JPMorgan Chase: China's AI demand accelerates expansion, model capability becomes the core competitive factor

The Chinese artificial intelligence foundational model industry is entering a phase of accelerated commercialization. JPMorgan believes that as the quality of models continues to improve and translates into faster demand growth, the capabilities of large models will determine pricing power, causing the gap between stronger and weaker companies to widen.

According to the Chasing Wind Trading Desk, on March 27, JPMorgan released a report that systematically addressed ten key questions from the market regarding demand growth, API pricing, competitive landscape, profitability, and global expansion risks.

The report states that 2026 is a crucial year for whether Chinese enterprise AI demand can replicate the growth curve of the U.S. market in 2025, with coding and agent applications becoming the most important demand catalysts.

Accelerated Demand: Non-linear Inflection Point Logic, Coding and Agents as Major Catalysts

JPMorgan believes AI demand should be understood as “inflection point driven” rather than linear growth—once model capabilities surpass a threshold sufficient to unlock real workflows on a large scale, demand will accelerate.

The U.S. market has set a precedent. According to data cited in the report, Anthropic’s annual recurring revenue (ARR) rapidly increased from $1 billion in December 2024 to $19 billion in March 2026, nearly a 19-fold increase in just 15 months.

JPMorgan believes that the conditions currently exist in China to follow a similar model: domestic model capabilities are approaching or even exceeding the level of leading U.S. models from a year ago, while domestic pricing is also more aligned with local economic efficiencies, both of which have improved the commercial returns on implementation.

The demand logic on the agent side is also strengthening. The report points out that OpenClaw has become an important catalyst, pushing use cases from single-turn interactions to multi-step executions, significantly increasing the token intensity of each task. Tencent, Alibaba, and ByteDance have integrated tools linked to OpenClaw into their respective ecosystems.

API Pricing: Differentiation is the Main Theme, Capability Determines Pricing Power

JPMorgan assesses that API pricing is unlikely to move unidirectionally but is more likely to trend toward differentiation.

On one hand, capability forms pricing power. If a model can uniquely unlock high-value tasks, such as agent coding, long-term workflows, or enterprise-level reliability, customers will pay a premium because the returns are quantifiable and independent of the token unit price.

On the other hand, as the efficiency of hardware, systems, and algorithms continues to improve, the unit cost of inference will continuously decline, putting pricing pressure on models that are “good enough but have stopped improving.”

The report concludes that models at the forefront of capabilities are expected to achieve simultaneous increases in both volume and price; whereas models that fail to continue improving are more likely to face price declines while experiencing growth in usage, with uncertain profit margin prospects.

Competitive Focus: Shifting from Price Wars to Model Capabilities

The report emphasizes that this is a key difference from discussions last year—especially in China, where the previous market focus was on comprehensive price competition.

In agent use cases, what customers are actually purchasing is not cheap tokens but the successful completion of tasks. The report cites calculations showing that in multi-step workflows, even a slight improvement in single-step reliability can lead to a significant enhancement in the final task completion rate (a single-step success rate improvement from 90% to 95% can increase the 20-step completion rate from 12% to 36%).

This means that models with a higher average token price but stronger reliability may actually have a lower actual cost per successful task.

JPMorgan believes that companies with strong cutting-edge models can typically extend more easily into lower-end markets, while companies focused on low prices find it difficult to enter higher-end markets. Therefore, competition is increasingly centered on absolute model quality and engineering efficiency rather than purely on price.

Industry Landscape: Survival of the Fittest, the Stronger Grow Stronger

JPMorgan maintains its judgment of a “survival of the fittest” scenario in the large language foundational model industry, with the core logic being: small technological gaps and endless product cycles, with business models gravitating toward API sales; companies that lose momentum can be quickly squeezed out.

The report points out that in China, the gaps between various large language model companies are often much smaller than investors believe, making the market highly unstable. Companies must continuously spend and iterate to avoid falling behind—standing still is not neutral but signifies a loss of market position.

Regarding the trend of internet giants venturing into the B-end AI field, the report believes this makes competition between independent model providers and large platforms more direct.

Alibaba has clearly identified cloud and AI as strategic priorities, and Tencent’s recently launched agent products have been segmented into personal, developer, and enterprise scenarios. JPMorgan assesses that as platforms more aggressively pursue B-end monetization, the advantage of “cloud neutrality” is weakening, with both parties’ competitive focus shifting toward the capabilities of the models themselves.

Profitability: Improvement in Gross Margin Expected, Operational Leverage Still to be Verified

JPMorgan believes that for large language model providers maintaining a top global position, gross margins should rise with improvements in model efficiency and inference chip efficiency, and higher-value workloads will also support a more favorable revenue structure. However, the more critical question of profitability is whether gross profit growth can exceed the growth rate of R&D expenditures.

The report references Anthropic—despite the company’s revenue level reaching $14 billion in February 2026, it simultaneously announced a round of $30 billion in financing and emphasized continued cutting-edge development, confirming that high revenue does not mean training intensity has normalized.

JPMorgan maintains “buy” ratings for Zhiyuan and MiniMax, with target prices of HKD 800 and HKD 1,100 respectively. It predicts that both Zhiyuan and MiniMax will turn profitable starting in 2029. The report also emphasizes that the continuous growth in usage and trend of improving unit economic efficiency are more important than the exact year in which profits are achieved.

The above exciting content comes from the Chasing Wind Trading Desk.

For more detailed interpretations, including real-time insights and frontline research, please join the 【**Chasing Wind Trading Desk ▪ Annual Membership**】.

![](https://img-cdn.gateio.im/social/moments-fbd1926258-e4f5fedf69-8b7abd-ceda62)

Risk Warning and Disclaimer

          

            Markets are risky, and investments require caution. This article does not constitute personal investment advice, nor does it take into account the specific investment goals, financial situations, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investing based on this carries your own responsibility.