Where has China's AI large model reached? Morgan Stanley report: The gap between China and the U.S. is narrowing, and commercialization is accelerating

According to the Zhitong Financial Bureau APP, an industrial transformation triggered by large models is accelerating in the Chinese market. Morgan Stanley's latest report outlines the true picture of China's AI industry: the technology gap is rapidly narrowing, price wars are forcing cost optimization, major companies are accelerating commercialization, unicorns are crossing the cash-burning phase, and the entire industry chain is undergoing a value reassessment.

China-U.S. Technology Gap: Rapidly Approaching, Already Equipped for Large-Scale Commercialization

Regarding the market's primary concern about the capability gap between China and the U.S. in large models, the report's core conclusion is that China's large models are quickly approaching the top levels in the U.S.

Whether in the field of cutting-edge closed-source models or open-source models, China's intelligence index is closely following the U.S., with no significant gap emerging. This means that the foundational capabilities of domestic large models are now sufficient to support large-scale commercial applications. The technological foundation is gradually solidifying, clearing fundamental obstacles for subsequent industrial explosions.

Pricing and Competition: Multi-Modal API Prices Plummet, Underlying Computing Power Optimization Behind the Intense Competition

At the critical juncture of commercialization, domestic large models are facing a brutal price war.

The report points out that the API prices for multi-modal models are experiencing a cliff-like drop. The API call prices for text-to-video and image-to-video generation have plummeted from Q1 2025 to Q3 2026. The incremental information behind this reflects the extremely fierce industry competition, with major companies drastically lowering prices to seize the developer ecosystem; but it also indicates that substantial progress has been made in optimizing underlying computing power and inference costs, and the price reductions are not merely "losing money to attract business."

Major Companies Racing: Alibaba "Making Real Money," Tencent Betting on Ecological Applications

In the AI commercialization race among internet giants, Alibaba (09988) and Tencent Holdings (00700) are showcasing distinctly different strategies.

Alibaba is currently the biggest beneficiary of a full-stack AI layout (IaaS + PaaS + MaaS). Core data indicates that Tongyi Qianwen's (Qwen) ARR (Annual Recurring Revenue) has doubled within three months, and it is expected to exceed 30 billion RMB by the end of 2026. In other words, Alibaba not only excels in model development but is also making real money. Its AI capital expenditure (Capex) can yield mid-double-digit returns on invested capital (ROIC), with a payback period of less than three years.

Tencent's strategy, on the other hand, relies on its vast social and office ecosystem (WeChat/QQ/Tencent Meeting, etc.) to create "AI-native applications." The WorkBuddy and CodeBuddy driven by the Hunyuan large model rank first among AI office assistants on the domestic PC side. This indicates that Tencent has significantly increased its AI investment in 2026, not rushing to compete on the underlying model scores, but focusing on the commercialization of high-frequency scenarios and user stickiness.

Unicorn Breakthrough: The Six Little Dragons Have Passed Their Darkest Hour and Entered a Positive Cycle

As representatives of China's AI entrepreneurial force, the survival status of the "Six Little Dragons" has always attracted market attention. The report shows that leading startups have survived the most challenging pure cash-burning phase.

Zhiyu (02513), focusing on foundational large models (GLM series), has seen its ARR continuously climb, with gross margins expected to reach 30% by 2026, gradually moving towards 50% MiniMax is betting on all modalities, relying on the dual drive of application end (MaaApp) and model as a service (MaaS), with gross margins also steadily improving. Overall, the revenue scale and gross profit level of leading startups have begun to enter a positive cycle.

Industry Chain Overview: Focus on the "Large Models" in the Spotlight, but Don't Forget the "Water Sellers" Behind the Scenes

For the capital market, the value distribution of China's AI industry chain is panoramic.

From energy cooling (such as Invec), to computing power chips (such as Semiconductor Manufacturing International Corporation/Huawei), to infrastructure (such as Foxconn Industrial Internet/Zhongji Xuchuang), basic models, and up to top-level applications (such as Kingsoft Office/Meituan), the capital market's focus is spreading comprehensively.

In addition to the large model manufacturers in the spotlight, the "water sellers" (computing power networks + energy temperature control) behind the scenes are also a highly certain core beneficiary segment.

Conclusion

China's AI industry has fully transitioned from a simple technological competition to a deep water zone of commercialization and ecosystem construction. The rapid iteration of underlying models and the drastic reduction in inference costs are reshaping the competitive landscape of the industry. Leading internet giants and AI unicorns have found monetization paths that align with their advantages and are gradually moving towards a positive profit cycle. Meanwhile, investment opportunities in the AI industry chain are showing characteristics of full-chain diffusion, with the deterministic advantages of infrastructure and underlying computing power segments becoming increasingly prominent