Repricing of China’s AI model layer after the rally: still positive on Zhipu AI and MiniMax, but entering a validation phase

Overview

This industry research focuses on the valuation, fundamentals, and trading path of listed Chinese AI foundation-model companies after their sharp rally. The report argues that Zhipu AI and MiniMax have been repriced by the market as the core listed names in China’s frontier model layer, and current share prices reflect not only end-2026 ARR targets but also expectations of continued rapid expansion after surpassing US$1bn ARR, similar to Anthropic. JPMorgan remains constructive overall and views technical pullbacks as potential buying opportunities, but emphasizes that the industry still faces uncertainty around competition, pricing, monetization, and fund flows.

Core views

The core views are: first, Zhipu AI and MiniMax remain structural beneficiaries of China’s AI adoption cycle, supported by strong API demand, improving monetization, and the early expansion of enterprise AI workloads. Second, Zhipu AI’s advantage lies in a clearer monetization framework and a more measurable execution path, with the market roughly pricing in ARR approaching US$1bn by year-end; MiniMax’s advantage is that if it can acquire and deploy compute more flexibly, it may achieve upside beyond the roughly US$700mn ARR implied by current pricing amid tight GPU supply. Third, the investment case going forward is no longer simply about 'having AI exposure,' but will shift toward selecting winners that can sustain leadership in model capabilities, coding/agent/enterprise use cases, pricing discipline, and ARR conversion.

Analysis framework

The report uses a four-dimensional framework to assess China’s listed LLM companies: first, the ARR, competition, and China-US technology gap already priced into share prices at the fundamental level; second, momentum and catalysts such as model launches, token growth, and pricing mechanisms; third, fund flow and technical factors including lock-up expiries, financing, index/Stock Connect inclusion, and scarcity premium; and fourth, the impact of ARR growth and potential IPOs of US frontier AI companies such as Anthropic and OpenAI on valuation anchors.

Methodology notes

• Valuation and fundamentalsARR and P/ARR benchmarking framework

  Using the historical valuation and ARR stages of Anthropic and OpenAI to infer the implied growth requirements for Chinese LLM companies.

  The report argues that at market capitalizations of about US$40-55bn, Zhipu AI and MiniMax are already being priced by the market on paths toward about US$1bn and US$700mn of ARR by end-2026, with implied rapid growth after surpassing US$1bn ARR.

• Competitive analysisRelative model capability ranking framework

  Observing both absolute model capabilities and relative positioning among domestic peers.

  New model launches need to be assessed for coding, reasoning, agents, long-context, and multimodal capabilities, and for whether their position versus competitors such as DeepSeek, Kimi, Alibaba, Tencent, and ByteDance is improving or deteriorating.

• Trading and fund flowsEvent timeline and fund flow framework

  Putting model launches, index inclusion, Stock Connect, lock-up expiries, refinancing, and new share supply into the same trading path.

  The report believes short-term lock-up expiries and financing may cause volatility, but inflows from index inclusion and Stock Connect can partly offset this; if fundamental execution remains unchanged, technical pullbacks can be viewed as better entry points.

Asset mapping & comparison

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

• Zhipu AI

  An independent Chinese LLM company; the report maintains an Overweight rating and views it as a listed proxy for China’s frontier model layer.

  Strengths

  Its monetization framework is relatively clear, API pricing and coding adoption are more visible, and it is progressing toward an end-year ARR path of about US$1bn.

  Weaknesses

  Current valuation already prices in high ARR growth and Anthropic-style follow-on growth; if growth normalizes, valuation multiples may compress.

  Comparison

  Compared with MiniMax, Zhipu AI has a more measurable execution path and a lower proportion of shares unlocking in the near-term 6-month window.

  Risks

  Declining relative model leadership, internet giants gaining a SOTA advantage, API price wars, a widening China-US gap, and weaker-than-expected ARR conversion.

• MiniMax Group Inc-H

  An independent Chinese LLM company; the report maintains an Overweight rating and assigns a HK$1,100 target price.

  Strengths

  If it can acquire and deploy compute more flexibly across regions, it may expand revenue faster in a GPU-constrained environment, with greater ARR upside elasticity.

  Weaknesses

  Near-term lock-up pressure is greater, with 39.0% unlocking at 6 months, and the market has already priced in about US$700mn of end-year ARR.

  Comparison

  Compared with Zhipu AI, MiniMax’s ARR upside potential depends more on compute deployment and model-cycle delivery, but short-term fund-flow pressure is heavier.

  Risks

  Unlocking supply, potential financing, weaker-than-expected model iteration, domestic peer divergence, and fading scarcity premium.

• China’s independent LLM model layer

  An industry thematic asset, driven by enterprise AI adoption in China, API demand, and model capability iteration.

  Strengths

  Token demand is strong, enterprise AI workloads are still in the early stage, and compute expansion can convert suppressed demand into revenue.

  Weaknesses

  The market is crowded with competition, and internet giants have advantages in distribution, data, compute, and subsidies.

  Comparison

  Compared with Anthropic/OpenAI in the US, China’s market is more fragmented, but catch-up speed and vertical-scenario capabilities may support valuations.

  Risks

  The revenue pool being fragmented across 6-8 players, breakthroughs by US frontier models, and weaker-than-expected validation of domestic GPU or inference compute.

Key data

• Post-IPO increaseabout 5-7xBoth Zhipu AI and MiniMax have undergone significant rerating since their IPOs in January 2026.
• Market-implied Zhipu AI end-2026 ARRabout US$1bnThe report argues that the current share price already prices in a path toward nearly US$1bn of API ARR.
• Market-implied MiniMax end-2026 ARRabout US$700mnIf compute acquisition and regional deployment become more flexible, MiniMax has ARR upside beyond current pricing.
• Anthropic March 2025 valuation referenceUS$61.5bn valuation, about US$1.4bn ARR, about 43.9x P/ARRUsed to illustrate the high-growth valuation anchor implied by the market for Chinese LLM companies.
• Assumed China-US model capability gapabout 9-12 monthsCurrent market pricing assumes Chinese players are still rapidly catching up with leading US models.
• Zhipu AI 6-month lock-up expiry5.8% post-IPO shares, July 2026Relatively lighter near-term lock-up pressure than MiniMax.
• MiniMax 6-month lock-up expiry39.0% post-IPO shares, July 2026Heavier near-term supply pressure, which may lead to a technical pullback.
• Scarcity premiumabout 25-30%The Cambricon case is used as a reference for the possible decline in scarcity premium after more AI companies list.

Impact & implications

The investment implication is that China’s AI model layer still has structural upside, but valuation tolerance has declined. If model cycles such as GLM 5.5 and MiniMax M3 continue to drive capability gains, usage growth, price increases, and ARR acceleration, current high valuations may still be supported; if growth falls short of an Anthropic-style path, domestic competition fragments the revenue pool, or the next generation of US models widens the technology gap again, share prices may face valuation compression. Short-term fund-flow volatility is viewed more as trading pressure than as invalidation of the fundamental thesis, provided the companies’ execution and competitive positions remain intact.

Risks

• ARR growth falls short of the path implied by current valuation, leading to P/ARR multiple compression.
• Zhipu AI or MiniMax loses relative model leadership in high-value use cases such as coding, agents, and enterprise workloads.
• Internet giants such as Alibaba, ByteDance, and Tencent launch SOTA models that are clearly superior to those of independent model companies, weakening the necessity of independent LLM companies.
• Broad API price competition resumes, preventing stronger models from translating into higher realized pricing.
• The next generation of US models delivers step-function capability improvements, widening the technology gap between China and the US again beyond 9-12 months.
• Strong token usage fails to convert effectively into recognized revenue and ARR.
• Lock-up expiries, potential equity financing, and more AI company listings weaken the scarcity premium.

What to watch

• The absolute capability improvement and domestic relative ranking changes after the releases of GLM 5.5 and MiniMax M3.
• The ARR trajectory, pricing resilience, and margin trends disclosed by Zhipu AI and MiniMax in mid-2026 and second-half 2026 results.
• Whether usage through OpenRouter, direct enterprise APIs, and cloud-platform channels can translate into higher recognized revenue.
• Whether DeepSeek V4 validates the feasibility of Huawei Ascend chips in frontier inference and alleviates China’s LLM compute bottleneck.
• Competitive pressure on independent model companies from internet-giant products such as Tencent Hunyuan 3.0, Alibaba Qoder/TRAE, and ByteDance Doubao.
• Passive and Southbound fund flows driven by inclusion in the Hang Seng Composite/TECH and Stock Connect.
• The major lock-up expiry windows in July 2026, October 2026, and January 2027, as well as potential equity financing.
• The impact of potential IPOs by Anthropic and OpenAI on global AI valuation anchors and public-market valuation methodologies.