Kimi K3 reprices China's frontier model premium, but Zhipu's 50%+ pullback is seen as overdone

Overview

This report discusses the impact of Kimi K3's release on China's AI frontier model competitive landscape, valuation multiples, and the investment ratings of Zhipu and MiniMax. J.P. Morgan believes Kimi K3 has significantly raised the domestic competitive standard for models, requiring the market to shift from a valuation framework based on a "persistent single leader" to one based on whether a company can repeatedly enter the frontier group across multiple model cycles. Even so, the report believes Zhipu's ARR trajectory has not been fundamentally weakened by Kimi K3, and the recent 50%+ share price decline has already over-reflected pessimistic expectations.

Core views

The core view is that Kimi K3 has indeed reduced the predictability of model leadership and therefore China's AI model companies should see a permanent reduction in leadership premium. However, the current Chinese model market is still constrained by compute supply rather than a fixed demand pool, meaning strong models can expand the market instead of merely redistributing revenue. Zhipu is still viewed as a member of China's frontier model group and has the conditions to continue expanding ARR; MiniMax has long-term optionality in multimodal and internationalization, but its model capability is still in catch-up mode, so the Neutral rating is maintained.

Analysis framework

The report evaluates company prospects by combining model release cadence, changes in frontier capability rankings, cross-sectional ARR comparisons, API pricing, compute supply-demand constraints, and the reset in valuation multiples. For Zhipu, the focus is on the current competitiveness of GLM-5.2, the iteration opportunity of GLM-5.3 and the 2T+ flagship model, and its commercialization capability in high-value workflows; for MiniMax, the focus is on its multimodal product portfolio, developer adoption, and whether it can re-enter the frontier model group.

Methodology notes

• Valuation methods2030E P/E multiple reset

  Cut the long-term valuation multiple from 30x 2030E P/E to 20x

  The report argues that Kimi K3 demonstrates that China's frontier model leadership will rotate frequently, making it unjustified to pay the prior premium for any single company's long-term technological lead; therefore, it adopts 20x 2030E P/E as the new valuation framework for Zhipu and MiniMax.

• industry_structurefrontier group membership

  Frontier group membership is more important than ranking No. 1 in any single instance

  In an environment of compressed model cycles, investors should focus on whether a company can repeatedly enter the frontier model group across multiple generations, rather than betting that one company will remain the sole leader over the long term.

• market_demandcompute-constrained market

  Compute-constrained market

  The report believes China's model market is currently mainly constrained by GPU and inference supply, and incremental service capacity will be absorbed relatively quickly by paid workloads; therefore, growth in Kimi K3 usage does not necessarily translate directly into revenue displacement for Zhipu or MiniMax.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Zhipu AI (2513.HK)

  Primary covered name, Overweight maintained

  Strengths

  It has repeatedly launched domestic SOTA-level GLM models; GLM-5.2 is still considered part of China's frontier group, with potential in high-value workflows, pricing power, and ARR expansion.

  Weaknesses

  Kimi K3 has shortened the leadership window of GLM-5.2; the market narrative has shifted from a sustained domestic leader to a frontier group member, and the valuation premium has been reduced.

  Comparison

  Compared with MiniMax, the report believes Zhipu better fits the framework of repeated SOTA visibility, high-value workflows, and pricing power.

  Risks

  Export controls, geopolitics, and entity list risks; intensifying competition; continued high R&D investment; uncertainty in commercialization and customer adoption; dependence on compute infrastructure and external suppliers.

• MiniMax Group Inc - H (0100.HK)

  Covered name under the same industry framework, Neutral maintained

  Strengths

  It has a multimodal + LLM product portfolio, 2C+2B product coverage, international expansion, and cost-performance advantages; OpenRouter usage indicates developer adoption.

  Weaknesses

  From a pure model capability perspective, it is still in catch-up mode and lacks clear evidence of a return to the frontier.

  Comparison

  Compared with Zhipu, MiniMax's strategic positioning has long-term optionality, but its current model leadership is insufficient to support a more positive rating.

  Risks

  Litigation with U.S. studios; intensifying competition; high R&D investment leading to execution and profitability pressure; uncertainty in commercialization and customer adoption; dependence on compute and external suppliers.

• Moonshot Kimi K3

  Industry catalyst and competitive benchmark

  Strengths

  It raises the standard of model capability in China and shows that Chinese labs can continue to approach U.S. frontier capabilities; higher API pricing reflects global commercialization potential.

  Weaknesses

  It creates competitive pressure on other Chinese model vendors and lowers market confidence in the valuation of any single company maintaining lasting leadership.

  Comparison

  Kimi K3 prompts the investment framework to shift from betting on a single winner to evaluating whether multiple companies can maintain frontier membership across cycles.

  Risks

  If compute supply improves and the market shifts toward demand constraints, competition may show up more directly as share and pricing pressure.

Key data

• Zhipu current priceHK$890.50 (20Jul26)The current price of 2513.HK disclosed in the report.
• Zhipu Dec-26 target priceHK$1,600.00Previously the target price was HK$2,400, lowered due to the valuation multiple reset.
• MiniMax current priceHK$193.10 (20Jul26)The current price of 0100.HK disclosed in the report.
• MiniMax Dec-26 target priceHK$160.00Previously the target price was HK$240, lowered due to the valuation multiple reset.
• Valuation multiple20x 2030E P/EBoth Zhipu and MiniMax were lowered from 30x to 20x.
• Discount rate15% WACCUsed in target price calculation to discount 2030E normalized earnings.
• Zhipu ARRapproximately US$1.0bnAs of July 2026, based on company communication, news, and market research.
• Combined ARR of four leading Chinese independent model vendorsapproximately US$2.1bnIncluding Zhipu, DeepSeek, MiniMax, and Kimi.
• Anthropic ARRapproximately US$69bnAs a reference for the commercialization scale of global frontier labs.
• Kimi K3 API pricingcached input US$0.30/mn tokens; standard input US$3/mn tokens; output US$15/mn tokensAbout 4x the previous-generation K2.7 Code version, reflecting its capability and cost positioning.

Impact & implications

The investment implication of the report is that after the release of Kimi K3, the valuation premium of China's AI model companies should be reduced, but the recent market may have misread competitive risk as a structural deterioration in Zhipu's commercialization trajectory. If Zhipu maintains its frontier group position with GLM-5.3 and the 2T+ flagship model, the current pullback could offer an opportunity to rebuild positions; MiniMax, by contrast, needs clearer evidence of a rebound in model capability, and its near-term risk-reward is less attractive than Zhipu's.

Risks

• Faster rotation in frontier model rankings, making it difficult for any single company's technological lead to persist.
• Export controls, geopolitics, and entity lists may constrain GPU access and global business expansion.
• Intensifying domestic AI model competition may compress pricing, traffic, and valuation multiples.
• Sustained high R&D and inference spending may delay the profitability inflection point and increase execution risk.
• There remains uncertainty around commercialization, customer adoption, and API demand growth.
• Dependence on compute infrastructure and external suppliers may create cost and availability risks.
• MiniMax also faces litigation risk related to U.S. studios.

What to watch

• Zhipu GLM-5.3's launch performance from late July to August, especially improvements in coding, agents, and production performance.
• Whether Zhipu's expected 2T+ flagship model in September to October can prove its training capability has entered the next scale phase.
• Whether Kimi K3's usage and pricing gains expand overall demand or translate into direct share pressure on other model vendors.
• ARR growth of China's leading independent model vendors versus the pace of compute capacity expansion.
• Whether enterprise customers continue to adopt multi-model routing, backup suppliers, and task-based workload allocation.
• Whether MiniMax shows clear evidence of returning to frontier model status, and whether its multimodal portfolio can translate into stronger commercialization.