For stock trading, check the Jin Qilin analyst research reports: authoritative, professional, timely, and comprehensive, helping you uncover potential thematic opportunities! CITIC Securities Research. By Lian Yixi, Xu Yingbo, Yang Haotian, Liao Yuan, Yang Zeyuan, Chen Junyun, Jia Kaifang, Ye Minting.
Although the current market attitude toward large models is trending cautious, we remain firmly bullish on the long-term commercial value and valuation growth space of both Chinese and US models. Currently, the RSI closed loop and model iteration are accelerating, and the combined ARR growth of OpenAI and Anthropic remains solid. Model payment and usage are highly uneven, exhibiting early commercialization characteristics similar to the mobile internet era. Even considering only the diffusion of demand to more enterprises, scenarios, and regions, the commercialization space remains broad. Domestically, demand for data annotation is exploding, the post-training data bottleneck is easing, and domestic model post-training will ultimately break through in the cracks. Looking ahead, we expect 26Q4 to be the key juncture for confirming large-scale enterprise-level deployment of overseas models. In August-September, due to factors such as competition and anti-distillation brought about by the release of GPT-6-Astra, Anthropic's ARR month-over-month growth rate declined somewhat. However, with the release of Opus5.5 at the end of September and expectations for next-generation model releases such as Fable 5.5 before its IPO, we believe that after October, Anthropic's ARR is expected to regain momentum, thereby repairing the market's currently overly pessimistic expectations for the AI sector. At present, domestic model ARR is only equivalent to the level of overseas models 18 months ago, while valuations are only one-tenth of overseas models. Once domestic model capabilities break through, the lower revenue base, faster growth rate, and more attractive valuations are expected to drive a systematic revaluation of domestic models.
Report Origin: Anthropic ARR Marginal Slowdown Triggers Market Volatility
However, the combined ARR growth of OpenAI and Anthropic remains solid, and we remain firmly bullish on the commercial value and valuation growth space of Chinese and US models. Since July, market concerns about Anthropic's ARR growth slowdown have intensified, mainly due to two reasons: first, previous expectations were overly optimistic, with the market linearly extrapolating the $17 billion single-month ARR increment in April to May-June; second, since July, OpenAI has continued to capture Anthropic's share. Confusion in third-party calculations has also exacerbated ARR expectation volatility. But from the combined scale of the two companies, ARR growth remains solid. The average monthly ARR increment in August-September was $15.9 billion, close to the combined increment of approximately $17 billion in April. From a full-year perspective, if extrapolated at a monthly increment of $15 billion, year-end ARR is expected to reach $180 billion, 5.9 times that of end-2025, faster than the 4.7 times and 3.8 times of the previous two years. Looking ahead, after Anthropic's November IPO, it may disclose more fundamental information and provide clearer guidance, at which point the model layer may usher in a new round of capital allocation and rebalancing.
Frontier Outlook: RSI Closed Loop Accelerates, New Scenarios Such as Cowork, AI4S, AIDD, and AI for Chip Design Land First
RSI progress is smooth. The proportion of R&D work led by Claude internally at Anthropic increased from 1% in March to 26% in August, and the proportion of work involving deep AI collaboration and above has exceeded 90%; OpenAI's agent research capability has reached a level of 3.1 Agents per human workday. However, achieving a fully closed RSI loop still requires time. Before RSI achieves a fully autonomous closed loop, scenarios such as Cowork, AI4S, and AI for Chip Design are expected to contribute incremental gains first. In AI4S, OpenAI and Anthropic released GPT-Rosalind and Claude Science in April and June respectively. In AI for Chip Design, OpenAI released its first self-developed inference chip Jalapeno mid-year, participating in chip design through internal models and reinforcement learning algorithms; Mythos 5.1 rewrote the GPU kernels of seven learning models, increasing inference speed by up to 2.5 times.
Commercial Deployment: In the Short Term, Enterprise-Level Deployment Enters the Acceptance Period at Year-End; on the C-End, Meta Muse Ignites Expectations for an Application Explosion
In the long term, we are still in the very early stages of AI commercialization. Currently, model capabilities maintain linear growth, but payment and usage remain highly uneven, exhibiting commercialization characteristics similar to the early mobile internet era. AI spending by the top 1% of enterprise employees is 10 times that of the top 10% and 576 times that of the top 50%. In terms of per capita usage, California is 6.5 times that of West Virginia; the US is 21 times that of India. Models exhibit commercialization characteristics highly similar to the early mobile internet era. The experience of the mobile internet shows that technology adoption requires a process from concentration to democratization. In 2010, the top 1% of users generated 52% of monthly data traffic in North America; by 2019, the top 1% of users accounted for only 5% of monthly data traffic. In the future, even considering only the diffusion of demand to more enterprises, scenarios, and regions, the commercialization space remains broad. In the short term, enterprise-level scenarios such as finance are expected to lead the acceleration of commercialization. Referring to SaaS historical experience, from small-scale validation to actual commercial deployment, it takes at least 6 months for FDE to help enterprises achieve large-scale deployment. If calculated on this basis, 26Q4 will be the key juncture for confirming large-scale enterprise-level deployment of overseas models. Once ARR re-accelerates, market sentiment is expected to reverse.
Domestic Models: Data Annotation Demand Explodes; If the Capability Gap Persists, a Systematic Revaluation Is Expected
Demand for data annotation is exploding, the post-training data bottleneck is easing, and domestic model post-training will ultimately break through in the cracks. At present, the combined ARR of domestic models is approximately $8 billion, only equivalent to the level of overseas models 18 months ago, while valuations are only one-tenth of overseas models. Once domestic model capabilities break through, the lower revenue base, faster growth rate, and more attractive valuations are expected to drive a systematic revaluation of domestic models.
Risk Factors
Risks include: model technology innovation and capability breakthroughs falling short of expectations; increased capital expenditure leading to depreciation eroding profits; computing power supply falling short of expectations; AI application commercialization falling short of expectations; intensifying industry competition; and computing power chain hardware export controls.
Investment Strategy
We recommend focusing on Anthropic's IPO progress, as well as domestic large model unicorns and internet companies. The RSI closed loop is accelerating, and model iteration continues to speed up. Anthropic's IPO window is approaching, coupled with the gradual landing of enterprise-level commercialization, bringing opportunities for model value revaluation. In August-September, due to factors such as competition and anti-distillation brought about by the release of GPT-6-Astra, Anthropic's ARR month-over-month growth rate declined somewhat. However, with the release of Opus5.5 at the end of September and expectations for next-generation model releases such as Fable 5.5 before its IPO, we believe that after October, Anthropic's ARR is expected to regain momentum, thereby repairing the market's currently overly pessimistic expectations for the AI sector. Overseas, focus on ARR growth after year-end FDE project acceptance; domestically, wait for model capability breakthroughs. Currently, domestic model capabilities lag overseas by about 6 months, while commercialization levels lag by about 18 months. If the capability gap remains stable, domestic models are expected to accelerate their pursuit of commercialization. With a lower revenue base, faster growth rate, and more attractive valuations, they are expected to usher in a systematic revaluation, and the investment value of domestic models and internet companies is expected to gradually emerge.
Comments