Why did Zhipu’s stock price plunge? Domestic large-model competition, financing expansion, and commercialization pressures are changing market expectations

On July 17, the Hong Kong AI sector continued to see pullbacks. Zhipu (2513.HK) saw a sharp intraday drop, at one point nearing 15%, trading at HK$1,316.00. The Hang Seng Tech Index opened slightly down 0.05%, and AI application stocks were generally under pressure. Since Zhipu’s intraday peak of HK$2,222 on July 9, its largest cumulative drawdown over six trading days has exceeded 40%. If measured from its historical high of HK$2,980, the cumulative drawdown is even more than 55%.

This round of decline is not an isolated event. Since July, Zhipu has undergone cornerstone unlocks, large placements, and sharp share-price volatility in succession—on July 2 it fell more than 16% intraday, on July 8 it surged nearly 15%, and after the placement announcement on July 9, it rose more than 20% at its intraday high. Multiple days of more than 10% gains or losses in a single month indicate that pricing disagreements for this AI blue-chip are rapidly widening.

The early rally was too fast, and valuation overhang triggered profit-taking

Since Zhipu’s listing in January 2026 at HK$116.2, the stock price reached a historical high of HK$2,980 within less than half a year, with a cumulative gain of more than 24 times. Behind this surge is concentrated market pricing of policy support for domestically developed large-model industries, expectations for AI agent commercialization, and growth in enterprise AI application demand.

However, after the rapid rise, the gap between valuation and fundamentals has started to be re-examined by the market. Zhipu’s current P/E ratio is -152.80, reflecting that the company is still in an loss-expansion stage. On July 10, Goldman Sachs initiated coverage of Zhipu with a “Neutral” rating, and its 12-month target price based on a cash-flow discount model is HK$1,880. This target price is clearly discounted versus the historical high, indicating that even in an optimistic scenario, the investment bank believes the prior valuation level already fully—or even excessively—reflected growth expectations.

When the market moves from the “storytelling phase” to the “results phase,” AI company valuations typically undergo repricing. Zhipu’s prior rally thesis was built on long-term narratives such as domestic substitution, technological breakthroughs, and market share expansion. But these narratives need to be converted into verifiable financial data to justify elevated valuations. The extreme volatility since July, at its core, reflects the market’s contest over this conversion process.

HK$31.4 billion placement: a double-edged sword of financing expansion and equity dilution

Financing is another key variable affecting Zhipu’s stock price. On July 9, Zhipu announced a placement of up to 19.78 million new H shares at HK$1,588 per share, raising an estimated HK$31.41 billion. The placing price was at a discount of about 12.99% versus the HK$1,825 closing price on July 8. The placing shares represent about 4.25% of issued shares after the enlargement. This scale set a new high for single-instance placement fundraising by Hong Kong tech companies in 2026.

The financing backdrop is that the pace of capital consumption is far faster than expected. As of June 30, 2026, Zhipu had used more than 93% of the net proceeds from its IPO. Capital expenditures by large-model companies are continuing to accelerate—the purchase of computing power, construction of GPU clusters, model training, and talent R&D all require large upfront investments.

In terms of the use of funds, the net proceeds of about HK$31.4B will be allocated to three main directions: R&D of foundational large models and computing-power infrastructure; commercialization expansion and strategic investments; and optimizing the capital structure and replenishing working capital. The company expects all proceeds to be fully used by the end of 2027.

Financing itself is not necessarily a negative, but market concerns are being amplified on two fronts: first, the short-term dilution effect from new share issuance—earnings per share are diluted and short-term supply increases; second, the AI business model still needs ongoing validation—answers to how much will be invested, how much revenue will be generated, and when profitability will be achieved remain unclear.

One noteworthy detail is that this placement was completed only about a week after Zhipu’s cornerstone unlock. Earlier in July, the company reached 6 months post-listing, and 25.68 million shares of cornerstone restricted stock were due to be released for trading. After the unlock, large-scale placement was launched immediately, sparking discussions in the market about “unlock equals financing.” Although six institutions took up all placing shares, the increase in near-term share supply still created downward pressure on the stock price. On July 14, only one trading day after the placement was formally completed, Zhipu’s intraday low fell to HK$1,473, already below the placing price of HK$1,588.

The large-model AI competition has entered the “elimination round” stage

Another pressure from the repricing comes from shifts in the competitive landscape. China’s large-model AI race is moving from a “technology contest” to a “commercialization contest.” Zhipu’s competitors include Alibaba’s Tongyi Qianwen, Baidu’s Ernie, ByteDance’s Doubao, DeepSeek, MiniMax, and others.

The core variables of competition are changing. Previously, model parameter scale and Benchmark performance were the key metrics the market focused on; now, commercialization capability is becoming the new focal point. Goldman Sachs divides the development of China’s large models into two stages: 2025 is the “DeepSeek moment”—the industry relies on the MoE architecture to sharply reduce inference costs, with the core advantage in cost efficiency; 2026 is the “Zhipu moment”—GLM 5.2 entering the global leading first tier in high-value tracks such as code and agents, demonstrating that domestic model performance can compete globally. After technological breakthroughs, how to convert technological advantages into sustainable commercial revenue is the key to determining long-term value.

Meanwhile, large-model companies’ C-end competition is moving from pure traffic contests into a value validation phase. After user scale grows, how to enhance willingness to pay, improve business models, and prove value becomes the industry’s next core question. 2026 is viewed within the industry as the “deep-water zone” for large-model scaling into mass deployment.

The same trend can be seen in valuation logic in the primary market. In 2026, the core metrics for institutional due diligence have shifted from parameter scale and leaderboard rankings to monthly operating revenue, computing-power cost amortization, and paid-user conversion rates. Projects unable to provide a clear commercialization path are increasingly unlikely to obtain new-round financing.

Valuation logic is shifting from “technology leadership” to “revenue verification”

Zhipu’s future valuation reshaping depends on answers to three core questions.

First, can the AI model generate sustainable revenue? Based on existing data, Zhipu’s commercialization progress shows positive signals. In Q1 2026, its API prices were cumulatively raised by 83%, yet the number of calls increased by 400%. The MaaS platform ARR reached RMB 1.7 billion. Registered enterprises and users exceeded 4 million, and services cover more than 218 countries and regions worldwide. But Goldman Sachs forecasts that China’s AI model API and subscription revenue will grow from RMB 35 billion in 2026 to RMB 879 billion by 2030—this scale of growth implies Zhipu needs to continuously expand market share amid intense competition.

Second, can computing-power costs effectively decline? Large-model companies face the core race between the speed of revenue growth and the speed of computing-power cost growth. GPUs, servers, and cloud resources make up the largest fixed-cost items for large-model companies. JPMorgan raised Zhipu’s target price from HK$2,000 to HK$2,400, believing that the HK$31.38B fundraise will effectively ease the company’s computing-power supply bottleneck. But the payback cycle and return on investment for computing-power spending remain key areas of market attention.

Third, will an ecosystem moat form? In the future, competition is not just about model capabilities, but rather a comprehensive contest of “model + agent + enterprise applications + developer ecosystem.” In an internal letter, Zhipu CEO Tang Jie clearly stated, “Amid the industry’s acceleration of monetization, we have decided to break through upward,” committing strategic investment to R&D of the next-generation AI model. Whether this long-term strategy can gain capital-market recognition in the short term remains highly uncertain.

Market signals: the AI rally isn’t over, but the pricing rules have changed

The sharp adjustment in Zhipu’s stock price sends a signal not that AI demand has disappeared, but that the capital market is now requiring AI companies to prove commercialization capability.

From a more macro perspective, in Q2 2026, global AI concept stocks are broadly under pressure. On July 17, affected by global “AI fatigue” sentiment, South Korean stocks fell 6.4% and Japanese stocks fell 2.8%. In the US stock market, renewed market concerns about AI spending were triggered again by TSMC’s results, leading to a significant selloff in tech stocks. Zhipu’s adjustment is part of the global AI sector’s valuation reconfiguration, not an isolated case.

However, structural support for China’s AI market still exists. Goldman Sachs launched a tradable “China AI value chain” basket and predicts that China’s AI model API and subscription revenue will grow by about 25 times over the next five years. An OpenRouter report shows that the token consumption of China’s AI models such as DeepSeek and Zhipu has rapidly surpassed US competitors. The long-term logic—growth in enterprise AI demand, ongoing development of domestic AI infrastructure, and expanded application space for large models—has not changed.

Zhipu’s future depends on one core variable: whether technological leadership can be converted into sustainable commercial leadership. The capital market is casting votes with prices—not denying AI’s value, but demanding a clearer commercialization path.

FAQ

Q: What is the direct reason for Zhipu’s stock price drop on July 17 (2513.HK)?

That day, Zhipu fell by nearly 15% intraday, to HK$1,316. The direct triggers include profit-taking after the stock’s early rally had been too large, short-term supply pressure caused by the HK$31.4B placement, and a broader market environment of weakness across the global AI sector. On the same day, the company completed an English abbreviation change to “Z.AI,” but it did not provide support for the stock price.

Q: Is the HK$31.4B placement a positive or negative for Zhipu’s stock price?

The financing itself provides funding for R&D and computing-power expansion, which is necessary in the long run. But in the short term, the market focuses on equity dilution (about 4.25%) and the placing price discount (about 13%), and the placement followed immediately after the cornerstone unlock, raising concerns about share supply. On July 14, the stock price had already fallen below the placing price of HK$1,588.

Q: Why did Zhipu and MiniMax show diverging stock performance?

After Zhipu’s unlock of 5.76% of shares in early July, its stock surged 13.66%, while after MiniMax’s unlock of 48.9% of shares, it plunged 17.98%. The difference stems from variations in unlock proportions, financing structure, and differing market expectations for the two companies’ commercialization prospects. JPMorgan raised Zhipu’s target price to HK$2,400, while lowering MiniMax’s target price to HK$240.

Q: What changes are happening in the competitive landscape of China’s large-model AI industry?

Competition is shifting from a “model capability contest” to a “commercialization capability contest.” Valuation logic in the primary market has moved from parameter scale and leaderboard rankings to monthly operating revenue, computing-power cost amortization, and paid-user conversion rates. Projects that cannot provide a clear commercialization path are unlikely to obtain financing.

Q: How should investors judge Zhipu’s long-term investment value?

It depends on three core questions: whether AI models can generate sustainable revenue, whether computing-power costs can effectively decline, and whether an ecosystem moat can form. Zhipu’s 2026 Q1 API saw a cumulative price increase of 83% while call volume grew 400%, indicating some pricing power. But Goldman Sachs gave it a “Neutral” rating, and there remains disagreement in the market about valuation versus the earnings outlook.