华尔街见闻·2026-02-10 11:52Core Insights - The core viewpoint of the article emphasizes that the ability to convert AI models into cash flow is becoming the true scarcity in the industry, as the Chinese AI market transitions from a "model war" phase to one where commercial viability and global positioning are key determinants of success [1][2]. Industry Overview - The report indicates that the number of capable and well-funded model developers in China has decreased from over 200 to less than 10, highlighting a significant consolidation in the AI market [1]. - The largest profit pool in the domestic AI industry is expected to shift towards platform giants that control distribution, while independent firms must find survival niches through "structural neutrality" [1][4]. Profit Distribution - The report asserts that the long-term profit pool of generative AI will likely be concentrated among large internet platforms like Tencent and Alibaba, which have established distribution and monetization channels [5][6]. - Platforms have high-frequency user touchpoints and mature application scenarios, making it easier to internalize AI capabilities as features rather than standalone products [6][7]. Independent Model Companies - Independent model companies like Zhipu and MiniMax are seen as having opportunities not through direct competition with platforms but by providing "structural neutrality" [11][12]. - These independent providers can monetize their models through APIs and enterprise licensing without creating competitive dependencies with their clients [14]. Financial Insights - Zhipu's revenue model is heavily focused on localized deployment, which accounted for 85% of its total revenue in the first half of FY2025, with a gross margin of 59.1% [16]. - MiniMax's revenue structure is notably global, with over 73% of its total revenue coming from markets outside China, providing it with significant economic flexibility [24]. Growth Projections - Zhipu is expected to achieve a revenue CAGR of 127% from 2026 to 2030, with profitability anticipated by 2029 and a normalized net profit margin of 20% by 2030 [19][20]. - MiniMax is projected to have a revenue CAGR of 138% during the same period, with profitability also expected by 2029 and a normalized net profit margin of 24% by 2030 [29][30]. Cost Structure Changes - The report highlights a significant shift in cost structure from "training-driven" to "inference-driven," with inference costs expected to dominate future expenditures [32][39]. - For Zhipu, the proportion of training costs is projected to drop from 93% in 2025 to 32% by 2030, while inference costs will rise from 7% to 68% [34][37]. Conclusion - The competitive landscape is shifting, with the focus moving from who can train larger models to who can achieve cheaper inference and higher utilization rates [40]. - The value of Zhipu and MiniMax lies not in challenging the platforms but in occupying indispensable positions outside of them [41].