The rapid deflation of the capital myth surrounding the so-called "first large language model stock" has been startling.

In early trading on June 10, shares of KNOWLEDGE ATLAS (HKG: 02513) plunged once again. By the midday break at 12 p.m., the stock price had fallen 8.45% to HK$1,040, giving the company a total market capitalization of HK$463.7 billion. From its historical intraday high of HK$1,993 on May 29, the stock has nearly halved in value, wiping out over HK$420 billion in market capitalization. This staggering loss is equivalent to the combined market value of a company like Baidu plus that of a company like XPeng.

It is hard to believe that just two weeks ago, this company was the brightest star in the entire Hong Kong stock market. Having listed on the Hong Kong Exchange on January 8 with an issue price of HK$116.2, it staged a capital market miracle in less than five months. Within 43 days of listing, its share price skyrocketed to HK$725, a gain of over 500%. The rally continued into May, with the stock surging 83% that month alone.

The deeper reason for the market's high valuation premium lies in the company's background, its top-tier model as the engine, its enterprise-level API as the primary product form, and its explosive growth achieved through integration into industrial ecosystems. Data shows that from 2022 to 2025, its revenue from cloud-based MaaS services grew from RMB 2.59 million to RMB 190.38 million, with its contribution to total revenue increasing from 4.5% to 26.3%. As of March 2026, the annual recurring revenue (ARR) for its MaaS platform was approximately RMB 1.7 billion, representing a growth of about 60 times over the past year.

However, the abrupt halt to the celebration caught everyone by surprise. On May 29, the very day the stock hit its all-time high of HK$1,993, it suddenly experienced a dramatic reversal, with an intraday swing exceeding 30%, ultimately closing down 1.42%. Since then, decline has been the sole theme. On June 5, the stock plummeted 9.05% in a single day, erasing HK$57.5 billion in market value. On June 9, it crashed another 13.55%, wiping out HK$79 billion in one session.

Even the long-anticipated positive catalyst of being included in the Hang Seng Tech Index failed to stem the tide. Since its formal inclusion, the stock has fallen more than 20% in just two trading days.

Furthermore, on the evening of June 1, the company unexpectedly announced the initiation of a listing process on Shanghai's STAR Market, aiming to raise up to RMB 15 billion. This comes less than five months after its Hong Kong listing, which raised net proceeds of approximately HK$4.173 billion from the global offering.

Financial reports reveal that the company posted a net loss of RMB 4.718 billion in 2025. Even after adjustments, the loss stood at RMB 3.182 billion, which is 6.5 times its revenue. A more fundamental issue is the complete disconnect between its valuation and its business fundamentals. At its peak market cap of HK$880 billion, the company had annual revenue of just RMB 724 million, resulting in a staggering price-to-sales (P/S) ratio exceeding 1,124 times. To put this in perspective, the P/S ratios for Microsoft and Apple are around 10 times. Even the most aggressive overseas AI companies are valued at less than one-tenth of this company's valuation. In other words, 99% of the stock price was based on future expectations, with only 1% grounded in reality.

Encroachment by industry giants has further squeezed the company's operating space. Competitors like Baidu, Alibaba, and ByteDance, with their own computing power clusters, have driven API call prices to rock-bottom levels, sacrificing profitability for market share. In contrast, this company lacks its own computing infrastructure. Each model iteration requires expensive GPU procurement, resulting in a per-token cost far higher than that of its rivals.

It is also noteworthy that on July 8, the company will face its first major post-IPO share unlock, involving approximately 11.9859 million shares held by cornerstone investors.