China's AI foundation-model sector is entering a capital-accelerated race, with clearer benefits for Zhipu AI while MiniMax still needs model validation

Overview

This report covers Zhipu AI and MiniMax within China's AI foundation-model industry. JPMorgan points out that competition in foundation models is shifting from simple model releases to a more capital-intensive cycle of sustained investment. Funding can support pre-training, post-training, reinforcement learning, evaluation infrastructure, and inference deployment, but capital itself cannot determine the competitive outcome. The report reaches differentiated conclusions on the two companies: Zhipu AI's financing more directly alleviates inference-service capacity bottlenecks and improves ARR conversion visibility; although MiniMax's financing strengthens financial resources, dilution costs occur immediately, while catch-up in model capability and commercialization still requires validation from future releases.

Core views

The core view is that capital is a necessary condition, not a sufficient one. Because of its already-strong demand, progress in the GLM model series, and commercial momentum, Zhipu AI is more likely to convert incremental inference resources into revenue over the next 12 months, leading to upward revisions in revenue forecasts and target price. MiniMax has long-term optionality value in multimodal, B2C and B2B products, and overseas expansion, but its relative model capability remains in catch-up mode, and the positive resource effect brought by financing still needs to be validated first through model upgrades and improved monetization.

Analysis framework

The report adopts a company comparison and valuation re-rating framework, separately assessing the impact of financing on training resources, inference capacity, ARR conversion, equity dilution, earnings forecasts, and target prices. For Zhipu AI, the focus is on the conversion path from demand to revenue; for MiniMax, the focus is on comparing the certain cost of financing dilution with the uncertain benefits of future model-capability improvement. Valuation for both uses 30x 2030E P/E, discounted back to the Dec-26 target price using a 15% WACC.

Methodology notes

• Valuation methodDiscounted forward P/E

  30x 2030E P/E discounted back using 15% WACC

  The report is based on normalized 2030E earnings, assigns a 30x P/E multiple, and discounts it back to the Dec-26 target price using a 15% WACC. This multiple carries a premium versus leading Chinese internet companies, mainly reflecting expectations of high revenue growth from 2026E to 2030E.

• Industry competition frameworkCapital conversion efficiency

  Conversion of capital investment into model capability, commercialization, and ARR

  Financing can extend the cycle of model training and infrastructure investment, but competitive advantage depends on technology direction, talent quality, engineering organizational efficiency, and the ability to convert compute investment into customer adoption and revenue growth.

• Company screening frameworkRepeated SOTA visibility, high-value workflow exposure, and pricing power

  repeated SOTA visibility, high-value workflow exposure and pricing power

  Zhipu AI is considered to fit this framework better, because the GLM series has repeatedly delivered leading domestic models and offers a clearer investment narrative in high-value workflows and pricing power.

Asset mapping & comparison

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

• Zhipu AI (2513.HK)

  Core positive target

  Strengths

  Financing eases constraints on inference capacity and training resources; the GLM model series has shown strong progress; demand, API services, and enterprise use-case momentum are solid; the ARR conversion path over the next 12 months is clearer.

  Weaknesses

  R&D investment remains high and profitability is still under pressure; free float is low and the year-to-date share-price increase has been very large, creating high volatility risk.

  Comparison

  Compared with MiniMax, Zhipu AI is considered closer to converting incremental capital into short-term revenue and long-term model leadership.

  Risks

  Export controls, geopolitics, entity-list risk, intensifying competition, compute supply and cost risk, and uncertainty in customer adoption.

• MiniMax Group Inc - H (0100.HK)

  Neutral watchlist target

  Strengths

  Financing strengthens financial resources; the company has foundations in multimodal models, consumer applications, B2B product coverage, and international expansion; it still has optionality value in developer adoption and long-term AI application scenarios.

  Weaknesses

  Relative to peers, model capability is still in catch-up mode; financing dilution is certain; evidence of improved near-term revenue growth and willingness to pay is insufficient.

  Comparison

  Compared with Zhipu AI, MiniMax's capital-conversion path depends more on future model-capability improvement and commercial validation, with lower certainty.

  Risks

  Litigation with U.S. film studios, intensifying competition, continued high R&D spending, profitability pressure, uncertainty in commercialization and customer adoption, and dependence on compute infrastructure and external suppliers.

• China AI foundation-model industry

  Thematic investment direction

  Strengths

  Larger financing scale enhances the ability to invest in model training, inference deployment, and commercialization infrastructure.

  Weaknesses

  Higher capital intensity raises barriers to entry, while also increasing cash burn and execution failure risk.

  Comparison

  Competition within the industry will shift from financing capability to capital-use efficiency, model iteration speed, and commercialization execution capability.

  Risks

  Misjudgment of technology paths, talent competition, price wars, regulatory and geopolitical constraints, and compute supply bottlenecks.

Key data

• Financing scale of China's independent LLM vendors since 2026More than US$20bnCumulative financing raised or announced through IPOs, placements, private rounds, and other methods.
• Zhipu AI financing sizeUS$4bn placement, about 4.4% of total share capitalThe financing will support model training and expansion of inference capacity, and the report believes it will help improve ARR conversion visibility.
• Change in Zhipu AI target priceHK$2,400, previous HK$2,000Overweight maintained; target price is based on 30x 2030E P/E and 15% WACC.
• Zhipu AI current priceHK$1,640.00As of 2026-07-10.
• Adjustment to Zhipu AI revenue forecast2026-30E revenue raised by 7-13%Reflecting better visibility on compute expansion and commercialization after stronger capital support.
• Zhipu AI free float ratioAbout 14% after placementThe report notes that after the sharp year-to-date share-price rise, the low free float ratio may still lead to extreme volatility.
• MiniMax financing sizeUS$2bnIncludes equity placement and convertible bonds, providing resources for model development, infrastructure, and commercialization.
• MiniMax dilution impactPlacement equals about 11% of share capital; an additional about 6% upon full conversion of convertible bondsThe report believes dilution costs are certain, while improvements in model capability and monetization still need to be proven.
• Change in MiniMax target priceHK$240, previous HK$300Neutral maintained; the cut mainly reflects financing dilution and limited confidence in the commercialization path.
• MiniMax current priceHK$268.60As of 2026-07-10.
• MiniMax EPS adjustment2026E adjusted net EPS cut by 4%, 2027-30E cut by 20%Mainly reflecting the impact of new share issuance and potential convertible-bond conversion.

Impact & implications

For investors, the industry's financing boom does not mean that all foundation-model companies will benefit equally. For Zhipu AI, the key implication is that incremental compute power is more likely to unlock existing demand and drive ARR upside over the next 12 months; for MiniMax, the key implication is that financing only extends the competitive window, and if subsequent models such as M3Pro fail to narrow the gap with peers, dilution will weigh on per-share value. At the industry level, model release quality, inference-resource utilization, customer adoption, and pricing power should be the core evidence used to judge capital returns going forward.

Risks

• Financing may fail to translate into improved model capability or growth in customer adoption.
• Intensifying foundation-model competition may lead to pricing pressure and delayed profitability.
• Heavy R&D and inference-infrastructure investment may compress margins and increase cash consumption.
• Export controls, geopolitics, and entity-list risks may affect the availability of compute power and supply chains.
• Zhipu AI's low free float ratio and sharp year-to-date gains may lead to severe share-price volatility.
• MiniMax faces certain equity dilution, while catch-up in model capability and commercial monetization still requires validation.
• MiniMax is exposed to legal risks including litigation with U.S. film studios.

What to watch

• Zhipu AI's GLM 5.5 release and whether it can continue to maintain leading domestic model performance.
• Whether MiniMax's M3Pro release can narrow the model-capability gap versus peers.
• The impact of peer model cycles such as DeepSeek V4 official on the industry's competitive landscape.
• Whether Zhipu AI's additional inference capacity can convert into ARR growth over the next 12 months.
• Whether MiniMax's multimodal products, API economics, and user willingness to pay improve.
• The impact of post-financing equity dilution, free float ratio, and trading volatility on valuation.
• Follow-on financing, compute investment, and commercialization efficiency among China's independent LLM vendors.