China AI 2.0: Application First, Monetization is King
The battlefield of Chinese AI is undergoing a fundamental restructuring.
Morgan Stanley’s latest flagship research report points out that Chinese AI has entered a new stage—the dominant narrative is no longer about catching up in capabilities, but about value capture; the core logic has shifted from training to inference, from technology to application, from potential to real profitability. This isn’t just a switch in topics, but a reshaping of the entire AI investment landscape.
This transformation has left clear marks at the corporate financial level. According to Morgan Stanley’s data, the next twelve months' earnings per share (NTM EPS) of Chinese AI adopters have cumulatively risen about 62% since the end of 2023, and EBIT margin is expected to expand from around 4% in 2021 to the 16%-17% range by 2027, significantly outperforming the MSCI China Index, which saw a roughly 10% increase in the same period.
The adoption side is also steadily expanding: Morgan Stanley’s latest 1H26 AlphaWise China CIO survey shows that 47% of surveyed CIOs plan to launch their first AI project in the coming year, up from 40% in the previous survey; among Greater China covered enterprises, the proportion of AI "enablers/adopters" has risen from 43% two years ago to 51%.
The report clearly states that the market is still underpricing the "rate of change" for China AI. Over the past year, the enabler layer—especially power, semiconductors, and infrastructure sectors—has consistently led gains; the report believes the application layer’s catch-up will be the next central theme.
Narrative Inflection: From Catch-Up to Monetization
The report opens with: "The Chinese AI story is no longer about catching up, but about rewriting the rules of the game."
Unlike the Western path focused on frontier model breakthroughs, Chinese AI is optimizing deployment speed, cost efficiency, and system-level integration capabilities. Morgan Stanley highlights three key trends that have emerged over the past 12 months:
First, AI is shifting from building infrastructure to profitable realization—adopters are already achieving quantifiable financial results, with upward profit expectations and margin expansion happening concurrently;
Second, bottlenecks have moved from computing power to electricity and deployment—the constraint is no longer whether AI can be built, but whether it can be powered, scaled, and delivered in real time, driving a new investment cycle at the energy system level;
Third, the new battlefield is extending toward embodied intelligence—from humanoid robots to autonomous driving, AI is rapidly moving into the physical world. With advantages in manufacturing, supply chain depth, and real-world scenario data, China has a unique competitive position.
Meanwhile, domestic semiconductor supply rates have steadily increased from 41% in 2025, and Morgan Stanley expects this to reach 86% by 2030, supporting a more resilient AI deployment system.
Accelerated Adoption, Profitable Realization
The penetration of AI among Chinese enterprises continues to exceed expectations, and its financial impact is beginning to materialize.
According to the fifth round of Morgan Stanley’s global AI mapping survey, the share of AI "enablers/adopters" in Greater China covered enterprises has gradually risen from 31% two years ago to 51% now; the proportion of enterprises listing AI impact as a "core argument" or "significant impact" has gone from 11% to 16%, while those with moderate AI impact have increased to over 37%, up from 32% two years ago.
The profitability data is even more intuitive. The NTM EPS for Chinese AI adopters has cumulatively increased about 62% since December 2023, with MSCI China staying roughly flat. EBIT margin is expected to reach 16%-17% by 2027, implying a cumulative expansion of 12 to 13 percentage points, confirming that AI’s impact on profit margins at this stage far outweighs its boost to revenue—about 91% of Chinese AI adopters’ primary benefit comes from cost efficiency, rather than revenue growth.
Large Model Layer: Pricing Power Returns, Commercialization Accelerates
An important turning point has occurred in the competition among large models—the "commoditization" trend is reversing.
Since price wars broke out in Q2 2024, the API prices of mainstream Chinese LLMs have generally dropped by 70%-90%; however, this trend peaked in Q2 2025. Alibaba, ByteDance, Baidu, Tencent, MiniMax, Z.ai, Moonshot, and DeepSeek have all raised prices for their new flagship models: According to Morgan Stanley, from Q2 2025 to Q1 2027, average API input prices have risen about 80%, average output prices about 36%, with Z.ai leading the trend—GLM-4.5 to GLM-5 saw over 200% increases.
Morgan Stanley emphasizes that this round of price increases is not due to rising costs but to continuous improvements in model performance—API gross margin expansion confirms the market is transitioning from pure commoditization to performance-based monetization.
Globally, according to OpenRouter data, the market share of token usage for leading Chinese LLMs has jumped from 5% in April 2025 to 32% in March 2026, while leading US models’ share dropped from 58% to 19%. The core efficiency advantage of Chinese AI is that, through architectural innovation (MoE mixture of experts), reinforcement learning, model distillation, and coordinated hardware/software optimization of inference infrastructure, Chinese models can deliver near-equivalent performance for only 15%-20% of the inference cost of US offerings.
The report lists MiniMax and Z.ai as core targets in the Chinese AI foundational model layer. Z.ai’s ARR is expected to grow from less than $100 million to $1 billion within a year; Morgan Stanley recently raised target prices for both by 18% and 77%, respectively. Alibaba is regarded as the best full-stack AI platform in the covered space, thanks to its chip (T-Head), cloud infrastructure (Alibaba Cloud), foundational models (Qwen series), and consumer app ecosystem; Tencent maintains strong competitiveness at the application layer, monetizing through the WeChat ecosystem.
Computing Power Independence: Semiconductor Chips Continue to Expand
The expansion of domestic chips continues.
Morgan Stanley expects Chinese AI chip market size to grow from about $19 billion in 2025 to $67 billion in 2030, with the supply rate increasing from 41% to 86%, and domestic vendor revenue by value to overtake imported chips in 2027.
From a competitive perspective, the market has shifted from "can we participate" to "how to compete for incremental share," driven by two structural trends: explosive expansion of AI inference workloads and tightening export controls.
Morgan Stanley channel research shows that domestic AI accelerators, compared to currently available Nvidia products in China, offer 30%-60% lower total cost of ownership (TCO), and inference cost per token is roughly comparable. Procurement decisions are shifting from chasing peak performance to “deployable cost efficiency.”
New Growth Poles: Power, Robots, and Autonomous Driving
The "powering AI" theme has led to a market value re-rating of more than $1.5 trillion.
Morgan Stanley notes that with the explosive growth of AI data center inference workloads, the bottleneck has shifted from power availability to power flexibility and grid access speed. Energy storage systems (ESS) are becoming the next core opportunity.
The report forecasts that global annual new deployment of data center ESS will reach about 321 GWh by 2030 (US 169 GWh, China 85 GWh, other regions 68 GWh), with a CAGR of roughly 30%, similar to the base of 325 GWh for global utility-scale ESS in 2025. Additionally, global gas turbine demand is expected to rise from about 60 GW/year to over 100 GW in the next decade, and high-voltage transformer prices are up 30%-50% from normal levels, with delivery lead times extended to two to three times the norm, giving China’s equipment exporters sustained flexibility.
Meanwhile, the scale tipping point for humanoid robots is approaching.
Morgan Stanley estimates that sales of humanoid robots in China will rise from 12,000 units in 2025 to 28,000 units in 2026, surpass 100,000 units by 2029, exceed 1 million units by 2034, and reach 54 million units by 2050, corresponding to about $1 trillion in potential market value, with a global market possibly reaching $7.5 trillion by then.
The competitive focus for 2026 has shifted from "movement capability" to "robot brain," with model capability as the key differentiator. China has unique advantages in broad real-scene data collection and multi-path parallel exploration. The 15th Five-Year Plan has, for the first time, listed robots as a strategic emerging industry, and sustained policy support will accelerate large-scale deployment.
The Chinese auto industry is also close to an AI-driven inflection point.
Regulators have issued the first batch of L3 autonomous driving permits to automakers including Changan, Huawei BAIC Arcfox, Lantu, Xpeng, Li Auto, and Xiaomi.
Morgan Stanley expects China’s L2+ intelligent driving penetration rate to rise from about 25% in 2025 to 32% in 2026, surpassing 50% by 2030. For Robotaxi, it is forecasted that by 2030, China will have 360,000 to 400,000 L4+ autonomous taxis, accounting for about 8% of the total taxi and ride-hailing fleet.
~~~~~~~~~~~~~~~~~~~~~~~~
The above exciting content is from Chasing Wind Trading Desk.
For more detailed interpretation, including real-time analysis and frontline research, please join [Chasing Wind Trading Desk Annual Membership]
Risk Disclosure and DisclaimerThe market carries risk; investment should be prudent. This article does not constitute personal investment advice, nor does it consider the particular investment objectives, financial situations, or needs of individual users. Users should consider whether any opinions, views, or conclusions herein are suitable for their specific circumstances. You are solely responsible for any decisions made based on this report.