On July 20, 2026, Hong Kong’s large AI model sector was hit by another wave of aggressive sell-offs. By the close of trading, Zhipu (02513.HK) stood at HKD 890.5, marking a single-day plunge of 19.56%. Trading volume reached 11.3789 million shares, with turnover at HKD 11.091 billion. Following the previous Friday’s (July 17) sharp drop of 28.49%, Zhipu’s cumulative decline over the past five trading days has widened to 46.23%. Another major player, MINIMAX (00100.HK), also weakened, closing at HKD 193.1, down 10.6% for the day, with an intraday low of HKD 191 and total turnover of HKD 1.104 billion.
This isn’t an isolated adjustment for these two stocks. Since late June, Hong Kong-listed AI large model companies have undergone a dramatic valuation reset. Zhipu has been sliding since its intraday record high of HKD 2,980 on June 22, retracing more than 70% in just one month and losing over HKD 800 billion in market capitalization. MINIMAX’s drop has been even more severe, plunging over 85% from its annual peak of HKD 1,330, now just a step away from its HKD 165 IPO price. While the Hang Seng Index gained 580 points (up 2.36%) that day, the sharp declines of these two AI giants stood out starkly against the broader market.
How a Trillion-Dollar Market Cap Collapsed in Just One Month
To understand the underlying logic behind this round of sell-offs, it’s essential to review how these two companies’ valuations soared to their peaks—and why they collapsed so quickly.
Zhipu listed on the Hong Kong Stock Exchange on January 8, 2026, at an IPO price of HKD 116.2, with an initial market cap of about HKD 52.8 billion. After listing, its share price climbed steadily, reaching an intraday high of HKD 2,980 on June 22—more than 24 times its IPO price, with market capitalization peaking at HKD 1.33 trillion. MINIMAX similarly surged to HKD 1,330 in mid-March, with a market cap approaching HKD 420 billion.
The key driver behind this valuation surge was the market’s high expectations for the commercialization prospects and technological advancement of domestic large AI models. As the AI narrative guided global capital flows, Hong Kong’s large model stocks became rare "pure AI assets," attracting concentrated investment. However, excessive valuation premiums inevitably face correction when not supported by sustained earnings. Zhipu’s price-to-earnings ratio is negative, with earnings per share at -HKD 11.17; MINIMAX is also loss-making. With profitability a distant prospect, a valuation collapse was only a matter of time.
Why the Unlocking of Restricted Shares Became the Direct Catalyst for Valuation Collapse
Valuation bubbles take time to form, but their burst is often triggered by a clear event. For Zhipu and MINIMAX, that trigger was the unlocking of restricted shares in early July.
Zhipu’s first round of share unlocking occurred on July 8, totaling 25.6816 million shares, or 5.76% of the company’s total equity, mainly held by 11 cornerstone investors. Before this unlocking, the company had only 11.74 million freely tradable shares; the event more than doubled the float. Shortly after, on July 13, Zhipu completed a placement of 19.78 million new H shares, further increasing tradable shares.
MINIMAX faced an even greater impact. On July 9, it unlocked 34.25% to 48% of its shares, corresponding to 107 million to 153 million shares. The unlocking included not only cornerstone investors but also many pre-IPO shareholders. Unlike Zhipu, where nearly 70% of cornerstone investors pledged long-term holding, MINIMAX’s unlocking was dominated by financial investors, naturally creating strong selling pressure. On the first day of unlocking, MINIMAX’s share price plunged nearly 18%, with market cap dropping to about HKD 93.3 billion. The supply-demand imbalance caused by unlocking is the most direct market force driving valuation correction—when the float suddenly expands and buying power is insufficient, prices inevitably seek a new equilibrium.
How Kimi K3’s Sudden Arrival Is Reshaping the Competitive Landscape
If share unlocking was a supply-side shock, then the abrupt shift in industry competition is shaking the long-term investment logic of AI large models.
In the early hours of July 17, Moonshot AI announced its new-generation large model, Kimi K3, boasting 2.8 trillion parameters—the largest open-source model globally. On the authoritative Frontend Code Arena leaderboard, Kimi K3 scored 1,679, surpassing Fable 5 (1,631) and GPT-5.6 Sol (1,618), becoming the first open-source model to outperform all overseas closed-source models on a major programming leaderboard.
This event’s industry significance goes far beyond a single product launch. Kimi K3’s open-source approach, achieving core capabilities superior to closed-source giants, directly challenges the high-priced business models of overseas closed-source models that rely on technological moats. After the announcement, the Philadelphia Semiconductor Index fell 12.5% in a week, and the US AI sector lost about USD 470 billion in market value within 72 hours. The shock quickly spread to Chinese AI peers—on K3’s launch day, Zhipu plunged 28.49% and MINIMAX tumbled 15.62%. Goldman Sachs noted in its July 18 research report that this reflects market concerns about the competitive landscape and long-term winners among Chinese AI models, with leadership and sustainability "still highly volatile."
As technological moats are eroded, the scarcity premium underpinning high valuations becomes unsustainable.
How US Regulatory Upgrades Are Increasing External Uncertainty
Beyond internal competitive shifts, external policy environments are also exerting sustained pressure on the AI sector.
According to sources, the US government has introduced new regulatory measures, strengthening control over the release and authorized use of cutting-edge AI models, gradually reclaiming decision-making autonomy from leading AI companies. The government now leads the list of enterprises and institutions allowed to access top AI models. Previously, OpenAI and Anthropic could independently select clients and control the scope of high-end model access, operating their flagship AI and cybersecurity models via the Daybreak and Project Glasswing initiatives.
Tighter regulation directly increases operational uncertainty for AI companies’ existing projects. At the same time, the US has rolled out supporting regulatory measures, signing relevant executive orders in June, launching the Gold Eagle security program this week, and planning to establish an official clearing center to vet model collaboration partners. For Hong Kong-listed AI companies highly dependent on global tech ecosystems and capital markets, external regulatory uncertainty undoubtedly heightens investor risk aversion.
Cross-Asset Mirroring: The Link Between AI Stock Sell-Offs and Crypto Market Dynamics
It’s worth noting that the sell-off in Hong Kong AI stocks isn’t an isolated event—it mirrors the performance of AI-themed tokens in the crypto market.
Globally, AI-related assets are undergoing a collective correction. Since June, Nvidia’s market cap has shrunk by more than 15%, wiping out USD 900 billion. Google, Amazon, and Meta have retreated by about 6%, 11%, and 12%, respectively. The Nasdaq 100 Index has pulled back nearly 6% over the same period.
In the crypto market, the AI narrative has also cycled from frenzy to cooling. In Q1 2026, AI agent tokens experienced an overall correction of 80% to 90%. Despite the sector’s total market cap rising from about USD 900 million at the start of 2025 to USD 2.2–2.7 billion by May 2026, internal valuation divergence and volatility have become the norm. This synchronized cross-asset correction reflects a systemic shift in global capital’s risk appetite toward AI—whether in traditional equities or crypto, investors are reassessing fair value for AI assets.
Capital Flows Reveal Market Divergence
Despite steep share price declines, capital flow data on July 20 showed signs of divergence.
Zhipu saw net inflows of HKD 4.2589 million in main funds, with super-large orders accounting for HKD 3.9919 million, large orders HKD 267,000, medium orders HKD 433 million, and small orders HKD 199 million. This indicates that while retail investors were panic selling, some institutional funds were stepping in against the trend. Investment banks mostly rated the stock as a "buy," with seven banks issuing buy ratings over the past 90 days, and the 90-day average target price at HKD 1,539.25.
MINIMAX, however, faces more severe selling pressure. The company is expected to see continued reductions, with many financial institutions seeking to exit and realize gains. As cornerstone investors’ "long-termism" clashes with financial capital’s "exit demands," whether this reshuffling of holdings will mark a turning point for valuation remains to be seen.
Where Does the Market Go After the Valuation Bubble Clears?
With Zhipu below HKD 890 and MINIMAX below HKD 200, the valuation bubble in Hong Kong’s AI large model sector is rapidly clearing. But is this the end of the correction, or the start of a new value reset?
Fundamentally, both companies remain loss-making, and high valuations will continue to face pressure without earnings support. From a holdings perspective, Zhipu will see a much larger unlocking event in January 2027, when about 40% of original shares will become tradable, meaning long-term selling pressure persists. Supply-side pressure hasn’t been fully released yet.
Industry-wise, technological iteration in large models is accelerating, and the competitive landscape is far from settled. MiniMax is developing a 2.7 trillion-parameter large language model, tentatively named M3 Pro, which could launch as early as Q3 this year and is planned to be open-sourced. Whoever successfully commercializes large models first will take the lead in the next round of valuation restructuring. As the AI industry’s pricing logic shifts from chasing model capabilities to validating commercial returns, the end of this valuation reset may depend on which companies can find a path to self-sustaining growth amid the cash burn race.
Summary
As of the close on July 20, 2026, Zhipu stood at HKD 890.5, down 19.56% for the day; MINIMAX closed at HKD 193.1, down 10.6%. This wave of sell-offs resulted from multiple factors converging: the supply shock from unlocking restricted shares, the competitive reset triggered by Kimi K3’s launch, increased external uncertainty from US regulatory upgrades, and systemic pressure from the global AI asset correction. Essentially, this is a long-overdue valuation reset—when market sentiment shifts from frenzy to rationality, excessive valuation premiums inevitably clear out. For investors, understanding the drivers behind valuation restructuring is far more valuable than chasing short-term price swings.
FAQ
Q: What were the closing prices for Zhipu and MINIMAX on July 20?
A: As of the close on July 20, 2026, Zhipu stood at HKD 890.5, down 19.56% for the day; MINIMAX closed at HKD 193.1, down 10.6%.
Q: How much has Zhipu retreated from its historical high?
A: Zhipu hit an intraday record of HKD 2,980 on June 22. By July 20, it closed at HKD 890.5, marking a one-month pullback of over 70% and a loss of more than HKD 800 billion in market capitalization.
Q: How far is MINIMAX’s current price from its IPO price?
A: MINIMAX’s IPO price was HKD 165. It reached a historical high of HKD 1,330 earlier this year. On July 20, it closed at HKD 193.1, only about HKD 28 above its IPO price.
Q: What impact did the launch of Kimi K3 have on Hong Kong’s AI sector?
A: On July 17, Moonshot AI launched Kimi K3, a model with 2.8 trillion parameters—the largest open-source model globally. The model outperformed overseas closed-source models on programming leaderboards, prompting a revaluation of the competitive landscape for domestic large models. On the day of K3’s launch, Zhipu plunged 28.49% and MINIMAX tumbled 15.62%.
Q: Has the valuation adjustment for the AI large model sector ended?
A: From a holdings perspective, Zhipu will see a much larger unlocking event in January 2027 (about 40% of original shares), so supply-side pressure hasn’t fully released. Whether valuations have bottomed depends on the competitive landscape, commercialization progress, and when global capital’s risk appetite for AI assets recovers.
