J.P. Morgan: China's AI demand is accelerating, and model capabilities have become a key competitive factor
```
China's foundational AI model industry is entering an accelerated phase of commercialization. JPMorgan believes that as model quality continues to improve and translates into faster demand growth, major model capabilities will determine pricing power, and the gap between stronger and weaker companies will widen significantly.
According to Chase Trading Desk, on March 27 JPMorgan released a report systematically answering ten key market questions about demand growth, API pricing, competitive landscape, profitability, and risks in global expansion.
The report argues that 2026 will be a pivotal year to see whether China's enterprise AI demand can replicate the growth curve of the U.S. market in 2025. Coding and agent applications are emerging as the most important demand catalysts.
Demand Acceleration: Nonlinear Inflection Point Logic, Coding and Agents as Main Catalysts
JPMorgan believes AI demand should be understood as "inflection point-driven" rather than linear growth—once model capabilities cross a threshold, unlocking real workflow at scale, demand then accelerates rapidly.
The U.S. market has already provided a precedent. According to data cited in the report, Anthropic's Annual Recurring Revenue (ARR) rapidly rose from $1 billion in December 2024 to $19 billion in March 2026, an approximately 19-fold increase within 15 months.
JPMorgan notes that China is currently equipped to follow a similar pattern: domestic model capabilities are already close to, or exceed, the level of leading U.S. models from a year ago, and domestic pricing aligns better with local labor economics—boosting commercial returns for deployment.
Agent-side demand logic is also strengthening. The report points out that OpenClaw has become a key 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 own ecosystems.
API Pricing: Differentiation Is the Main Theme, Capability Determines Pricing Power
JPMorgan judges that API pricing is unlikely to move in just one direction, but more likely toward differentiation.
On the one hand, capability establishes pricing power. If a particular model can uniquely unlock high-value tasks such as agent coding, long-duration workflows, or enterprise-level reliability, clients will pay a premium, as the returns are measurable and independent of the per-token cost.
On the other hand, as hardware, system, and algorithm efficiency improves, unit inference costs will continue to decline, which will put pricing pressure on models that are "good enough but stop progressing."
The report concludes: Models staying at the cutting edge in terms of capability are expected to achieve both volume and price growth; models that fail to keep improving may see usage rise but prices fall, with uncertain profit margin prospects.
Competitive Focus: Shift from Price Wars to Model Capability
The report emphasizes this is the key difference from last year's discussions—in China especially, market focus previously centered on all-out price competition.
In agent use cases, customers are actually purchasing not cheap tokens, but the successful completion of tasks. Cited calculations show that in multi-step workflows, slight improvements in single-step reliability result in large gains in overall task completion rates (raising single-step success from 90% to 95% boosts 20-step completion rate from 12% to 36%).
This means that models with higher average token prices but stronger reliability may actually have lower actual cost per successful task.
JPMorgan believes companies with strong cutting-edge models can usually extend more easily into the low-end market, while companies relying on low prices find it harder to move upmarket. Therefore, competition increasingly centers on absolute model quality and engineering efficiency, not just price.
Industry Landscape: Survival of the Fittest, The Strong Get Stronger
JPMorgan maintains its judgment of a "life-and-death struggle" in the large language model foundational industry. The core logic: technical gaps are small, product cycles are endless, the business model converges to API sales, and companies that lose momentum can be swiftly eliminated.
The report notes that in China, the gap between companies in the large language model space is often much smaller than investors realize, making the market highly unstable. Companies must continue spending and iterating to avoid falling behind—standing still is not neutral, but implies loss of market position.
Regarding the trend of internet giants moving into B-end AI fields, the report believes this makes competition between independent model providers and big platforms more direct.
Alibaba has already clarified its strategic focus on cloud and AI; Tencent's newly launched agent products are now segmented for personal, developer, and enterprise scenarios. JPMorgan assesses that as platforms push B-end monetization more aggressively, the advantage of being "cloud-neutral" is weakening, with both sides now focusing more directly on model capability itself.
Profitability: Gross Margin Improvement Expected, Operating Leverage Yet to Be Verified
JPMorgan believes that for large language model providers maintaining a world-leading position, gross margins should rise as model and inference chip efficiency improves, with higher-value workloads supporting a more favorable revenue structure. However, the key question for profitability is whether gross profit growth can outpace the growth rate of R&D expenditures.
Drawing on Anthropic as a reference—even when its revenue reached $14 billion in February 2026, it simultaneously announced a $30 billion round of fundraising and emphasized ongoing frontier development, proving that high income does not mean training intensity has normalized.
JPMorgan maintains an "overweight" rating for Zhipu and MiniMax, with target prices of HKD 800 and HKD 1,100 respectively. It forecasts both Zhipu and MiniMax will turn profitable from 2029 onward. The report emphasizes that more important than the exact year of profitability is the trend of continued usage growth and improvement in unit economics.
~~~~~~~~~~~~~~~~~~~~~~~~
The above excellent content comes from Chase Trading Desk.
For more detailed analysis, including real-time dissection, frontline research, etc., please join [Chase Trading Desk ▪ Annual Membership]
Risk Disclaimer and Liability ClausesThe market has risks, investment needs caution. This article does not constitute personal investment advice, nor does it take into account the individual investment goals, financial situations, or needs of any user. Users should consider whether any opinions, viewpoints, or conclusions in this article fit their particular circumstances. Responsibility for investment based on this is borne by the user. ```