J.P. Morgan Raises Zhipu AI Target Price to HK$1,800, Maintains Overweight

Overview

This report primarily analyzes Zhipu AI's (2513.HK) newly released GLM-5.2 model and its impact on industry pricing logic. J.P. Morgan maintains its 'Overweight' rating on Zhipu AI and raises its Dec 2026 target price from HK$1,400 to HK$1,800. The core thesis is that GLM-5.2 achieved an effective price increase through pricing structure adjustments while improving performance via algorithmic optimization rather than compute scaling, thereby enhancing unit economics. The report notes that the LLM market is experiencing price divergence: mature capabilities face downward pricing pressure, while models with frontier capabilities can still secure premiums by addressing high-value workflows.

Core views

GLM-5.2 Achieves Substantive Price Increases and Margin Improvement. Although GLM-5.2's listed price appears similar to GLM-5.1, it eliminates previous low-price tiers, applying a higher unified rate across all usage volumes. Data shows GLM-5.2's blended API price is 13% higher than GLM-5.1. In peer comparison, its pricing is 1.2x that of Kimi K2.7-Code, 2.5x MiniMax M3, 4.9x DeepSeek V4 Pro, and 14.4x DeepSeek V4 Flash. More importantly, GLM-5.2's performance gains are primarily driven by reinforcement learning and post-training optimization, without significant increases in model parameter scale (744B total parameters, 40B active). This implies that with a broadly stable cost base, the company has secured higher effective pricing, thus lifting gross margins. Structural Divergence in LLM Pricing. The report highlights that as model capabilities proliferate and inference costs decline, prices for 'mature intelligence' are trending downward—DeepSeek exemplifies this trend, lowering the market-clearing price for 'good enough' solutions. However, GLM-5.2 demonstrates the other side of the pricing curve: newly unlocked frontier capabilities (e.g., coding, agents, enterprise workflow automation, long-context tasks) improve task completion rates, reduce retries, and save labor, making customers willing to pay premiums. This divergence suggests that pure capability commoditization leads to price wars, whereas model vendors continuously advancing toward high-value tasks can retain pricing power. Earnings Forecast Revisions and Valuation Support. Based on improved visibility into high-quality revenue growth driven by GLM-5.2, J.P. Morgan raised Zhipu AI's 2026-2030 revenue forecasts by 7%-16%. Consequently, adjusted net loss expectations have narrowed, with profitability potentially achievable in 2028 (net profit of RMB 1.287 billion). The target price is set at HK$1,800, based on a 30x P/E multiple applied to 2030E normalized EPS, discounted at a 15% WACC. This valuation multiple exceeds that of tier-1 Chinese internet companies, primarily reflecting expectations of >100% revenue CAGR from 2026-2030.

Analysis framework

J.P. Morgan employed a 'Capability Curve and Pricing Power' analytical framework. First, it assessed changes in unit economics by dissecting GLM-5.2's pricing structure and cost drivers (distinguishing between model scaling and algorithmic optimization). Second, it segmented the LLM market into 'mature intelligence' and 'frontier capabilities,' comparing pricing strategies across vendors (e.g., DeepSeek vs. Zhipu/Kimi) to demonstrate the inevitability of pricing divergence. Finally, leveraging the company's sustained leadership in benchmarks such as WebDev Arena (ability to repeatedly deliver SOTA), it evaluated whether competitive advantages can be maintained amid compressed model cycles, thereby deriving long-term monetization potential and valuation premiums.

Methodology notes

• Industry/Sector Analysis FrameworkSupply-demand framework

  Supply-Demand and Price Divergence in the LLM Market

  The report segments the LLM market by capability maturity: on the supply side, oversupply of mature capabilities drives prices down (e.g., DeepSeek); on the demand side, frontier capabilities exhibit relatively rigid demand due to their ability to solve complex, high-value tasks (e.g., coding, long-text), with relatively scarce supply supporting premium pricing. This segmented supply-demand perspective explains why leading frontier models can raise prices despite industry-wide price cuts.

• Company Fundamentals and Financial Framework

  Unit Economics Analysis

  The report focuses on the relationship between 'price' and 'cost' for GLM-5.2. By highlighting that performance gains derive from algorithmic optimization rather than compute stacking, it demonstrates that costs remain relatively stable while revenue-side pricing increases, directly improving gross margins. This is a key micro-level indicator for assessing profit inflection points in technology-driven companies.

• Valuation MethodologyPE/PEG valuation

  Forward P/E Valuation

  Given the company's current loss-making status, the report adopts forward valuation, using 2030E normalized earnings as the base, applying a 30x P/E multiple, and discounting back to present value at 15% WACC. This approach suits high-growth tech stocks not yet profitable in the near term, aiming to capture cash flow value upon long-term maturity.

Asset mapping & comparison

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

• Zhipu AI (2513.HK)

  Beneficiary, leveraging GLM-5.2's frontier capabilities and pricing strategy to enhance revenue quality and gross margin outlook.

  Strengths

  Repeated delivery of domestic SOTA models; GLM series ranks highly on WebDev Arena, demonstrating a track record of consistently proving frontier delivery capabilities.

  Weaknesses

  Currently loss-making with substantial R&D spending; dependent on computing infrastructure and external suppliers.

  Comparison

  Versus DeepSeek's ultra-low-price strategy, Zhipu AI pursues a high-end frontier path, with pricing significantly above the DeepSeek V4 series and closer to the premium tiers of Kimi and MiniMax.

  Risks

  Geopolitical risks, export controls, Entity List designation; intensified competition eroding pricing power; commercialization falling short of expectations.

Key data

• GLM-5.2 Blended API Price Increase+13%Compared to GLM-5.1 blended pricing
• GLM-5.2 vs. DeepSeek V4 Pro Price Multiple4.9xDemonstrates premium pricing power of frontier models
• 2026-2030 Revenue Forecast Revision7%-16%Reflects enhanced growth visibility from GLM-5.2
• 2028E Adjusted Net ProfitRMB 1.287 BillionAccelerated timeline to breakeven/profitability
• New Target PriceHK$1,800Previous target was HK$1,400

Impact & implications

For Zhipu AI, the successful launch of GLM-5.2 validates its logic of sustaining pricing power through technological iteration; if subsequent API usage remains resilient, it will confirm customer willingness to pay for superior task outcomes rather than merely seeking low-cost tokens. For the industry, this signals that competition among model-layer companies will shift from pure parameter scaling or low-price strategies toward the ability to continuously unlock high-value workflows (e.g., coding agents, enterprise automation). Investors should distinguish between price cuts driven by commoditization and price increases driven by capability enhancements—the former squeezes margins, while the latter may drive margin expansion.

Risks

• Export controls, geopolitical risks, and Entity List designation risks.
• Intensified industry competition, particularly as domestic peers narrow capability gaps through model iterations.
• Execution risks and earnings pressure from high and sustained R&D investment.
• Uncertainties in commercialization progress and customer adoption.
• Dependency on computing infrastructure and external suppliers may lead to cost and supply risks.

What to watch

• Demand resilience for GLM-5.2 following effective price increases, especially sustainability of API usage and enterprise workflow demand.
• Model iteration progress among domestic competitors, focusing on releases of Kimi K3 and DeepSeek V4.1 and their potential impact on Zhipu's pricing position.
• Zhipu AI's own GLM-5.5 release (expected August), which will be critical in testing its ability to continue breaking upward along the capability curve.