ByteDance’s All-In Strategy on AI Development

ByteDance, China’s internet giant and parent company of TikTok, is making an aggressive push into artificial intelligence with a comprehensive, saturated approach across the entire AI value chain, according to a report by Latepost

The company has established two independent AI organizations – Flow and Seed – which operate separately from ByteDance’s existing business units and report directly to the founders. This structure allows the AI division to bypass traditional expansion rhythms, compensation structures, and evaluation mechanisms.

According to the February all-hands meeting, ByteDance’s three major AI goals for 2025 are: exploring the upper limits of intelligence, investigating new UI interaction forms, and strengthening network effects. The AI organization, comprising Flow, Seed, and Stone divisions, has nearly 2,500 employees.

Seed, responsible for model development, is now co-led by Zhu Wenjia and Wu Yonghuei, who joined in February 2024. Wu, a former Google Fellow with over 40,000 paper citations, leads the core model development team focusing on large language models, vision models, speech models, and multimodal & world models.

Flow, led by Zhu Jun, oversees AI product development, including Doubao (豆包), which has become China’s most active AI application. The company’s AI exploration covers various directions: AI character interaction, image generation, AI education, and developer tools.

ByteDance has allocated substantial resources to AI development. By 2023, the company had secured over 100,000 GPUs, with orders exceeding $1 billion to Nvidia in the first half of 2023 alone. The AI computing power procurement budget for 2025 is reported to be at least 90 billion yuan.

However, ByteDance faces new challenges. The emergence of DeepSeek, a small technical team of less than 200 people producing impressive models, and Tencent’s unexpected surge in AI users through DeepSeek integration have created competitive pressure. Despite these challenges, ByteDance maintains its characteristic approach: saturated investment from talent to infrastructure, exploring all business directions without overlooking any possibilities.

The company’s founder, Zhang Yiming, has returned to startup mode, frequently meeting with AI researchers and personally studying technical papers with assistance from a dedicated research team.

Discussion about this post