ByteDance CEO Liang Rubo prioritizes in-house AI models despite lagging LLMs

ByteDance CEO Liang Rubo used an internal companywide meeting on Aug. 6 to acknowledge a hard truth: ByteDance's large-language models have fallen further behind leading overseas models. According to people familiar with the meeting, Liang said the company's Doubao consumer AI app remains competitive and its Seedance video-generation model is at the technological frontier. But he made clear that ByteDance will not pivot away from its in-house AI strategy. Instead, it will continue developing its own foundation models, accepting short-term losses in the global AI model race to build a stronger technological foundation for the future.

This stance comes as ByteDance faces significant pressure on its AI investments. The company recently saw profits plunge 70% on aggressive AI spending, and it has reshuffled its workplace tool Feishu as part of a broader AI push. By insisting on proprietary models, ByteDance is betting that owning the full AI stack will pay off over time, even if it means lagging on current benchmarks. Doubao and Seedance give ByteDance credible consumer and video-generation products to build around, but the core LLM gap remains the company's biggest vulnerability. Liang's message signals a deliberate trade-off: immediate competitive positioning sacrificed for long-term technological independence and integration.