom AI silicon, securing major wins from Tesla and Anthropic. As previously noted by AlexTech.ai, Anthropic is also exploring Samsung's 2nm capabilities and advanced packaging technologies to bring its main processors closer to memory chips, enhancing overall system efficiency.
The 2nm Frontier
The transition to the 2nm node is expected to deliver superior processor density and power efficiency, which are essential for scaling large-scale AI models. To ensure success, Meta is reportedly collaborating closely with Samsung’s System LSI division on the chip's architecture, aiming to maximize throughput while minimizing the energy footprint of its global data center network.
No comments yet. Be the first!