StepFun Trillion-Parameter MoE Enterpris Diagram

Strengths

• Trillion-parameter MoE architecture reduces inference costs by up to 9x compared to Dense models
• Established professional moat in the financial sector with the 'AI Xiao Cai Shen' application
• Deep integration with terminal manufacturers such as Honor, OPPO, and ZTE

Weaknesses

• Lack of phenomenal consumer-facing (C-end) applications, resulting in relatively limited market visibility
• Closed-source model faces cost-based pressure from open-source alternatives
• Overseas commercial team is still in the early stages of development

Opportunities

• Explosive growth in demand for enterprise-grade Agents and private deployments in 2026
• Hong Kong IPO provides capital injection and brand endorsement
• Domestic chip ecosystem alliance reduces compute costs

Threats

• Competition from domestic models like Moonshot AI and DeepSeek offering lower pricing
• Performance gap between frontier models narrowing to 2.7%, reducing the advantage of technical generation gaps
• High industry-wide compute bills compressing gross margins