According to a research report from Open Source Securities, Gartner data indicates the global AI industry will reach $1.8 trillion by 2025, with AI models accounting for 1% of that total, showing a two-year compound annual growth rate (CAGR) of 74%. Furthermore, IDC data projects the number of active AI Agents will surge from approximately 28.6 million in 2025 to 2.216 billion by 2030, indicating substantial growth potential. The top three players hold over 50% market share, with domestic large language models emerging as new disruptors. In mid-February, MiniMax's M2.5 model ranked first on the OpenRouter leaderboard for weekly token consumption. Mainstream large models still have significant room for improvement in practical scenario solutions and long-context learning. The next phase of AI may prioritize evaluation over training, signaling an industry paradigm shift from "comprehensive" to "precise and accurate." The competition between OpenAI and Anthropic highlights differing growth dynamics: Anthropic's annual recurring revenue (ARR) grew nearly tenfold annually from 2023 to 2025, while OpenAI's grew around threefold. By February 2026, OpenAI achieved an ARR of $25 billion, with Anthropic reaching $20 billion by early March 2024, showing Anthropic is accelerating its catch-up. Regarding profitability, both companies maintain a medium to long-term gross margin outlook above 70%, though this has been revised downward due to higher-than-expected inference computing costs. In 2024, OpenAI's computing costs—comprising training (R&D expenses) and operations (inference/operational costs)—reached 125% of its revenue, with its computing investment significantly exceeding Anthropic's. By 2028, Anthropic's revenue scale is expected to approach OpenAI's, while its computing investment may only need to be about one-third of OpenAI's. In consumer markets, major players are accelerating penetration during seasonal peaks like the Chinese New Year, while in enterprise markets, domestic models are joining the top tier. ChatGPT maintains stable monthly visits exceeding 5.5 billion, significantly leading competitors, while Gemini steadily catches up leveraging multimodal capabilities. China's large model market is growing rapidly, with Qwen emerging as the leader in the enterprise sector, capturing a 32% market share in the second half of 2025. Seasonal promotions and deep Agent integration are driving monthly active user growth, with Alibaba, ByteDance, and Tencent employing distinct strategies based on their respective resources. The enterprise API call market is highly concentrated, with Anthropic holding nearly half the share. Domestic large model firms are gradually entering the top tier, and the current competitive landscape may shift. Leveraging advantages in "open-source and cost-efficiency," domestic models are poised to maintain long-term competitiveness. On valuation, Anthropic's valuation updates have typically been 2-3 times its previous level, but its rapid ARR growth provides a foundation for valuation digestion. In its early stages (May 2023), when the business was essentially in a 0-to-1 phase, its valuation reached 46 times that year's ARR. Since 2025, Anthropic's peak valuation has been over 20 times, typically dropping below 10 times the following year. Applying this to domestic models, such as Knowledge Atlas, matching Anthropic's valuation anchor would require Knowledge Atlas to achieve a 176% revenue CAGR from 2025 to 2027. Given that Agent-driven model demand is not simply linear, the ARR growth of domestic large models is expected to adequately absorb current valuation levels. Investment recommendations focus on opportunities in models, AI infrastructure, and computing power amid the exponential leap driven by Agents. Domestic large model capabilities have progressively reached global top-tier levels. Beneficiaries include
Comments