The AI sector is currently a fiercely competitive landscape with numerous players. Investors in large language models find it difficult to place all their bets on a single company. Before a clear winner emerges, they tend to spread their investments across several leading model firms.
On July 17, 2026, Hong Kong's AI sector experienced a 'Black Friday.' The Hang Seng Tech Index plummeted sharply, with its decline widening to 5%, closing at 4592.58 points. AI large model concept stocks were hit hard, with KNOWLEDGE ATLAS (HKEX: 02513) and MiniMax, known as the "Twin Titans of Large Models," leading the losses.
By the close, KNOWLEDGE ATLAS shares had plunged 28.49% to HK$1107, wiping out approximately HK$200 billion in market value in a single day. Its latest total market capitalization fell back to around HK$515.4 billion. MiniMax dropped 15.63% on the same day, closing at HK$216 per share.
It's worth noting that on June 22 this year, KNOWLEDGE ATLAS stock price hit a historical high, reaching an intraday peak of HK$2980 and closing at HK$2410, achieving a market cap of HK$1.07 trillion and a year-to-date surge of nearly 20 times. This means that in less than a month, the stock price has fallen by 54%, erasing over 550 billion in market value.
Just six trading days prior, KNOWLEDGE ATLAS had completed a HK$31.4 billion share placement. Institutions that participated in that placement are now sitting on paper losses of 30%.
On July 9, KNOWLEDGE ATLAS announced a placement plan to issue 19.78 million new H shares at HK$1588 per share. If fully placed, this would raise HK$31.411 billion from the capital market, more than six times the HK$5 billion raised in its January IPO. The placement price represented a discount of approximately 13% to the closing price on July 8.
The market reaction was initially enthusiastic. On the day of the placement announcement, KNOWLEDGE ATLAS shares surged over 20% intraday, finally closing at HK$2032. Participating institutions saw paper gains of nearly 28% that day.
However, the specific list of institutions involved in this placement was not publicly disclosed. Public information indicated that CICC was the sole placing agent, and the placement attracted participation from various institutions including international long-term funds and sovereign wealth funds.
On July 14, KNOWLEDGE ATLAS stock price broke below the placement price of HK$1588 intraday. By July 17, it closed at HK$1107, marking a staggering 39% decline over just seven trading sessions. Based on the placement price of HK$1588, participating institutions are facing a 30% paper loss, making them significant contributors to the losses in this downtrend.
Reasons for the Rapid Fundraising
Why was KNOWLEDGE ATLAS in such a hurry for such a large-scale fundraising? Data shows that after its January IPO and the full exercise of the over-allotment option, the company raised approximately HK$4.896 billion in net proceeds. However, by June 30, 2026, less than six months after listing, HK$4.588 billion of this had already been used, leaving only HK$308 million on the books. In other words, 95% of the IPO funds have been spent. Over 60% of the newly raised HK$31.4 billion from the placement is earmarked for continued investment in proprietary large model training and computing power clusters.
Explaining the Sharp Price Correction
The market had expressed concerns about potential stock price volatility ahead of KNOWLEDGE ATLAS's lock-up expiration. However, on the actual expiration date, the stock closed up 13.35%, turning those worries into a false alarm.
On July 8, the first batch of restricted shares for KNOWLEDGE ATLAS became tradable. On that day, approximately 25.68 million shares held by 11 cornerstone investors were released, representing about 5.76% of the total share capital, with a market value exceeding HK$40 billion. Despite this supply, the stock defied the trend, opening nearly 3% lower before quickly reversing to close up 13.35%, adding over HK$100 billion in market value. Nearly 70% of cornerstone shareholders had expressed their intention to hold long-term before the expiration, including several state-backed institutions like the Beijing Artificial Intelligence Industry Investment Fund and the Beijing Jingneng Green Energy M&A Investment Fund. On the same day, J.P. Morgan raised its target price for KNOWLEDGE ATLAS from HK$1800 to HK$2000.
Wang Yajun, Head of Equity Capital Markets for Asia ex-Japan at Goldman Sachs, recently commented, "AI is a high-risk industry, and fluctuations in stock prices and market caps are normal. Without tolerance for volatility, it's difficult to cultivate outstanding enterprises. A market must tolerate companies going 'from 0 to 1' and also 'from 1 to 0' to foster good companies." Regarding whether lock-up expirations would affect the overall trend of the Hong Kong market, Wang analyzed that from the perspectives of fundraising amount and total market cap proportion, such expirations would not alter the market's structural upward trend, stating, "Hong Kong is an international market; a single company's lock-up expiration is insufficient to change the overall direction."
However, the strong performance on the expiration day did not last. The real crisis stemmed from the dual supply shock of the lock-up expiration and the share placement.
On July 17, the same day KNOWLEDGE ATLAS fell 28.49%, a piece of positive news emerged: by July 2026, KNOWLEDGE ATLAS's Annual Recurring Revenue (ARR) had reached $1 billion, achieving its full-year target ahead of schedule. From January to July this year, its ARR grew 15-fold. It took the company only five months to grow from $100 million to $1 billion in ARR, faster than Anthropic's 15 months. Furthermore, the current ARR of KNOWLEDGE ATLAS alone is equivalent to the estimated combined ARR of three leading domestic model firms: DeepSeek, Kimi, and MiniMax.
Despite these impressive commercialization results, why has the stock price continued to fall recently? Market analysis points to three main reasons.
First, the lock-up expiration increased the free float of KNOWLEDGE ATLAS from 3.9% to 9.7%, an expansion of about 2.5 times. The placement added another 19.78 million shares. This dual supply shock coincided with a "global AI tech sell-off panic." On July 17, Japan's Nikkei 225 and Taiwan's Weighted Index plunged over 4% at one point, with Japanese memory giant Kioxia falling 16% and TSMC down nearly 3%. The sudden tightening of external liquidity exacerbated the internal selling pressure on KNOWLEDGE ATLAS, rapidly exhausting market buying power.
Furthermore, the market has begun to doubt the sustainability of growth fueled by heavy spending. The fact that 95% of the HK$4.896 billion IPO proceeds were burned through in less than six months indicates that ARR growth comes at the high cost of computing power investment.
Finally, the competitive landscape is undergoing profound changes. DeepSeek's pre-money valuation is $71 billion with an ARR nearing $500 million. Kimi's ARR surged from $100 million to $200 million in four months, and its K3 model with 2.5 trillion parameters was released in July. Market observers note that KNOWLEDGE ATLAS has a higher valuation than DeepSeek, a less compelling growth story than Kimi, and more expensive API pricing. The scarcity premium of being the "world's first listed large model company" is being diluted by its peers.
Commenting on the competitive dynamics of the large model industry, Wang Yajun stated, "Currently, large model companies are numerous and fiercely competitive. However, ultimately, the large model field will see a winner-take-all outcome. For users, the switching cost from Model A to Model B is low. Considering industry development patterns and user behavior, it's difficult for investors in large models to bet everything on a single model company. Before a winner is determined, investors will place bets on several leading companies simultaneously."
Divergence Between KNOWLEDGE ATLAS and MiniMax
KNOWLEDGE ATLAS and MiniMax listed on the Hong Kong exchange almost simultaneously, operate in the same AI large model sector, and are both in a loss-making state. Yet, their market capitalizations differ by a factor of 6.5.
According to 2025 financial reports, KNOWLEDGE ATLAS reported annual revenue of 724 million yuan, a year-on-year increase of 131.9%. MiniMax reported revenue of 569 million yuan, up 158.9% year-on-year. Their revenue sizes differ by only 155 million yuan. However, as of July 17, KNOWLEDGE ATLAS has a market cap of approximately HK$515.4 billion, while MiniMax is below HK$80 billion.
The difference in post-lock-up performance was a key turning point. The expired shares for KNOWLEDGE ATLAS represented only 5.8% of total shares, and nearly 70% of its cornerstone investors were state-backed with strong holding intentions. In contrast, MiniMax saw 146 million shares become tradable on July 9, representing 46.4% of its total shares, increasing its free float from 5.4% to 51.9%, with a higher proportion of financial investors. Since hitting its historical high on March 19, MiniMax's stock price has accumulated a decline of nearly 80%.
Institutional expectations have also diverged significantly. J.P. Morgan raised its target price for KNOWLEDGE ATLAS to HK$2000, believing its GLM-5.2 model is globally competitive and its open-weight strategy can create a growth path of "wider distribution - larger user scale - stronger paid conversion." Conversely, Citigroup lowered its target price for MiniMax from HK$1330 to HK$533, citing a lukewarm market response to its M3 model released in June and uncertainties in user retention and monetization strategies.
On July 10, MiniMax's founder and CEO, Yan Junjie, announced in an internal letter that he would forego his salary until the achievement of AGI (Artificial General Intelligence) and planned to allocate personal shares equivalent to 4% of the company's total equity over the next four years to incentivize the team.
On July 11, KNOWLEDGE ATLAS founder Tang Jie published an internal letter titled "The Giant Wave Has Arrived." He stated that the company would launch a two-year strategy called the "Touch High Plan," clearly indicating that "while the industry is generally accelerating commercial monetization, we have decided to break through upwards," and that the future focus would be "not pursuing short-term application monetization, but aiming directly for the next peak of AGI."
Currently, although KNOWLEDGE ATLAS has seen its market value shrink significantly during the decline, it still enjoys a substantial valuation premium compared to MiniMax. Market participants believe this reflects the market's continued higher regard for KNOWLEDGE ATLAS, backed by state support and possessing stronger commercialization certainty. However, the short-term liquidity crisis and shifts in the competitive landscape have prompted capital to exit temporarily and adopt a wait-and-see approach.
Comments