ByteDance Pre-Training 10-Trillion-Parameter AI Model as Global Race for Frontier AI Intensifies

ByteDance is reportedly pre-training a 10-trillion-parameter AI model, according to reports from August 2026, in a move that signals China's push to compete with Western AI labs at the frontier of model scale.

The scale of the reported model would place it among the largest AI systems ever trained. For comparison, most publicly known large language models operate in the range of hundreds of billions of parameters. A 10-trillion-parameter model would require enormous compute resources and represents a significant investment in frontier AI development.

ByteDance's move comes as Chinese AI companies accelerate their efforts to close the gap with US labs. Alibaba also released its Qwen3.8-Max model in August 2026, positioning it as a competitor to leading Western models.

The global AI race is intensifying on multiple fronts. In the United States, Meta released Muse Glimmer, a 30-billion-parameter open-weight model optimized for local, offline use on consumer hardware. Meta CEO Mark Zuckerberg has framed open-weight models as a counterweight to proprietary systems from OpenAI and Anthropic.

On the hardware side, Tesla and SpaceX committed $16.8 billion to build a semiconductor manufacturing complex in Texas called Terafab. Intel raised $15 billion through a stock offering to expand its AI chip capabilities. AMD acquired the startup Taalas to embed AI workloads directly into silicon.

Nations are also treating AI as a matter of national security. South Korea committed billions to its semiconductor supply chain, and companies like Naver are partnering with Nvidia to build sovereign AI data centers.

Analysts said the scale of investment across both the US and China suggests the AI infrastructure buildout will continue at a rapid pace through the remainder of 2026 and into 2027.