营收过亿，这个多模态生成式AI黑马开启新一程

Core Insights - The article highlights the contrasting development states of AIGC companies in China and the US, with Chinese firms like Zhixiang Future showing significant commercial growth while US counterparts struggle with high costs and low user retention [1][2]. Company Overview - Zhixiang Future has achieved over 100 million RMB in revenue for 2025 and recently peaked in downloads for its C-end product vivago.ai, gaining nearly 10 million new users in January alone [1][2]. - The company has released major models such as HiDream-I1 and HiDream-E1, which have gained recognition in the AI evaluation rankings [1][4]. - The founder, Mei Tao, has a strong background in multimedia analysis and computer vision, having published over 300 papers and received multiple awards [2][3]. Business Model and Financing - Zhixiang Future is currently in the process of B-round financing, with both rounds of financing amounting to several hundred million RMB [2][11]. - The company has transitioned its business model from MaaS to SaaS and now to RaaS, focusing on delivering results and sharing revenue with clients [13][14]. - The company has a comprehensive content library, covering 70% of domestic film data and generating over 100 million AIGC derivative materials [10]. Market Potential - The global generative AI market is expected to grow at a compound annual growth rate of 63.8%, reaching $284.2 billion by 2028, with Zhixiang Future positioned to benefit from this growth [13]. - AI-generated videos are becoming a mainstream choice in marketing and content creation, with the global market for AI video generation projected to reach $6.2 billion in 2024 and $25.6 billion by 2032 [15]. Technological Innovation - Zhixiang Future has developed the Sparse DiT architecture, balancing generation quality and operational speed, and has made significant advancements in video generation capabilities [7][8]. - The company is also working on a new generation of multi-modal generative architecture that enhances reasoning capabilities and supports multi-task scaling [14].