China's foundational AI model industry is transitioning from an &#34;expectation-driven&#34; to a &#34;demand-driven&#34; phase. In a recent research report, J.P. Morgan systematically addressed ten core questions from investors regarding this sector, asserting that model quality has become the primary determinant of market structure, and industry divergence will accelerate.
According to the report released by J.P. Morgan on March 27, the Chinese AI market is at a distinct inflection point. Demand growth in coding and agent scenarios is accelerating. Domestic model capabilities have reached or surpassed the level of leading U.S. models from a year ago, while local pricing is more economically viable, collectively improving the returns on deployment.
2026 will be a pivotal year for Chinese enterprises' AI demand to potentially replicate the growth trajectory seen in the U.S. in 2025. Using Anthropic as a benchmark, its Annual Recurring Revenue (ARR) surged from $1 billion in December 2024 to $19 billion in March 2026, an approximate 19-fold increase within 15 months.
The Chinese market possesses the conditions to follow a similar path, especially in the coding domain. Internet giants such as Tencent, Alibaba, and ByteDance have integrated relevant tools into their existing ecosystems, driving demand from standalone demonstrations to comprehensive deployment. The firm maintained its &#34;Overweight&#34; rating on Zhipu and MiniMax, with target prices of HKD 800 and HKD 1100 respectively.
<h2>Question 1: Is AI demand linear or explosive at an inflection point?</h2>
Demand is driven by inflection points, not linear growth.
As long as model quality is sufficient to unlock real-world application scenarios, usage will shift from linear growth to an &#34;upward convex curve&#34; explosion. The most compelling evidence comes from the U.S. market: Anthropic's Annual Recurring Revenue (ARR) soared from $1 billion in December 2024 to $19 billion in March 2026, a nearly 19-fold increase in just 15 months.
China currently possesses the foundational conditions for a similar boom: domestic model capabilities have surpassed those of leading U.S. models from a year ago, and local pricing aligns better with China's economic realities, collectively improving the expected returns on AI implementation.
On the agent side, OpenClaw has become a significant catalyst, pushing use cases from single-turn interactions to multi-step task execution, thereby substantially increasing the token consumption per task. Internet giants like Tencent, Alibaba, and ByteDance have integrated OpenClaw-related tools into their existing ecosystems, signaling a transition from &#34;developer experiments&#34; to &#34;comprehensive ecosystem deployment&#34;.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/f91ead11-e0af-4276-adfa-0118c70edc49.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/f91ead11-e0af-4276-adfa-0118c70edc49.png"/>
<h2>Question 2: Will API pricing rise, fall, or diverge?</h2>
Pricing will not move unilaterally; divergence is the main theme.
On one hand, models with superior capabilities command pricing power. If a model can uniquely unlock high-value tasks (e.g., agent coding, long-term workflows, enterprise-grade reliability), customers are willing to pay a premium because the returns are quantifiable. On the other hand, as hardware and algorithmic efficiency continue to improve, the cost per unit of inference will steadily decline, exerting price pressure on models with stagnant capabilities.
The eventual outcome is a differentiated pricing structure: models that consistently maintain cutting-edge capabilities can achieve simultaneous increases in both volume and price; models that fail to iterate continuously will face price declines, and even with growing usage, their profit margins will become uncertain.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/af694845-d761-4a9f-82bd-5ee2e6f5fca7.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/af694845-d761-4a9f-82bd-5ee2e6f5fca7.png"/>
<h2>Question 3: If pricing isn't the main battlefield, where is the focus of competition?</h2>
The main battlefield has shifted from token prices to model capabilities.
This is a key change compared to last year – the focus in the Chinese market in 2025 was a comprehensive price war, whereas now, quality is far more important than unit price in the fastest-growing coding and agent scenarios.
In multi-step workflows, customers are essentially buying &#34;successful task completion,&#34; not &#34;cheap tokens.&#34; The research report provides an intuitive mathematical example: if the success rate of a single step increases from 85% to 98%, the final completion rate for a 20-step task will jump from 4% to 67%. Under this logic, the model with the lowest price per token might actually have the highest effective comprehensive cost per task.
The report also points out that companies with strong cutting-edge models can easily extend into the low-end market, but companies relying solely on low prices struggle to move upmarket.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/ce6868eb-2702-46a8-b480-680549258b02.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/ce6868eb-2702-46a8-b480-680549258b02.png"/>
<h2>Question 4: Why is the foundational large model sector still a &#34;life-and-death struggle&#34; industry?</h2>
Small technical gaps, endless iteration cycles, and converging monetization models – these three factors dictate the industry's extreme ruthlessness.
The capability gaps between different large model companies in China are often smaller than investors anticipate, making the market highly unstable. In this industry, &#34;standing still&#34; is not a neutral outcome but signifies a loss of position – companies must continuously invest and iterate to avoid falling behind.
The convergence of business models exacerbates the pressure of elimination. Both revenue growth and profit margins primarily depend on product strength, and switching costs remain low. This means companies that lose technological momentum will rapidly lose their defensive capabilities commercially and financially, and the number of truly reliable companies in the industry will gradually decrease.
<h2>Question 5: What determines profitability?</h2>
The core issue is whether gross profit growth can consistently outpace the growth of R&amp;D expenditure.
The basic economic model for the token business is clear: Revenue = Token Volume × Price, with the main cost being inference computation and the largest operational expense being training-related R&amp;D. As model efficiency and inference chip efficiency continue to improve, the gross profit margin for cutting-edge models should gradually increase.
However, the outlook for operating profit is more complex. Anthropic serves as a cautionary example: even with monthly revenue reaching $14 billion in February 2026, the company announced a new $30 billion funding round concurrently, emphasizing continuous cutting-edge development – high revenue does not equate to normalized training intensity.
The baseline scenario is that both Zhipu and MiniMax are projected to achieve profitability from 2029 onwards. The report emphasizes that more important tracking metrics than the specific year of profitability are the sustained growth trend in usage volume and the continuous improvement in unit economics.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/d8e6048b-6799-403d-9766-88651489b0cc.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/d8e6048b-6799-403d-9766-88651489b0cc.png"/>
<h2>Question 6: How should investors track model strength?</h2>
It requires combining token price, usage volume, and third-party evaluations; a single metric is insufficient.
<ul>
<li>
Token Price: This is the most important indicator as it represents the company's real-time expression of its product's market positioning. The price difference with the best models is becoming a good proxy for actual competitive strength.
</li>
<li>
Token Usage Volume: Actual consumption reflects the genuine choices of users and developers. Third-party API aggregators like OpenRouter can serve as references, with particular attention needed for the growth of agent-based workloads, as these consume significantly more tokens per task than simple workflows.
</li>
<li>
Third-Party Evaluations: Artificial Analysis provides structured evaluations, and LMArena reflects the blind preference of real users. These two complement each other to form a more complete external perspective.
</li>
</ul>
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/88127787-a0c0-4cd4-a273-e71f17ebed0f.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/88127787-a0c0-4cd4-a273-e71f17ebed0f.png"/>
<h2>Question 7: With internet giants heavily entering the B2B space, what is the future for independent model companies?</h2>
Competitive boundaries are converging, ultimately returning to the competition of model capabilities.
Alibaba has clearly prioritized Cloud and AI, deeply integrating model development with enterprise workflows. Tencent's agent products now cover personal, developer, and enterprise scenarios. OpenAI is also shifting its commercial focus to enterprise products and coding deployment. The direction for leading companies is consistent: AI is evolving from a &#34;consumer-side feature&#34; to a &#34;tool that directly generates enterprise revenue.&#34;
In this context, independent model companies can no longer rely solely on the &#34;cloud-neutral&#34; label for a moat, nor can internet giants fully compensate for model deficiencies solely through ecosystem traffic advantages. When deploying AI, enterprise customers primarily purchase model quality – stronger coding inference capabilities and more reliable workflow completion rates.
<h2>Question 8: What factors determine a company's survival?</h2>
Talent first, compute second, organization third – all three are indispensable.
<ul>
<li>
Top Research Talent: This remains a research-driven industry. High-level technical judgment is itself a competitive factor; management's ability to make correct decisions on research directions directly impacts the company's technological trajectory.
</li>
<li>
Compute Power and Capital: Cutting-edge training is prohibitively expensive, and inference economics depend on infrastructure quality. Weak compute acquisition capability is a structural disadvantage – it not only affects model training efficiency but also weakens the ability to respond to demand at a reasonable cost.
</li>
<li>
Organizational Execution: In a rapidly iterating market, the ability to translate research findings into products, products into usage, and usage into monetization is almost as important as the model itself.
</li>
</ul>
<h2>Question 9: If everyone is progressing, will models eventually converge? Will there be a winner-take-all market?</h2>
Overall strength will approach, but not converge; the market will not form a winner-take-all structure.
Different companies have divergences in architectural choices, training data, product focus, and technological paths. These differences will continue to yield distinct capability advantages. The report suggests that in a still rapidly expanding market, multiple companies can grow simultaneously, even with some overlap in capabilities – the expansion of the overall market at this stage is far more significant than prematurely worrying about commoditization.
In the long run, a more realistic market endgame is not &#34;one dominant player and the rest exit,&#34; but rather a few truly capable companies remaining, each with its own areas of strength, competing in a market large enough to support multiple winners. As AI expands from productivity tools to consumer scenarios, differences in individual tastes, styles, and preferences will further reinforce this diversified landscape.
<h2>Question 10: How to unify understanding of open-source/closed-source, model iteration, and global expansion risks?</h2>
Iteration is mandatory, open-source/closed-source are strategic choices, and the core risks of global expansion lie in compute power and compliance.
Regarding model iteration, the expected pace is roughly one flagship model generation per year (e.g., from GLM 4.7 to GLM 5, MiniMax M2 series to M3 series), with minor upgrades driven by reinforcement learning in between. Halting iteration means losing competitive standing.
On open-source vs. closed-source, the report believes the answer is not either/or. Closed-source models offer stronger commercial defensibility and reduce the risk of disintermediation; open-source aids ecosystem building, increases adoption rates, and accelerates technical feedback. Therefore, most Chinese model companies will ultimately adopt a hybrid strategy: keeping the latest and strongest models closed-source while open-sourcing some other versions.
In terms of global expansion, the biggest risk remains compute power acquisition. Both training and inference are highly dependent on high-performance chips, and tightening export controls will simultaneously weaken model progress speed and cost competitiveness. The second risk is data and security compliance: if model deployment, user services, and data storage can be localized overseas, cross-border data transfer issues are relatively manageable; however, local privacy regulations and the determination of data access rights for Chinese affiliated entities remain sources of uncertainty.

00100

02513

J.P. Morgan systematically answers ten core questions about the AI industry. China's AI sector is shifting from a price war to a focus on model quality. In agent scenarios, 'task completion rate' is far more critical than 'token price' for customer retention. The core profitability challenge is whether gross profit growth can consistently outpace R&D expenditure growth. Both Zhipu and MiniMax are projected to achieve profitability from 2029 onwards, and the overall market will not be a winner-take-all scenario

- China's artificial intelligence (AI) foundational model industry is shifting from an "expectation-driven" phase to a "demand-driven" stage, according to a JP Morgan report addressing ten key investor questions.  
- The report notes a significant turning point in China's AI market, with domestic model capabilities approaching or surpassing those of leading U.S. models from a year ago and an increasing demand for coding and intelligent agent scenarios.  
- It highlights 2026 as a critical year for Chinese enterprises to replicate the growth trajectory witnessed in the U.S. in 2025, with major firms integrating related tools into existing ecosystems to drive demand.

"Token New Era" – A "Ten Questions, Ten Answers" on China's AI Industry