China’s large-model industry is shifting from a fundraising race into a capital-market pricing competition.

On May 29, MiniMax officially entered A-share IPO advisory, with CITIC Securities acting as sponsor. Earlier, Zhipu AI had already appointed Guotai Junan Securities and China International Capital Corporation (CICC) to advance its listing preparation on the Shanghai Stock Exchange. The back-to-back moves by two leading foundation model companies signal a structural change: China’s core AI assets are no longer confined to early-stage venture capital, Hong Kong listings, or narrative-driven trading. They are now entering the domestic A-share pricing system.

This is not simply a case of “two more AI companies preparing to list.” It marks a shift in how China’s AI industry is being defined by capital markets. A-shares have historically priced peripheral AI exposure—compute, semiconductors, servers, optical modules, application software. The next phase will involve direct pricing of foundation model companies themselves: revenue quality, user growth, compute burn, monetization capacity, and valuation logic tied to core model platforms.

Why MiniMax and Zhipu?

The overlap is clear. Both are not narrow tool providers. They sit closer to “model foundation + application gateway” structures within China’s large-model ecosystem.

MiniMax is more exposed to consumer applications and AI-native products: Hailuo AI, Talkie, voice systems, multimodal capabilities, and agent-based applications. This positioning makes it easier for capital markets to interpret it as a next-generation AI application platform, closer to end users and more consumer-facing.

Zhipu AI, by contrast, is positioned as a domestic foundation model stack aligned with enterprise services, government clients, and industrial applications. It fits a different logic: under the constraints of localization, enterprise digitization, industrial AI adoption, and AI safety requirements, it aims to become enterprise-grade AI infrastructure in China.

If MiniMax represents the “consumerization of AI traffic,” Zhipu represents the “industrialization of AI infrastructure.” The market’s dual-labeling of them as “twin leaders” reflects two parallel commercialization paths rather than direct substitution.

Why List in A-shares?

Three forces are at work.

First, China’s RMB capital market needs core AI assets.
Over the past two years, A-share AI trades have been dominated by the compute supply chain: Nvidia-linked proxies, optical modules, servers, liquid cooling, PCBs, and data centers. These are essentially “AI shovels.” The real long-term value drivers—models, user gateways, application monetization, and ecosystem control—have been absent from domestic listings.

MiniMax and Zhipu’s A-share pipeline marks a shift from pricing AI infrastructure inputs to pricing AI platform outputs.

Second, foundation models require continuous capital intensity.
Large models are not asset-light internet businesses. Training, inference, data pipelines, engineering systems, safety compliance, and application ecosystems all require sustained capital. As domestic models enter price competition and freemium expansion, usage growth directly translates into compute burn.

Listing is not an endpoint. It is a financing mechanism and balance-sheet expansion tool.

Third, A+H structures enable dual capital ecosystems.
Hong Kong provides international capital access and growth re-rating; A-shares provide policy alignment, domestic institutional demand, and industrial narrative support. For foundation model companies, dual listings expand not only funding channels but also valuation anchors and pricing influence.

Who Is Closer to the “Compute and Application No.1 Stock”?

If the metric is “application leadership,” MiniMax carries stronger upside optionality. Its product velocity, user interaction intensity, and consumer-facing applications create more asymmetric valuation outcomes. AI companionship, video generation, voice systems, and agents can produce rapid sentiment-driven re-rating if breakout products emerge. The market would likely treat it as a native AI platform company.

But the constraint is clear: user growth does not automatically translate into profit. High-frequency AI usage often implies high inference cost. If users remain price-sensitive, scale can amplify losses rather than earnings. The key test is whether MiniMax can establish sustainable monetization through subscriptions, enterprise services, APIs, content ecosystems, or agent-based transactions.

If the benchmark is “compute and industrial AI infrastructure,” Zhipu holds stronger policy and enterprise depth. Its alignment with government clients, state-owned enterprises, and regulated industries provides structural demand. In A-share logic, “domestic substitution + strategic technology + policy support” carries clear valuation premiums.

The limitation is commercialization latency. Enterprise AI cycles are longer, revenue recognition is slower, and project-based income introduces variability in margins and cash collection. Markets will eventually shift from model narratives to financial durability: repeat purchase behavior, unit economics, and scalability.

The real competition is not model strength. It is translation of model capability into financial language that capital markets can price.

Three Questions A-shares Will Eventually Ask

First, what is the revenue structure?
Is it one-off project revenue or recurring subscription income? Integration-heavy low-margin work or high-margin API/platform revenue? Subsidy-driven user acquisition or genuine willingness to pay?

Second, can compute costs be absorbed by the business model?
The key risk is simultaneous revenue expansion and loss expansion. The winner will be the company that can compress inference costs and achieve positive unit economics at scale.

Third, can applications create defensible moats?
Model capabilities iterate quickly. Parameter scale alone is not durable. Defensibility comes from data loops, embedded use cases, workflow integration, domain expertise, and ecosystem control points.

This is where MiniMax and Zhipu diverge most sharply: one must prove it is not a transient breakout application; the other must prove it is not a project-based service provider, but a scalable AI infrastructure layer.

What This Means for China’s AI Equity Market

The listing of MiniMax and Zhipu would reshape the structure of China’s AI equity trade.

So far, the dominant exposure has been upstream infrastructure—chips, optics, servers, and data centers. The next phase introduces direct pricing of foundation model operators themselves: who owns the model layer, who owns user distribution, and who captures application revenue.

Two structural effects follow.

On one side, valuation expands from hardware certainty to application growth optionality. If foundation model companies successfully list, China’s AI market will move from an Nvidia-linked proxy trade toward a domestic AI pricing system.

On the other side, dispersion increases. Pure narrative-driven AI firms without revenue, customers, or product closure will face valuation compression as capital concentrates in core model assets.

For A-shares, this is not noise. It is normalization. Capital markets ultimately require assets that reflect the direction of industrial transformation.

Conclusion: The First Stock Is Not a Label. It Is a Pricing Regime

The contest between MiniMax and Zhipu for the “compute and application No.1 stock” title cannot be resolved in simple terms.

MiniMax leads in consumer imagination and user access. Zhipu leads in industrial depth and policy certainty. One resembles a platform candidate for AI applications; the other resembles a candidate for domestic AI infrastructure.

But the more important shift is not listing order. It is that China’s foundation model companies are now entering hard market constraints: financial transparency, cost visibility, growth accountability, and valuation discipline.

From this point forward, model companies can no longer rely on parameter scale or product demonstrations. They must answer a basic capital-market question:

Can you make money?
Can you keep making money?
Can you, after burning massive compute resources, build durable application and industrial value?

The first company that convincingly answers these questions will define what “China’s first foundation model stock” actually means.