The market capitalization of KNOWLEDGE ATLAS has broken through the one trillion Hong Kong dollar mark as of this morning's reporting. In the four trading days leading up to the Dragon Boat Festival holiday, its stock staged a robust rally. It climbed for four consecutive sessions, accumulating gains of over 70%, with its total market cap reaching HK$933.6 billion at the close on June 18.

If you had invested HK$1 million in its shares on the January 8 listing date, your account would now hold HK$18 million. On investment platforms, some are openly sharing their success in making a HK$500,000 profit from KNOWLEDGE ATLAS in just a few days. However, the true beneficiaries of this wealth feast are not short-term retail investors, but rather the KNOWLEDGE ATLAS employees who, back in 2019 in a Beijing office, signed a few pages of "equity incentive agreements."

In recent years, such paper agreements had fallen out of favor at companies with sluggish growth. But in the AI era, at a company like KNOWLEDGE ATLAS that champions "common prosperity," these agreements, though still paper until the lock-up period expires, now carry vastly different weight and potential value.

Employee Wealth Multiplies Post-IPO

During the Dragon Boat Festival holiday, parents from across the country continued to queue with their children outside the gates of Peking University and Tsinghua University in Beijing's Haidian District. Within the AI circle, the Tsinghua alumni network is particularly prominent. KNOWLEDGE ATLAS, which created a batch of new fortunes just before the holiday, is a purebred "Tsinghua-affiliated" company.

Rewind to 2019, in Tsinghua Science Park just outside Tsinghua University's south gate, around 30 people from the university's Knowledge Engineering Laboratory founded KNOWLEDGE ATLAS and began their entrepreneurial journey.

That year was on the eve of the AI wave, with "computer vision" being the hot term. Sensetime, the so-called "first AI stock," was still operating in the same building. The industry had little understanding of KNOWLEDGE ATLAS, often simplistically comparing it to a "second Sensetime."

What KNOWLEDGE ATLAS later became is now well-known: a leading Chinese company building the foundational models for AGI. In terms of corporate governance, it designed a structural opportunity for "common prosperity" between leadership and employees.

In 2021, KNOWLEDGE ATLAS established two employee shareholding platforms—"Hui Hui" and "Zhi Deng"—which collectively hold approximately 14.96% of the company's shares.

According to its prospectus, as of the end of June 2025, KNOWLEDGE ATLAS had 883 employees, of which 452 held company shares, representing over half of the workforce.

The rationale for this design lies in AI being a knowledge-intensive industry where talent is the sole hard asset. The logic is similar to Tencent welcoming Yao Shunyu or Xiaomi recruiting Luo Fuli.

But hiring is easier than retaining. AI companies burn cash rapidly. Before its IPO, KNOWLEDGE ATLAS completed eight funding rounds over five years, raising over RMB 8.3 billion cumulatively, while incurring losses exceeding RMB 6 billion during the same period. Amidst fierce competition for talent where companies are burning cash, continuously raising salaries to retain staff in the short term is unrealistic. The alternative is to offer a future-oriented incentive.

"Common prosperity" is also standard practice among industry pioneers. Third-party estimates suggest OpenAI employees receive an average of about $1.5 million in equity compensation per person, roughly 34 times that of other large tech companies. Therefore, KNOWLEDGE ATLAS's approach is not impulsive but follows "international norms."

In hindsight, the employees of KNOWLEDGE ATLAS "bet correctly." Over the past six months, their paper wealth has multiplied in three significant leaps.

The first leap was the IPO. On January 8, 2026, KNOWLEDGE ATLAS listed on the Hong Kong Stock Exchange at an issue price of HK$116.2, with an initial market cap of approximately HK$52.8 billion.

The Hui Hui platform held an effective 8.97% stake post-listing. Based on the market cap at that time, excluding the two employees with the largest holdings in the platform, the remaining 424 employees had an average paper value already in the tens of millions of Hong Kong dollars.

The second leap occurred from February to May. Within 43 days of listing, KNOWLEDGE ATLAS's stock surged over 524% cumulatively, and its market cap once breached HK$323.2 billion. The average paper wealth of the 424 ordinary employees in the Hui Hui platform climbed another notch.

The third leap happened on June 18. While e-commerce platform employees were concerned about their sales and commission performance during the 618 shopping festival, KNOWLEDGE ATLAS directly boosted the paper wealth in its employees' accounts, with over 400 employees seeing their average paper wealth rise to staggering new heights.

The direct driver behind this was the official launch of GLM-5.2 on June 17, coupled with KNOWLEDGE ATLAS's STAR Market IPO entering the "guidance acceptance" phase. These multiple positive catalysts propelled the stock price to successfully surpass HK$2,094, an 18-fold increase from the issue price, culminating in this morning's market cap breaking the trillion-Hong-Kong-dollar mark.

Three waves of surges, three levels of magnitude increase.

Naturally, the biggest winners are KNOWLEDGE ATLAS's three founders. According to the 2025 annual report disclosure, the founders and core personnel collectively hold approximately 29.84% of the shares. Based on the latest market cap, the estimated paper wealth of core figures like Liu Debing, Zhang Peng, and Tang Jie has reached the billions or even tens of billions level.

The fact that KNOWLEDGE ATLAS is willing to pursue "common prosperity" already puts it ahead of many companies. It's no wonder rumors circulate that its employee turnover rate is notably low compared to others.

It also offers inspiration to many graduates of the class of 2026: how important it is to choose the right direction, the right company to work for, and a boss willing to share the wealth—which, of course, includes gaining entry into a top-tier university within the AI sphere.

Academic Roots Pave Way to Programming Elite

As ordinary investors, a common trait is being "slow to react."

Upon its release, GLM-5.2 was benchmarked against Anthropic's Claude Opus 4.8. However, five years ago in 2021, when KNOWLEDGE ATLAS first released its self-developed GLM pre-training framework, the industry's reaction was lukewarm. The prevailing question was whether this was just another academic group focused on publishing papers.

Tang Jie, co-founder and chief scientist of KNOWLEDGE ATLAS, recalled in an internal letter that from 2021 to 2022, large model development faced difficulties. Most people did not accept the "crazy plan" of making machines think like humans, akin to a moon landing, nor did they believe it could be a significant technological transformation opportunity, or they feared failure.

Wu Xi, a partner at Fortune Capital, also recalled that when first introduced to KNOWLEDGE ATLAS in 2021, "very few people could understand what KNOWLEDGE ATLAS was doing at the time."

In September 2022, KNOWLEDGE ATLAS open-sourced China's first trillion-parameter foundation model, GLM-130B. It spent a month on GitHub barely gathering 1,000 stars, gaining only modest recognition within technical circles, with little public awareness. It wasn't until the emergence of ChatGPT at the end of 2022 that people realized large models were the future.

Even during the domestic "Hundred Models War" from 2023 to 2025, the mainstream narrative from the industry to the media was: buying chips, stacking parameters, and chasing benchmark scores.

Investors were more interested in "how large is your model" rather than "how intelligent is it." KNOWLEDGE ATLAS chose the latter, pursuing the "path towards unseen intelligence." During this period, KNOWLEDGE ATLAS itself gained visibility—its ability to attract significant investment in a short time was proof—but its technical approach was not necessarily widely "understood."

This divergence began to change only in 2025. The market realized that after parameters are stacked to a certain point, marginal returns diminish severely. A larger model is not necessarily smarter but is certainly more expensive to run. KNOWLEDGE ATLAS persisted with its "efficient parameters" route—not blindly pursuing total parameter count but focusing more on the "utilization efficiency" of parameters—which allowed it to gradually carve its own path.

For the full year 2025, KNOWLEDGE ATLAS achieved revenue of RMB 724 million, a year-on-year increase of 131.9%. Its MaaS platform registered users exceeded 4 million. Among China's top ten internet companies, nine have deeply integrated the GLM model. Even with API call pricing increasing 83% compared to the end of last year, the market remains in a state of undersupply.

On June 17, 2026, KNOWLEDGE ATLAS released and open-sourced GLM-5.2, primarily designed for handling complex programming tasks and building AI agents. In simple terms, GLM-5.2 can autonomously complete development, integration, testing, and deployment, delivering a complete application within hours—a task that previously required a team working for weeks.

In the FrontierSWE benchmark, GLM-5.2 trails the current global strongest model, Claude Opus 4.8, by only about 1% to 4%, making it the highest-ranked open-source model. On the Arena.ai programming evaluation system CodeArena, GLM-5.2 scored 1595 points, ranking first among globally available models.

What was once an academic product has now become a vital assistant for programmers.

Previously, the recognized "big three" in AI programming were only Anthropic and OpenAI. Now, with KNOWLEDGE ATLAS joining, it completes the true "Coding Big Three" landscape.

Looking at KNOWLEDGE ATLAS's positioning domestically now: DeepSeek focuses on being the king of cost-effectiveness, Tencent's Hunyuan and Alibaba's Qwen leverage ecosystem synergy, Kimi handles underlying architecture personalization, and Doubao covers the entire population with entertainment and chat.

Meanwhile, KNOWLEDGE ATLAS, while peers compete on price, scenarios, or modalities, has chosen to relentlessly pursue the "upper bound of intelligence." And who is ultimately paying for this strategic direction?

Capital Shifting from DeepSeek's Logic to Atlas's Logic?

Just in the past week, media reported that DeepSeek completed its first funding round, raising over RMB 50 billion, with a valuation exceeding $50 billion, approximately RMB 340 billion. Tencent plans to invest RMB 10 billion, CATL plans RMB 5 billion, and Liang Wenfeng personally contributed RMB 20 billion, forming a prestigious capital lineup.

However, KNOWLEDGE ATLAS, with its market cap now exceeding one trillion Hong Kong dollars, has, in a sense, overshadowed DeepSeek.

Behind this, capital trends appear to be shifting from the "DeepSeek logic" to the "KNOWLEDGE ATLAS logic."

In recent years, DeepSeek gained fame internationally by championing "extreme cost-effectiveness, making AI cheap," also altering the logic in the Chinese market of valuing AI companies based on chip stockpiles. But this path is hitting a ceiling—compared to "high-order intelligence tokens," general-purpose tokens are rapidly depreciating, heading towards zero gross margin.

On May 22, 2026, DeepSeek announced a price reduction for its entire API input lineup, with its flagship model DeepSeek-V4-Pro seeing a 75% cut, and it was permanent. This set a new historical low for pricing among leading global large models, with input (cache hit) prices as low as RMB 0.025 per million tokens, an order of magnitude lower than the then-cheapest mainstream large model globally.

In contrast, KNOWLEDGE ATLAS has become the "contrarian," not lowering prices but instead raising API prices multiple times this year. Its logic is: premium for high-order intelligence, making AI indispensable.

The official pricing for the latest flagship GLM-5.2 is RMB 8 per million tokens for input and RMB 28 per million tokens for output, continuing its premium pricing strategy. Paradoxically, despite the price hikes, call volume has not decreased but instead increased by 400%.

This demonstrates KNOWLEDGE ATLAS's "real trick": there is genuine market demand, certified by capital as "real capability." Consequently, J.P. Morgan assigned KNOWLEDGE ATLAS an "overweight" rating.

This validates a statement once made by KNOWLEDGE ATLAS CEO Zhang Peng: "When a model is sufficiently powerful, the API itself is the best business model. High-order intelligence is the scarce resource of the moment; whoever masters the upper bound of intelligence holds the pricing power."

Beyond successful technology monetization, the soaring stock price of KNOWLEDGE ATLAS also involves capital market calculations.

KNOWLEDGE ATLAS has a relatively small proportion of freely tradable shares relative to its total share capital. On June 8, 2026, KNOWLEDGE ATLAS was included in the Hang Seng Tech Index. Index funds tracking the Hang Seng Tech Index "must allocate according to its weight," bringing substantial passive buying pressure.

However, there are also "cautious voices." A Jefferies research report provided a figure: KNOWLEDGE ATLAS's EV/S valuation is approximately 253x, which is indeed quite high. It maintained a "hold" rating with a target price of HK$912.55.

Currently, the risks facing KNOWLEDGE ATLAS remain significant and cannot be ignored. These include sustained substantial losses due to an unproven business model, impending massive pressure from lock-up share expirations, the divergence between extremely high valuation and limited fundamentals, and uncertainty regarding commercial willingness-to-pay.

Some believe that capital's relentless push to elevate KNOWLEDGE ATLAS represents a collective effort to trade time for space. Behind the scenes, various "major players" need to keep the game of passing the parcel going, allowing KNOWLEDGE ATLAS's stock price to gradually revert to its appropriate level.

Others believe that what capital is doing is using today's price to buy an option on the "final outcome of Chinese AI." One thing is certain: the future market does not need this many large model players battling it out, similar to today's new energy vehicle sector, where the market rewards only the leaders.