ByteDance's AI strategic focus may shift from mass consumer to enterprise services

According to LatePost, ByteDance is adjusting its AI resource allocation strategy, shifting its focus from consumer-facing products like "Doubao" to products that serve enterprise clients. This change is partly influenced by the high operational costs of AI and challenges to existing business models.

It is reported that "Doubao" has over 200 million daily active users, but due to the enormous computational costs associated with inference and multimodal capabilities, it incurs daily expenses of tens of millions, while its monetization channels, such as e-commerce, generate less than one million daily. In contrast, ByteDance's video generation model Seedance demonstrates strong profitability. Thanks to its efficient MoE architecture (200 billion parameters) and relatively low reliance on inference computing power, Seedance 2.0 currently has a gross margin of 70%, with the vast majority of its revenue coming from enterprise clients, and its current annual recurring revenue (ARR) has reached 2 billion dollars.

Insiders reveal that ByteDance's large model data review team has expanded to over 3,000 people this year, primarily to clean training data for programming models. Meanwhile, the MaaS business of Volcano Engine has also been placed in a more important position, with ByteDance's top management setting a goal to increase revenue tenfold and accelerate international expansion. Additionally, ByteDance executives recently visited Anthropic, which has achieved significant growth in both revenue and valuation through enterprise-level AI programming services, providing a reference for ByteDance's strategic adjustments.