JPMorgan: Under the open-source trend, China’s AI enters an era of “winner takes more”; raises its target price for Zhipu to 2,000 HKD.

```

The wave of open weights is reshaping the competitive landscape of China's large language model (LLM) market. The latest research report from J.P. Morgan points out that the open-source strategy is evolving into a "winner-takes-more" business model—the stronger the model, the more it can convert open distribution into paid monetization; those without sufficient model differentiation face faster price competition and the risk of traffic diversion.

Accordingly, J.P. Morgan gave divergent ratings for two Chinese AI companies. The report raised the target price of Zhipu from HKD 1,800 to HKD 2,000 through December 2026, maintaining an “Overweight” rating, citing that GLM-5.2 reinforced the argument that "open weight commercialization can create significant option value for leading model providers." At the same time, MiniMax’s target price was lowered from HKD 400 to HKD 300 with a “Neutral” rating, as M3 has yet to demonstrate sufficient model-driven pricing power.

J.P. Morgan noted that open weights expand the distribution range of models, but monetization capability still highly depends on model quality, iteration speed, and endpoint reliability. For leading models, open weights can leverage external GPU resources via cloud service providers (CSP), inference platforms, and private deployments to scale usage, while achieving monetization through multiple channels such as official APIs, licensing, platform SaaS, and revenue sharing; for less differentiated models, broader access actually accelerates price competition and traffic switching.

Open Weights: A Structural Shift in Monetization Logic

J.P. Morgan believes the market generally interprets the release of open weights as monetization leakage, which is directionally correct but not complete. The more critical issue is whether LLM providers can convert weakened access control into broader distribution and paid conversion.

The report points out that open weights do not make all API endpoints homogeneous. Official APIs still have systematic advantages in terms of model timeliness, caching strategies, throughput, latency, feature support, and service reliability.

The release of open weights is usually a published checkpoint, while the official API is a continuously evolving product—after release, model providers can continuously optimize instruction following, tool usage, long-context stability, and inference efficiency according to real-time traffic, and such improvements often do not fully sync to the open weights package. Therefore, under the same model name, the user experience between the official endpoint and third-party deployments may diverge over time.

J.P. Morgan illustrates this framework with two cases. In the DeepSeek V4 Pro case, according to OpenRouter data, the official path, with a lower price and higher cache hit rate (93.5%), has an estimated monthly workload cost of about $24 to $41, while some third-party paths can reach $85 to $196—a price difference of about 6 to 12 times. In the MiniMax M3 case, the nominal output prices across providers are similar, but the official path has advantages in cache hit rate, actual input cost, and response speed. According to Artificial Analysis, MiniMax is among the fastest M3 providers.

Zhipu: GLM-5.2 Strengthens Leading Position, Target Price Raised to HKD 2,000

J.P. Morgan maintains its “Overweight” rating for Zhipu and raises its target price to HKD 2,000, mainly because GLM-5.2 enhances the monetization potential of open weights.

The report believes Zhipu’s open weight strategy is more composed: it expands usage with broad access, while positioning official channels and higher service versions (such as the GLM-Turbo series) to meet quality-sensitive demands. GLM-5.2 uses the MIT license and continues to expand its distribution on global cloud platforms and inference providers—on AWS, GLM-5.2 is listed as an official SaaS API product; on Azure, Microsoft Foundry distributes GLM-5.2 through Fireworks AI.

J.P. Morgan points out that based on the current valuation, the market has largely priced in Zhipu’s guidance of $1 billion ARR by the end of the year. Further upside depends on whether strong open-weight models can scale through external infrastructure and distribution channels, rather than relying solely on Zhipu’s own GPU capacity. This is an option value rather than a certain near-term revenue, and can only scale if Zhipu maintains its model leadership. Key observation points include the performance comparison of GLM-5.5/6, Kimi K3, and DeepSeek V4.1.

MiniMax: Lack of Pricing Power, Open Weight Pressure Amplifies Downside Risks

J.P. Morgan maintains a “Neutral” rating for MiniMax and lowers its target price from HKD 400 to HKD 300. The main reason is that open weight commercialization is becoming a “winner-takes-more” framework, and M3 has yet to show sufficient evidence of model-driven pricing power.

The report argues that M3’s permanent 50% discount is a key signal, indicating the model has not achieved a capability premium compared to domestic leaders. The discount may support short-term usage, but it reduces confidence in model-driven monetization. With expanded access through open weights, undifferentiated models face quicker price competition and easier traffic diversion—broader distribution in this context magnifies risks rather than creating upside potential.

J.P. Morgan also notes MiniMax’s relative advantages: The company remains relevant in multimodal AI, overseas usage, and agent workflows. M3, with 1 million token context, native multimodality, and MiniMax Code, improves its product narrative, and OpenRouter usage indicates developer adoption has reached considerable scale. However, workflow monetization requires stronger model pull—code or agent products must drive significant task improvement to change user habits, rather than just providing another extensively accessible model path.

J.P. Morgan also highlights that both companies are at a capital-intensive stage, and financing needs pose potential risks.

Analysts expect both companies to conduct two more financing rounds in 2026 and 2027. Although current cash balances are sufficient to support operations under the base scenario, accelerated model iteration, larger-scale overseas deployment, or higher-than-expected inference costs may cause both to seek more external funding.

Risk Disclosure and DisclaimerThe market involves risks, investments require caution. This article does not constitute personal investment advice, nor does it take into account the specific investment objectives, financial situation or needs of any individual user. Users should consider whether any opinions, views, or conclusions in this article fit their own circumstances. Investments based on this information are at your own risk. ```