The competition in the artificial intelligence (AI) sector is intensifying, with significant demand for NVIDIA's AI chips that power large language models (LLMs). In response, ByteDance is reportedly planning to develop its own chips to reduce reliance on NVIDIA's products. According to The Information, ByteDance is collaborating with TSMC, a key manufacturing partner of NVIDIA, to produce two self-developed AI chips based on a 5 nanometer (nm) process by 2026. These chips are currently in the design phase, with plans likely to evolve.
The 5nm process is only one generation behind TSMC’s most advanced 3nm process and is on par with the technology used in NVIDIA’s upcoming flagship AI chip, Blackwell architecture. By developing its own chips, ByteDance aims to decrease its dependence on costly NVIDIA components for developing and operating AI models. This year alone, ByteDance has ordered over 200,000 H20 chips from NVIDIA, valued at over $2 billion, and is awaiting the full delivery. Additionally, ByteDance plans to engage TSMC to manufacture hundreds of thousands of its proprietary training and inference chips, potentially saving them billions compared to purchasing directly from NVIDIA.