智谱市值上万亿 半个公司身家过亿

This article tells the turnaround story of Chinese large language model (LLM) firm Zhipu AI: just six months after its public listing, its market cap surged over 18x from HK$52.8 billion to surpass HK$1 trillion. Key takeaways are as follows:

1. Zhipu's growth journey: After its early consumer-facing products failed to gain traction, Zhipu pivoted to customized private deployments for government and enterprise clients—a segment once dismissed by capital markets. For years, it focused on this unglamorous work, accumulated scenario-specific data and built up adaptation capability for domestic chips. It later launched a series of world-class open-source SOTA (state-of-the-art) LLMs, driving explosive growth in its API revenue and completely reversing market perception of the company.

2. Key lessons: Startup success does not always require chasing hot, "sexy" asset-light business models. Deep diving into underserved vertical scenarios and focusing on sustained technology iteration can instead build defensible moats that competitors cannot replicate. For early employees, joining a technology-driven startup and securing equity can deliver outsized financial returns.

This article offers valuable takeaways for brands operating in the LLM industry, as outlined below:

1. Pricing and competition: Against an industry-wide trend of cutthroat price competition, Zhipu AI raised prices despite its leading SOTA technology. Following the price hike, model call volumes grew 400% year-over-year. This demonstrates that core technological capability wins far more market recognition than competing on low prices, and technical moats are the core foundation of a brand's pricing power.

2. Go-to-market strategy: Zhipu first focused on building trust and refining its technology through serving high-requirement government and enterprise clients, before extending that trust into general commercial markets and consumer segments. This path is far more sustainable than burning cash to win mass consumer adoption upfront, and serves as a strong reference for technology-focused brands.

3. Brand building: When top closed-source LLMs were pulled from the Chinese market, Zhipu seized the industry pain point and open-sourced a full LLM of comparable capability, positioning itself around the tagline "Technology should not be limited to a small few". This quickly won recognition from global developers and built a strong positive brand reputation.

This article outlines key opportunities, lessons and risk warnings for industry practitioners and sellers in the LLM space:

1. Market opportunities: There is enormous room for growth in open-source LLMs, government and enterprise private deployments, and domestic technology adaptation for China's "Xinchuang" domestic digital innovation initiative. Global developers have strong, unmet demand for cost-effective open-source LLMs that match the capability of top closed-source models while allowing full independent control. Meanwhile, government and enterprise clients have long-term stable demand for on-premise deployment driven by strict data security requirements.

2. Key lessons: Instead of blindly pursuing light, standardized business models, focusing on unglamorous deep work in vertical industry scenarios allows players to accumulate unique scenario data and build technical moats. Capabilities honed through B-side projects can in turn empower core products, driving growth across the entire business.

3. Risk warnings: Valuation bubbles are already obvious in the current LLM industry. Zhipu supports a HK$1 trillion market cap with just HK$700 million in annual revenue, resulting in a 1,000x price-to-sales ratio that is unsustainable. The scarcity premium driven by low free float is likely to fade after lock-up expirations. In addition, the industry's pace of technological iteration is extremely fast: companies that cannot keep up with the SOTA frontier will be淘汰 quickly, requiring sustained heavy R&D investment.

This article offers key insights for traditional manufacturing factories looking to integrate AI and advance digital transformation:

1. Business opportunities: Virtually all physical industry sectors—including finance, energy, petrochemicals and government services—now require custom LLMs adapted to their specific use cases, with strong demand for private deployments that keep data within internal networks and adaptation solutions for domestic chips. Factories can leverage their deep industry experience to partner with LLM vendors and expand into this AI service business, unlocking new growth.

2. Insights for digital transformation: Zhipu's path of starting with small use cases, gradually accumulating data to refine core capabilities, and ultimately achieving a market breakthrough is applicable to factory digital transformation. Factories do not need to pursue a perfect, general-purpose solution from the start. They can instead start by addressing core needs in their own key production scenarios, and iterate and refine capabilities incrementally.

3. Guidance on partnerships: Factories that lack in-house LLM R&D capability can partner with established players like Zhipu AI. They can build industry-specific LLMs adapted to their own production processes based on Zhipu's standardized base model, and leverage the vendor's training systems to address gaps in in-house secondary development capability, lowering overall transformation costs.

This article summarizes key industry trends, pain points and actionable solutions for AI industry service providers:

1. Industry trends: LLM competition has now entered an era defined by SOTA (state-of-the-art) technological competition. Players that cannot keep up with the iteration pace will be immediately pushed out of the top tier of the industry. Open-source LLMs have become the mainstream direction in the global market, with developer demand for cost-effective top-tier open-source models surging. Adaptation to China's domestic Xinchuang innovation framework is the core moat for domestic service providers.

2. Core customer pain points: For government and enterprise clients, the top priority is data security, with a requirement to keep all data within internal networks, while most lack in-house capability for LLM secondary development. For global developers, there is a widespread lack of cost-effective open-source alternatives that match the performance of leading closed-source models.

3. Actionable solutions: For government and enterprise clients, service providers can adopt a "standard base model + customized private deployment" model, partner with domestic chip makers to deliver integrated on-premise solutions to meet data security requirements, and set up dedicated training systems to support clients' secondary development. This model maintains healthy gross margins, while the project experience gained can be used to iteratively improve the core base model.

This article offers actionable references for AI-related platform operators on operations, business development and risk mitigation:

1. Market demand: Global developers currently have strong demand for accessing and using top-tier open-source LLMs. Adding high-quality leading open-source LLMs to a platform can quickly attract developer traffic and boost platform influence, so platforms should prioritize this type of resource introduction.

2. Business development and operations direction: The most promising LLM projects are not always hot projects that achieve mass consumer traction in the short term. Underappreciated early-stage projects like Zhipu AI, which focused on deep technology and scenario accumulation, can deliver far higher returns. Platforms should not evaluate projects solely based on short-term traffic, and should prioritize technical depth and scenario accumulation when sourcing new projects.

3. Risk mitigation: The LLM industry's pace of technological iteration is extremely fast, so platforms need to watch for the risk of portfolio projects falling behind the technology frontier. Obvious valuation bubbles have already emerged in the industry: the scarcity premium driven by 1,000x price-to-sales ratios and low free float is unsustainable, and valuations are likely to correct after lock-up expirations. Platforms should communicate these risks clearly, guide the market to focus on projects' actual profitability and sustainable business models, and discourage pure concept-driven speculation.

This article provides research materials on new industry trends, open questions and business model innovation for AI industry researchers:

1. New industry trends: China's LLM industry has developed a distinct growth path different from global peers. Unlike OpenAI's pure standardized API model or the pure project outsourcing model used by China's early leading AI firms, Zhipu AI has forged a new path of "standard base model + customized private deployment". It refined its model through B-side scenario accumulation, then used that improved model to drive growth in API services and the general market, creating a new archetype for LLM development in China.

2. Key open research questions: Valuation logic in the current AI industry is highly volatile. The market rejected heavy-asset LLM models early on, then delivered a trillion-dollar valuation to Zhipu in a very short period. The sustainability of the 1,000x price-to-sales valuation bubble and the scarcity premium driven by low free float are all areas worthy of in-depth research. In addition, how companies maintain competitiveness amid extremely rapid technological iteration is an important research direction.

3. Business model innovation: Zhipu's model challenges the traditional assumption that government and enterprise customization necessarily carries high marginal costs and cannot scale. By building on a standardized base model, the company maintains a healthy industry-leading gross margin above 40%, and has validated the technology spillover effect of B-side government and enterprise projects. This opens up a new research direction for LLM business model innovation.