同为大模型 为何智谱万亿 Minimax却腰斩？

This article analyzes the starkly different stock performance of two leading Chinese large language model (LLM) firms, Zhipu AI and MiniMax, which both listed on the Hong Kong Stock Exchange in the same week. Zhipu AI’s market capitalization surpassed HK$1 trillion six months after its IPO, while MiniMax’s share price has fallen by half from its peak. The divergence stems from two key dimensions: core technological strength and capital market structure. For general readers interested in the AI industry or AI-related investments, the key takeaways are:

1. The LLM industry has moved past the unregulated "war of a hundred models" phase and entered a second half defined by hard technological strength and business certainty. Valuations are no longer driven by narrative, but by fundamentals. To assess an LLM company’s value, the first question is whether it has built unreplaceable competitive barriers.

2. When investing in LLM-related stocks, investors need to pay attention to shareholding structure in addition to business fundamentals. Lock-up expiration scales and shareholder background both materially impact price volatility. Stocks with a high share of financial investors and large near-term unlocking volumes face greater selling pressure and more significant valuation downside.

3. The certainty of a future secondary listing on mainland China’s A-share market also affects valuations in Hong Kong. Companies with clear, well-advanced A-share listing plans earn an additional scarcity premium, while high uncertainty around this process erodes the premium and weighs on share prices.

This analysis of valuation divergence among LLM companies offers multiple actionable takeaways for brands operating in the LLM space or adopting AI for their businesses:

1. Brand marketing: Zhipu AI capitalized on the opportunity created by Anthropic’s ban on overseas access, crafting the precise positioning that "cutting-edge AI should not belong to a small handful of players." It leveraged this high-profile event to upgrade its brand positioning, and further entered the global top-tier AI discussion circle via interactions with industry icon Elon Musk. This opportunistic marketing strategy is well worth learning from for brands.

2. Product R&D and pricing: If a product has built sufficient irreplaceability, price increases can actually drive higher sales. Conversely, without core competitive barriers, even rapid price cuts after a hike cannot win back user trust, and will only expose a company’s lack of pricing power.

3. Customer strategy: Against the current geopolitical backdrop, focusing on B-end clients in government, finance, energy and other sectors with strong demand for on-premise data storage and supply chain security allows companies to build a stable core business, gain institutional access barriers, and capture niche market dividends.

For sellers building LLM-related businesses, this article provides key insights on industry opportunities and risks:

1. Opportunities: The growing risk of overseas LLM supply cuts has spurred surging domestic demand for autonomous, controllable cutting-edge LLMs. Both B-end private deployment and cloud-based MaaS (Model-as-a-Service) for developers have strong high-growth potential. Players that build irreplaceable technological barriers will be able to capture broad industry dividends.

2. Risks: For C-end AI application developers, if core advantages only extend to interaction experience and cost competitiveness, it is easy for large tech incumbents to erase those gaps via subsidies and feature iterations. Meanwhile, cross-border business faces geopolitical risks—if foreign governments introduce restrictive policies, growth will be hit hard.

3. Capital strategy: For companies preparing for an IPO, it is critical to plan shareholding structure and lock-up expiration schedules in advance, prioritize bringing in strategic shareholders, and avoid selling pressure from concentrated unlocking by a large number of financial investors. Companies should also map out a clear A-share secondary listing path early to secure certainty, which supports sustained valuation in capital markets.

For traditional manufacturers looking to leverage LLMs to drive digital transformation and expand business opportunities, this article offers the following key insights:

1. Demand perspective: There is now strong domestic demand for autonomous, controllable LLMs, and mature domestic LLM providers already support almost all domestic computing platforms including Huawei Ascend and Cambriccon. These solutions meet factories’ security requirements for on-premise data storage, meaning manufacturers do not need to over-rely on overseas LLMs for digital transformation—viable domestic alternatives are already available.

2. Opportunities: As the LLM industry enters its second half, demand for infrastructure, computing adaptation and scenario implementation will increasingly concentrate on leading autonomous LLM players with strong hard capabilities. Manufacturers that develop implementation scenarios or build hardware adaptations around leading domestic LLMs will gain access to more business opportunities.

3. Transformation guidance: When advancing digital and e-commerce-related intelligent upgrades, factories should prioritize partnering with domestic LLMs that have irreplaceable capabilities and high long-term growth certainty. Avoid partnering with model providers with unstable valuations and insufficient barriers, to reduce the risk of supply disruptions or falling behind on technological iterations down the line.

For service providers serving the AI industry or enterprise digital transformation, this article outlines the latest LLM industry trends and core client pain points, with the following key takeaways:

1. Industry trends: The LLM industry has finished its unregulated "war of a hundred models" phase and entered a second half where valuations are tied to hard technological strength. Industry divergence is accelerating: leading players with solid competitive barriers will capture more resources and higher valuations, while the space for small and mid-sized players continues to shrink.

2. Core client pain points: For current B-end clients, especially those in finance, energy and government sectors, the top pain points are supply chain security and data security amid rising geopolitical risk. Demand for autonomous, controllable LLMs is extremely urgent, and clients also have high requirements for technological capability and cost control.

3. Strategic direction: Service providers can expand supporting services centered around leading autonomous LLMs, including implementation services for private deployment, adaptation and debugging services for domestic computing power, and scenario development services for MaaS offerings. This allows players to capture the current domestic substitution opportunity and grow their business by addressing core client pain points.

For platform operators building AI and LLM-related businesses, this article outlines the current logic of industry divergence and key risk points, with the following key takeaways:

1. Demand perspective: Developers and B-end clients now prioritize autonomy, controllability, technological leadership and supply stability when choosing LLM platforms. When onboarding LLM service providers, platforms should prioritize leading domestic LLMs with irreplaceable barriers and autonomous controllability to better meet core client needs.

2. Business development and operations: Platforms should prioritize partnering with leading LLM companies backed by strategic capital and with high certainty of an A-share listing. These firms have strong growth certainty and can bring stable traffic and business growth to platforms.

3. Risk mitigation: Platforms should be wary of LLM companies that lack core barriers, have a high share of financial investors in their shareholding structure, and face large unlocking pressure. These firms have high valuation volatility and high business uncertainty, so platforms should fully assess valuation risk and business stability before cooperation to avoid unnecessary negative impacts.

For researchers studying the LLM industry, this article identifies the latest development trends and core issues for the industry’s second half, with the following key insights:

1. New industry dynamics: China’s LLM industry has exited the narrative-driven growth stage and entered a second half where valuations are tied to hard capabilities. Valuation divergence is determined by two core dimensions: irreplaceability sets the ceiling for valuation premiums, while shareholding (chip) structure determines the depth of drawdowns. This pattern has become the new valuation logic for China’s LLM industry.

2. New open questions: While leading domestic autonomous LLMs have earned high market valuations, they also face the risk of overvaluation. Zhipu AI currently trades at a price-to-sales ratio of 450x, far exceeding OpenAI’s 34x and Anthropic’s 22x. Even with a sustained 100% compound annual growth rate, it would take roughly four years for earnings to catch up to the current valuation, indicating notable valuation bubble risk.

3. Business model validation: This article verifies the feasibility of the popular LLM business model that uses private deployment as a stable base and cloud MaaS to open up long-term growth space. Zhipu AI’s MaaS business has already achieved a proven positive flywheel of growing call volume → economies of scale → higher gross margins → reinvestment in R&D, providing a mature, researchable case study for the industry.