China AI Narrative Shift, Profitability Becomes the New Mainline

Morgan Stanley's latest China AI research report, released in May, defines the current phase as "China AI 2.0" — the market narrative is undergoing a clear shift: from model training to inference deployment, from infrastructure investment to application realization and profit realization. Over the past year, the market has systematically underestimated the "profit change rate" brought by AI commercialization, which is the most important pricing error at present.

The most illustrative data is: enterprises that have adopted AI technology have an EPS expectation increase of over 60% in the next 12 months, significantly outperforming the market. The benefits of AI have begun to penetrate from the capital expenditure statements of tech giants to the profit and loss statements of ordinary enterprises. Alibaba is listed as the best-positioned full-stack AI platform within the scope of China, Tencent as the best AI application and ecological platform, and the compound annual growth rate of China's AI cloud market from 2024 to 2029 is expected to reach 72%.

At the level of large models, the gap between China's top models and the United States has been narrowed to 3 to 6 months, and the inference cost is only 15% to 20% of that in the United States. Since the second half of 2025, various manufacturers have begun to increase API prices, declaring the end of pure price wars. MiniMax and ZhiPu AI, which completed their listing on the Hong Kong Stock Exchange in January this year, are key basic model targets in the coverage list.

At the infrastructure level, the bottleneck constraining the expansion of AI has shifted from computing power to electricity. The demand for electricity by data centers has been upgraded from "available" to "dispatchable at any time", and energy storage systems have become key infrastructure due to their cost advantages and the ability to delay conventional power grid capital expenditures. It is expected that by 2030, the global data center energy storage deployment will reach 321GWh, and the core beneficiaries will be Contemporary Amperex Technology Co. Limited (CATL), Siyuan Group, and Yingli Green Energy Holding Company Limited (Yingli Green Energy).

At the chip level, the process difference is no longer the core competitive dimension — through advanced packaging and software-hardware collaborative optimization, the overall cost of ownership of domestic AI accelerators is already 30% to 60% lower than American products under inference scenarios. The self-sufficiency rate is expected to rise from 41% in 2025 to 86% in 2030, and Cambricon, Gowin Semiconductor, SMIC, and Northern Huabao are key coverage targets.

Humanoid robots and autonomous driving are classified as "physical AI". It is expected that by 2026, the delivery volume of Chinese humanoid robots will double to 28,000 units and the penetration rate of L2+ intelligent driving will reach 32%. On a macro level, AI is expected to raise China's potential GDP by about 3.5 percentage points by 2035, effectively offsetting the pressure of population aging. However, short-term risks should not be ignored: within the next 2 to 3 years, junior white-collar and labor-intensive service industries face significant replacement risks, and if policy responses are inappropriate, it may exacerbate deflationary pressures.

Among the 65 core targets, the best AI adopters include Beisen Holding, Meitu, Roborock, Midea Group, and Ecovacs Robotics, all of which have achieved quantifiable cost reduction or product upgrades through AI. The investment main line of the next stage will belong to those enterprises that can effectively convert the popularization of AI into incremental revenue and profit margin expansion.