ByteDance Quietly Develops Advanced AI Models to Rival Industry Leaders

ByteDance has started training an artificial intelligence model designed to reach 10 trillion parameters, according to people familiar with the initiative. The effort signals a sharp increase in compute ambition, putting the planned system at a scale that is described as roughly three times larger than the biggest Chinese models currently available.

Officials familiar with the project said the target size would also exceed widely cited estimates for leading US-based AI systems. The company is pursuing what it describes as an independent development path, with an explicit decision to avoid model distillation techniques in order to prioritize long-term performance and competitiveness.

Supporters of the strategy argue that owning more of the stack—from infrastructure to training methodology—can improve control over performance, deployment choices, and product integration. In this framing, the initiative is not only a research milestone but also a platform investment tied to commercial AI services.

Hardware constraints and regulatory scrutiny remain uncertainties

People familiar with the effort cautioned that the project could face headwinds linked to hardware procurement. The availability of advanced computing components can shape both training timelines and the achievable scale of experiments, making supply constraints a practical risk to execution.

Officials also pointed to the possibility of international regulatory scrutiny as a complicating factor for large-scale AI development. How oversight evolves—and what constraints could apply—was described as an open question for the initiative.

If successfully deployed, the model would represent a major step in the capabilities of Chinese firms building at the frontier of AI scale. Officials said such an outcome could influence the current balance between domestic and international AI labs, though the pace and feasibility of reaching the stated training target remain to be proven in practice.