ByteDance begins training 10-trillion-parameter AI model

According to a report circulating via Google News aggregation, ByteDance has begun training an AI model with 10 trillion parameters. The news spread through aggregator channels on August 8 and drew attention across the industry.

A 10-trillion-parameter model represents a step change in scale, demanding GPU clusters, high-bandwidth interconnect, and power supply on a level few organizations can sustain over the long run.

Public information so far is limited to the training plan itself; the report does not provide further details on the model's architecture, training timeline, or compute scale.

If confirmed, ByteDance would join a very small group of companies worldwide attempting models at the 10-trillion-parameter scale. The plan signals that the race to ever-larger foundation models is heating up again.

Why it matters: model scale is directly tied to capability ceilings and cost structures. Training at 10 trillion parameters tests not only algorithmic and engineering skill, but also a company's long-term commitment to infrastructure investment.

What to watch next: whether ByteDance officially confirms the plan, when training is expected to complete, and how the model will eventually reach user-facing products. The competitive landscape for frontier-scale models could be redrawn.