ByteDance creates top-level AI data unit amid 10T model push

ByteDance has established a new first-tier department called "AI Data & Security," elevating the company's approach to data governance as it embarks on training what would be one of the world's largest AI models.

The new unit, reported by Chinese tech outlet 36Kr, operates at the same organizational level as ByteDance's Seed AI research division, Flow, and Douyin, according to PANews. It is led by Adam Wang, who previously oversaw TikTok's platform responsibility and live streaming operations.kucoin

Consolidating the Data Pipeline

The department's core mission is to provide cross-modal data services for all of ByteDance's large models. It integrates the Global Data team founded by Fu Yue, a founding member of TikTok, in 2023, along with the company's data middleware DMC and the AI Data Platform under Flow. The unit covers the entire data lifecycle, from standardization and sourcing to synthesis, cleaning, quality evaluation, and security compliance.kucoin

The Anti-Distillation Context

The restructuring comes days after ByteDance founder Zhang Yiming told the company's Seed AI research team to avoid improving models by distilling rival systems. "AI development requires long-termism and delayed gratification, rather than using others' output to achieve short-term leaderboard rankings," Zhang said at a recent internal meeting, according to a Reuters report citing China's The Paper.mix929

Zhang told staff ByteDance should be willing to "sacrifice some short-term gains for long-term goals," adding that distillation interferes with genuine long-term technological breakthroughs.mix929

The directive carries added weight given the broader geopolitical environment. The Financial Times reported on August 7 that ByteDance is pre-training a model with as many as 10 trillion parameters using a mixture-of-experts architecture, requiring approximately 30,000 GPUs and three to six months of continuous pre-training. The new data unit appears designed to ensure the training pipeline for that effort — and future ones — runs on properly governed, internally sourced data rather than outputs scraped or distilled from Western systems.wionews

Structural Signal

By placing AI data governance at the same tier as its core AI research and product divisions, ByteDance is making an organizational statement that data provenance and compliance are not afterthoughts but prerequisites for its frontier ambitions. The move also positions the company to defend itself against future accusations of intellectual property misuse at a time when U.S. officials have warned that "sanctions and Entity List designations will be on the table" for Chinese firms suspected of distilling American AI systems.wionews