字节芯片，隐秘布局｜36氪独家

Core Viewpoint - The article discusses the emergence of AI-driven industrial revolution, focusing on ByteDance's advancements in self-developed chips and their implications for cost reduction and strategic independence in the AI sector [2]. Group 1: Chip Development and Strategy - ByteDance's chip team has expanded to over 1,000 members, with more than 500 focused on AI chips and around 200 on CPU development [6]. - The company is prioritizing its chip business due to the increasing demand for computing power, leading to a recent team restructuring [7]. - ByteDance plans to invest 160 billion yuan (approximately 22.4 billion USD) in AI by 2026, with 85 billion yuan (around 12 billion USD) allocated for AI processor procurement and development [8]. Group 2: Product Segmentation - ByteDance's chip initiatives began in 2020 and are categorized into four main areas: AI chips, CPUs, VPUs, and DPUs, all closely aligned with its core business [10]. - The AI chips are intended for inference tasks related to the Doubao model, while CPUs are designed for general computing in data centers [10]. Group 3: Production and Partnerships - ByteDance aims to produce at least 100,000 self-developed AI chips by the end of this year, with plans to increase production to 350,000 chips, potentially partnering with Samsung for manufacturing [11]. - The company has been iterating its AI chip designs for three to four generations and has engaged with various foundries, including Samsung [12]. Group 4: Current Resources and Future Plans - As of the end of 2025, ByteDance has secured nearly 120,000 NVIDIA GPUs and close to one million NVIDIA chips, indicating a dual strategy of self-development and external procurement [13]. - Additionally, ByteDance has accumulated a significant number of domestic AI chips from various manufacturers, further diversifying its hardware resources [14].