Hong Kong-listed AI large model stocks slide across the board: MiniMax and Zhipu plunge hard—has the AI sector entered a knockout stage?

The narrative around AI in the capital markets is undergoing a subtle yet profound shift.

As of July 22, 2026 (Beijing time), Hong Kong-listed AI large-model stock names saw clear sell pressure. MiniMax-W (HKG: 0100) closed at HK$197.00, down 11.42% for the day; Zhipu (HKG: 2513) closed at HK$1,175.00, down 3.61%. Over the same period, the Hang Seng TECH Index fell by about 1.8%, and the AI concept sector broadly underperformed the broader market.

This trend is not an isolated case. Since the second quarter of 2026, the Hong Kong AI sector has repeatedly seen adjustments like this. The market can’t help but ask: Is this merely normal profit-taking after an overly strong prior run-up, or is the valuation logic of the AI track undergoing a structural change?

Starting from short-term market performance, this article will analyze the pressure test facing MiniMax and Zhipu by combining the evolution of the global AI competitive landscape and changes in the valuation frameworks for large-model companies, and will also explore the next stop in AI investment logic.

Short-term pullback: profit-taking and valuation digestion go hand in hand

From a trading perspective, this pullback clearly has short-term triggers.

Since the first quarter of 2026, MiniMax’s cumulative gain at one point exceeded 80%, and Zhipu’s gain over the same period was also above 45%. In the absence of short-term performance catalysts, some capital chose to lock in profits. The drop on July 22 came with an increase in trading volume; MiniMax’s trading value on the day was about twice the average of the prior five trading days, indicating that sell orders were relatively concentrated.

Global risk appetite contracted in tandem. With the Federal Reserve’s July policy meeting approaching, uncertainty around the pace of rate cuts has increased; the U.S. Dollar Index has recently strengthened, and technology sectors in emerging markets have faced overall pressure. As an offshore market, Hong Kong equities are more sensitive to expectations for global liquidity, and high-beta AI concept stocks are often hit first.

These factors can explain the stock-price volatility in the short term, but they are not enough to account for the clear differentiation within the sector—MiniMax’s decline far outpaced Zhipu and the broader market. This suggests the market is not indiscriminately dumping AI assets, but rather reflects stronger company-specific factors or differentiated pricing.

The deeper question is: in today’s AI companies’ stock prices, how much is being priced in for a “future technological lead,” and how much is being priced in for “current commercial monetization”? Over the past two years, market valuations for AI large-model companies have been built more on narratives of technical capability and parameter scale. When the narrative enters the realization period, investors begin demanding to see more “traditional” financial metrics such as revenue, customers, and profit margins—loosening valuation logic is inevitable.

Competitive landscape: from “a technical sprint” to “a multidimensional war”

The competitive posture in the AI large-model industry has clearly escalated.

In the past two years, the market’s mainstream focus has been on “whose model has more parameters” and “whose benchmark rankings are higher.” Technical breakthroughs alone are enough to drive a round of valuation re-rating. But as 2026 arrives, this single-dimension competition is being replaced by multidimensional commercial warfare.

From a global perspective, OpenAI, Google Gemini, Anthropic, and Meta AI form the first tier. Their shared characteristics are: a parent company or financing capability on the scale of hundreds of billions of dollars, ecosystem entry points covering hundreds of millions to tens of billions of users, and large-scale self-built compute cluster groups. These elements create a “flywheel effect” for ongoing model iteration—more users generate more feedback, more feedback optimizes the models, and improved models attract more users.

In the Chinese market, competition is equally intense. Alibaba’s Tongyi Qianwen relies on Alibaba Cloud’s enterprise customer network and e-commerce scenarios; Tencent’s Hunyuan is embedded in the WeChat and gaming ecosystem; Baidu’s Ernie has a search entry point and years of AI accumulation; and ByteDance’s Doubao leverages Douyin’s traffic advantages to scale up quickly. What these large enterprises share is that model capability is not their only competitive moat, and may even not be their most important moat.

For standalone large-model companies like MiniMax and Zhipu, competition has shifted from “can we build a great model?” to “can we establish a sustainable commercial closed loop in the gap between the giants?”

Valuation logic switching: the market is redefining “moats”

The valuation framework for large-model companies is undergoing a three-stage transition.

The first stage is driven by technological leadership. The capital market is willing to pay a high premium for metrics such as parameter scale, benchmark scores, and paper publications. In this stage, a company’s value is mainly determined by the “potential” of its technical capabilities.

The second stage is driven by commercialization validation. The market begins to focus on indicators such as the growth rate of API calls, the number of enterprise paying customers, changes in average revenue per customer, and revenue structure. In this stage, investors care more about whether “technological advantages can be converted into revenue advantages.”

The third stage is driven by ecosystem and cost. As base-model capabilities gradually converge, the core of competition will shift to two points: first, whoever has lower unit inference costs gains pricing initiative, thereby expanding the customer base and forming a data flywheel; second, whoever has more application scenarios and user touchpoints will be able to build a positive loop of data and traffic.

The market is currently at a critical node transitioning from the first stage to the second stage. Companies that can continuously disclose commercialization progress and show a clear path to revenue will receive valuation support; conversely, companies that are still stuck in the “technology narrative” stage will face ongoing valuation contraction pressure.

Judging from the different drawdowns between MiniMax and Zhipu, the market may already be executing this screening logic—investors are pricing the commercialization prospects of different companies differently.

Decomposing competitive pressure on independent AI vendors

MiniMax’s core advantage lies in its multimodal capabilities and its initial overseas market layout. But its challenges are equally concrete: at the C-end application layer, it needs to compete for user time with products that have huge traffic entry points, such as ByteDance’s Doubao and Baidu’s Ernie; in the B-end market, it needs to compete with cloud vendors like Alibaba and Tencent for enterprise budgets with bundled “model + cloud service” solutions; and on the cost side, independent vendors are at a natural disadvantage in GPU procurement scale and bargaining power, which directly affects their competitiveness in API pricing.

Zhipu’s technical accumulation and open-source ecosystem are important moats, especially where it has a first-mover advantage in the enterprise services market. However, its main pressures come from: competition with the cloud ecosystems of internet giants—enterprise customers often prioritize AI solutions tied to the cloud service providers they already use; commercialization speed—compared with big players that can allocate AI costs across multiple business lines, independent vendors need every dollar of R&D investment to be returned by the AI business itself; and in a context where base-model capabilities trend toward homogenization, how to build differentiated technical labels.

The shared pressure for both companies is: with high ongoing base-model investment, can revenue growth cover costs—or at least allow the market to see a clear path to cost coverage?

Conclusion

This pullback in Hong Kong AI large-model stocks is more likely the beginning of a structural valuation reshaping than the end of the AI trend.

With competition among global and China-based AI companies—such as OpenAI, Google, Alibaba, Tencent—advancing in full swing, the core competitive strengths of large-model companies are shifting from single “technological leadership” to multidimensional comprehensive competition: “model performance + cost efficiency + commercialization capability + ecosystem scale.”

For investors, this means the opportunity window for AI investments is narrowing—from “industry beta” to “individual stock alpha.” The future winners will not be every model company, but those that can be the first to prove that revenue growth is sustainable, cost structures can be optimized, and customer relationships can be deeply retained.

AI technology itself is still evolving rapidly, but capital markets’ patience is shrinking. When the tide goes out, the difference between those swimming naked and true long-distance swimmers will become clearer than ever.

FAQ

1. What are the main reasons for the stock price declines of MiniMax and Zhipu?

In the short term, after AI concept stocks surged significantly, the market faces profit-taking pressure, and global risk appetite has cooled somewhat due to uncertainty around Federal Reserve policy. But more fundamentally, the market’s valuation logic for AI companies is shifting from the “technological leadership narrative” to “commercialization validation.” Investors have started to assess each company’s revenue growth, customer acquisition, and cost control abilities more cautiously; companies without a clear path to profitability face valuation discounts.

2. What main competitors do China’s AI large-model companies face?

In international markets, major competitors include OpenAI, Google Gemini, Anthropic, and Meta AI—these companies are backed by massive capital and global user ecosystems. In China’s market, the competitive landscape is also fierce, with main rivals including Alibaba Tongyi Qianwen, Tencent Hunyuan, Baidu Ernie, and ByteDance’s Doubao. These internet giants not only have strong R&D capabilities, but also rich application scenarios and large user bases, enabling ecosystem synergy advantages beyond the models themselves.

3. What changes have occurred in the valuation logic of large-model companies?

Previously, valuations mainly anchored on “potential indicators” such as model parameter scale and technology leaderboard rankings. Now the market is gradually shifting toward requiring companies to demonstrate quantifiable commercialization outcomes, including API call revenues, the number of enterprise paying customers, and revenue from deployed AI applications. Meanwhile, compute cost control capability (including GPU investments and inference costs) and the breadth of ecosystem applications have also become important valuation dimensions. Investors are paying more attention to the sustainability of profitability rather than simply a technical vision.

4. Does the adjustment in Hong Kong’s AI sector mean the end of AI investment opportunities?

No. It does not mean the end of the AI trend; it may instead signal a switch in investment logic from “track-based opportunities” to “individual stock selection opportunities.” In the future, the market will be more inclined to reward companies with clear revenue growth paths, cost advantages, and ecosystem moats, while targets lacking fundamental support may face sustained pressure. For investors, the opportunity window shifts from going long the entire industry to carefully choosing companies with sustained competitive strength.

5. What core challenges does commercialization face for AI large-model companies?

The primary challenge is whether high R&D and compute costs can be covered by revenue growth, including GPU cluster investment, data training costs, and model inference costs. Next comes customer acquisition and retention—independent AI vendors need to compete with internet giants that already have established enterprise customer relationships. Furthermore, as base-model capabilities gradually converge, each vendor needs to find differentiated application scenarios and industry solutions to avoid getting trapped in pure price competition. In addition, losing API pricing power and lacking an ecosystem are also risks that independent vendors need to watch out for.