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

In a report released on March 27, JPMorgan Chase systematically addressed ten key market concerns, including demand growth, API pricing, competitive landscape, profitability, and risks associated with global expansion.

The report identifies 2026 as a pivotal year for determining whether enterprise AI demand in China can replicate the growth trajectory seen in the US market in 2025. Coding and AI agent applications are emerging as the most significant demand catalysts.

**Demand Acceleration: A Non-linear Inflection Point, with Coding and Agents as Key Drivers**

JPMorgan Chase posits that AI demand should be understood as "inflection point-driven" rather than linear. Once model capabilities surpass a certain threshold, enabling the large-scale automation of real-world workflows, demand is expected to expand rapidly.

The US market provides a precedent. According to data cited in the report, Anthropic's Annual Recurring Revenue (ARR) surged from $1 billion in December 2024 to $19 billion by March 2026, representing an approximate 19-fold increase within 15 months.

JPMorgan Chase argues that China currently possesses the conditions to follow a similar pattern. Domestic model capabilities have reached or even surpassed the level of leading US models from a year ago. Furthermore, domestic pricing is more aligned with local labor economics, collectively improving the commercial returns on implementation.

The demand logic for AI agents is also strengthening. The report highlights OpenClaw as a significant catalyst, shifting usage scenarios from single-turn interactions to multi-step executions, thereby substantially increasing token intensity per task. Tencent, Alibaba, and ByteDance have already integrated tools linked to OpenClaw into their respective ecosystems.

**API Pricing: Divergence is the Trend, Capabilities Dictate Pricing Power**

JPMorgan Chase anticipates that API pricing is unlikely to move in a single direction and is more likely to diverge.

On one hand, capabilities create pricing power. If a model can uniquely enable high-value tasks—such as agentic coding, long-horizon workflows, or enterprise-grade reliability—customers will pay a premium because the return is quantifiable and independent of the token price.

On the other hand, as hardware, systems, and algorithmic efficiencies continuously improve, the unit cost of inference will keep declining. This will exert pricing pressure on models that are "good enough" but have stopped advancing.

The report concludes that models remaining at the capability frontier are poised to achieve both volume and price increases simultaneously. Conversely, models that fail to improve continuously are more likely to face price declines even as usage grows, casting uncertainty over their profit margin outlook.

**Competitive Focus: Shifting from Price Wars to Model Capabilities**

The report emphasizes a key difference from discussions held last year—particularly in China, where the market's focus was previously on comprehensive price competition.

In AI agent usage scenarios, customers are essentially purchasing the successful completion of a task, not just cheap tokens. Citing calculations, the report notes that in multi-step workflows, a minor improvement in single-step reliability can lead to a substantial increase in the final task completion rate (e.g., increasing single-step success from 90% to 95% can boost the completion rate for a 20-step workflow from 12% to 36%).

This implies that a model with higher average token prices but superior reliability might actually result in a lower effective cost per successfully completed task.

JPMorgan Chase believes companies with powerful, cutting-edge models can typically extend their reach into the lower-end market more easily, whereas companies built on low prices find it difficult to move upmarket. Consequently, competition is increasingly concentrating on absolute model quality and engineering efficiency, rather than on price alone.

**Industry Structure: A Battle for Survival, Favoring the Strong**

JPMorgan Chase maintains its assessment of the large language model industry as a "battle for survival." The core logic rests on small technical gaps, endless product cycles, business models converging on API sales, and the rapid obsolescence of companies that lose momentum.

The report points out that in China, the capability gaps between major LLM companies are often much narrower than investors perceive, making the market highly volatile. Companies must continuously invest and iterate to avoid falling behind—stagnation is not neutral but equates to a loss of market position.

Regarding the trend of internet giants expanding into enterprise AI, the report suggests this makes competition between independent model providers and large platforms more direct. Alibaba has clearly positioned cloud and AI as strategic priorities, while Tencent's recently launched agent products are already segmented for individual, developer, and enterprise scenarios. JPMorgan Chase judges that as platforms more aggressively pursue enterprise monetization, the positioning advantage of "cloud neutrality" is weakening, and the competitive focus for all players is converging on model capability itself.

**Profitability: Gross Margin Improvement Expected, Operating Leverage Yet to Be Proven**

JPMorgan Chase believes that for LLM providers maintaining a top-tier global position, gross margins should rise alongside improvements in model efficiency and inference chip efficiency. Higher-value workloads will also support a more favorable revenue mix. However, the more critical question for profitability is whether the growth rate of gross profit can outpace the growth rate of research and development expenditures.

The report uses Anthropic as a reference—even as the company reached a revenue run-rate of $14 billion by February 2026, it concurrently announced a new $30 billion funding round emphasizing continued frontier development, underscoring that high revenue does not mean training intensity has normalized.

JPMorgan Chase maintains "Overweight" ratings on both Zhipu AI and MiniMax, with target prices of HKD 800 and HKD 1,100, respectively. It forecasts that both Zhipu AI and MiniMax will achieve profitability starting in 2029. The report also emphasizes that more important than the exact year of achieving profitability are the trends of sustained usage growth and improving unit economics.