On July 15, 2026 (Beijing time), Hong Kong’s AI sector once again took center stage in the market. The Hang Seng Index opened up 0.87%, while the Hang Seng Tech Index rose 0.78%. AI concept stocks rebounded across the board, with Zhipu (02513.HK) opening over 6% higher, surging more than 5% during the session, then narrowing gains to 0.19% at the time of writing, closing at HKD 1,603. MiniMax (00100.HK) jumped 7.48%, closing at HKD 247.2. The semiconductor sector also strengthened, with SMIC up 2.5% to HKD 80.3 and Hua Hong Semiconductor up 2.6% to HKD 175.9.
This trend is not an isolated event. Since July 8, Zhipu and MiniMax have experienced a series of critical milestones, including the lifting of lock-up restrictions, large-scale refinancing, and founder internal communications. Their stock prices have diverged sharply—Zhipu soared 13.85% on its unlock day, while MiniMax plunged nearly 18% in a single day after almost half its shares became tradable. By the close of July 13, the market cap gap between the two companies had widened to more than tenfold.
The market is now grappling with a deeper question: Are Hong Kong-listed AI large model companies transitioning from the "valuation hype" phase to the "commercial value validation" phase?
Global AI Investment Enters Second Phase: From Computing Infrastructure to Models and Applications
To understand the changing pricing logic in Hong Kong’s AI sector, it’s essential to first clarify the evolution of the global AI industry cycle.
He Pianpian, Chief Overseas Tech Analyst at Huatai Securities, uses Jensen Huang’s "baseball nine innings" theory to divide the stages of AI industry development: 2023–2025 marks the first inning, with the focus on large model training, centered on GPUs and HBM as core computing resources. In 2026, the industry officially enters the second inning, shifting to agent intelligence and low-latency inference scenarios. Goldman Sachs, in its July 14 report, further notes that the AI industry is moving toward multipolar competition, where the core of the second half is "effective output"—that is, which companies can generate the highest "useful output" per dollar of AI investment.
This shift in the industry cycle is directly reflected in capital market pricing logic. Over the past three years, the global AI investment narrative has centered on a "computing power arms race"—whoever owns more GPUs and larger parameter models commands a valuation premium. But as we enter 2026, this logic is being revised. The market is now asking: Can computing power investments translate into sustainable revenue? Can model capabilities generate commercial value in real-world scenarios?
UBS Securities, in its mid-July outlook, clearly lists "model capability, monetization, and Token ROI" as the three key themes for China’s AI models in the second half of 2026. This signals that Chinese AI companies’ valuation anchors are shifting from "technology narrative" to "commercial validation."
Hong Kong, as the world’s only capital market with pure large model companies listed, naturally becomes the most direct observation window for this shift in valuation logic. Zhipu and MiniMax both debuted on the Hong Kong Stock Exchange in January 2026, receiving enthusiastic market attention—Zhipu’s IPO price was HKD 116.2, and MiniMax’s IPO was heavily oversubscribed. Six months later, their stock prices and market caps have fundamentally diverged, reflecting the market’s differentiated pricing for distinct commercialization paths.
Zhipu and MiniMax: Two Distinct Paths to AI Commercialization
Though Zhipu and MiniMax are both considered the "dual giants" of large models on the Hong Kong market, they differ fundamentally in their technical approach, market positioning, and commercialization strategies.
Zhipu (02513.HK): Enterprise AI and Domestic Large Model Ecosystem
Zhipu’s core narrative centers on enterprise AI services and domestic large model infrastructure. Originating from a Tsinghua University lab, its GLM series models are recognized as leading domestic large models. By July 2026, Zhipu had built a full-stack technical ecosystem from chip adaptation to model deployment, open-sourcing over 60 cutting-edge models globally under MIT licenses. International downloads surpassed 100 million, serving over 5 million enterprise users and developers.
On the commercialization front, Zhipu demonstrates "price setter" behavior. According to Yuanta Securities, Zhipu doubled its API prices while maintaining steady business growth. Changjiang Securities forecasts company revenues of RMB 2.5 billion, RMB 6.5 billion, and RMB 12.5 billion for 2026–2028, with year-on-year growth of 244%, 162%, and 91%. JPMorgan raised Zhipu’s target price twice in one week—from HKD 1,800 to HKD 2,000, then to HKD 2,400—believing its fundraising will effectively ease computing power supply bottlenecks.
On July 11, Zhipu founder Tang Jie issued an internal letter announcing the launch of the "Touch High" initiative, committing to strategic investment in AGI fundamental research over the next two years, rather than pursuing short-term application monetization. He wrote, "Since the end goal is AGI, short-term interests or industry trends are merely scenery along the way." This strategic choice won long-term capital support—on the unlock day, Zhipu’s stock price surged 13.85%, delivering a rare "unlock day celebration."
MiniMax (00100.HK): Multimodal AI and Global Consumer Applications
MiniMax has taken a different path. Its core strengths lie in multimodal capabilities and consumer product deployment. Its flagship M3 model is natively multimodal, supporting image, video input, and desktop operations. On the application side, MiniMax has launched Talkie/Xingye (emotional interaction), Hailuo AI (content creation), and other consumer products, serving over 200 million users worldwide, with Talkie/Xingye users averaging over 70 minutes of daily usage.
MiniMax’s commercialization goals are notably aggressive. Goldman Sachs’ July 3 report shows management is confident in achieving USD 1 billion in annual recurring revenue (ARR) by the end of 2026.
However, the capital market’s pricing for MiniMax is far less optimistic than for Zhipu. On July 9, nearly 150 million shares—representing 48.9% of total equity—were unlocked, with the free float expanding from less than 3% to nearly 50%. The stock plunged 17.98% on the unlock day, closing at HKD 297.4. The decline continued, hitting a historic low of HKD 209.2 during trading on July 14. By the July 13 close, MiniMax’s market cap had fallen from its March peak of over HKD 410 billion to under HKD 70 billion.
On July 10, MiniMax founder Yan Junjie announced he would forgo all compensation until AGI is achieved, putting up personal shares equivalent to 4% of the company’s total equity for team incentives. That same day, the company announced a capital raise of about HKD 16.04 billion through new share placement and convertible bond issuance. Yet these moves failed to stem the decline—JPMorgan twice lowered MiniMax’s target price in one week, from HKD 300 to HKD 240.
These two divergent paths and outcomes reflect a fundamental shift in how the capital market values Chinese AI large model companies.
Reconstructing Valuation Logic: From User Numbers and Traffic to Model Capability and Commercial Revenue
The divergence in Zhipu and MiniMax’s market caps essentially mirrors the shift in valuation anchors for AI companies.
Previous Valuation Logic: User Numbers + Traffic
In the first phase of AI investment, the market focused on metrics like user scale, DAU, and call volume. MiniMax, thanks to its consumer products like Talkie/Xingye and its overseas expansion narrative, once enjoyed a substantial valuation premium—soaring 109% on its IPO day, reaching HKD 1,330 in March, and breaking HKD 410 billion in market cap. However, the traffic story failed to sustain market confidence.
Future Valuation Logic: Model Capability + API Revenue + Agent Ecosystem + Enterprise Service Revenue
As we enter 2026, the market’s focus is undergoing a structural shift. According to Securities Star, the first batch of pure large model companies listed in Hong Kong are experiencing sharp valuation corrections: "The market frenzy is fading, capital is no longer simply paying a premium for ‘scarce listed targets,’ and the domestic large model industry is officially bidding farewell to the parameter narrative and scarcity premium era. Valuation anchors are shifting entirely to commercialization capability, computing cost control, and sustainable cash flow."
Specifically, the new valuation framework includes at least four dimensions:
Model capability is the basic threshold. Zhipu’s GLM-5 series leads SWE-bench Pro programming evaluations, with GLM-5.2 topping CodeArena’s global model rankings. MiniMax’s M3 model has carved out differentiated advantages in multimodal fields. Yet model capability alone is no longer sufficient—the market cares more about whether these capabilities can be monetized.
API revenue is the most direct indicator of commercialization. Zhipu maintained business growth even after doubling API prices, demonstrating pricing power; MiniMax expanded its developer ecosystem by open-sourcing M3 model weights. The scale and growth rate of API revenue are now key metrics for assessing the commercial value of large model companies.
Agent ecosystem is seen as the core form of next-generation AI applications. Zhipu lists "autonomous intelligent systems" as one of the four technical engines in its "Touch High" initiative; MiniMax’s agents are positioned as "general intelligent agents capable of completing long-range complex tasks." Huatai Securities notes that the industry’s main line in 2026 has shifted to agent intelligence. Whoever builds a sustainable agent ecosystem first will gain an edge in the next phase.
Enterprise service revenue is becoming a crucial variable for distinguishing valuation tiers. Zhipu’s enterprise-focused positioning aligns with current market preferences—serving over 5 million enterprise users and developers, combined with the policy narrative of domestic substitution, makes it easier to secure a valuation premium. In contrast, MiniMax’s consumer-driven model faces greater challenges in user retention, paid conversion, and competition.
Tang Jingcao of Shuimu Capital points out that the model layer is undergoing a major reshuffle: "The current coexistence of dozens of large model companies in China is unsustainable. As model capabilities converge, open-source ecosystems mature, and giants continue free strategies, by 2027, fewer than 10 general large model companies are expected to survive and thrive independently." Against this backdrop, the capital market will inevitably scrutinize every AI company’s business model and sustainability with stricter standards.
Value Reassessment or Valuation Correction?
Returning to the question posed at the start: Are Hong Kong-listed AI large model companies undergoing a value reassessment?
The answer may be twofold—Zhipu is receiving a "value reassessment" premium, while MiniMax is facing a harsh "valuation correction."
Zhipu’s post-unlock rally and successive target price upgrades from institutions show the market’s endorsement of its "long-termism" strategy and enterprise positioning. On July 14, northbound capital net bought HKD 2.989 billion of Zhipu in a single day, with sustained inflows from long-term funds supporting its stock price. JPMorgan predicts that Zhipu’s newly acquired inference computing resources could convert into annual recurring revenue over the next 12 months.
MiniMax’s challenges are more complex. The unlock fundamentally altered its share structure—free float expanded from less than 3% to nearly 50%—changing its pricing logic. Add to this the pressure from financial investors exiting, dilution effects from fundraising, and doubts about the sustainability of its consumer business model, and MiniMax’s valuation correction may not be over yet.
However, it’s worth noting that CICC maintained a "buy" rating for MiniMax on July 14, citing its status as a rare resource in the AI large model field, rapid revenue growth, and infrastructure deployment exceeding expectations. This suggests the market’s pricing for MiniMax is not uniformly bearish but is undergoing a dramatic repricing process.
From a broader perspective, the fundamentals of China’s AI large model industry remain strong. According to OpenRouter data, China’s weekly AI large model call volume reached 23.45 trillion tokens, up 15.01% week-on-week, and has led the world for ten consecutive weeks, surpassing the US. The top six global call volumes are all Chinese models. The industry’s high prosperity stands in stark contrast to the capital market’s volatility—a typical feature when the industry cycle shifts from "expectation-driven" to "validation-driven."
For investors, Hong Kong’s AI large model sector is entering a phase where discernment is more crucial than ever. The era of traffic stories and scarcity premiums is fading, replaced by refined assessments of model capability conversion efficiency, API revenue growth, agent ecosystem development, and enterprise service depth. The divergence between Zhipu and MiniMax may just mark the beginning of this value reassessment process.
FAQ
Q1: What are the stock codes for Zhipu and MiniMax in Hong Kong?
Zhipu’s Hong Kong stock code is 02513.HK, listed on January 8, 2026, with an IPO price of HKD 116.2. MiniMax’s code is 00100.HK, listed on January 9, 2026. Both are known as the "dual giants" of large models on the Hong Kong Stock Exchange and are currently the only pure large model companies listed there.
Q2: Why did Zhipu’s stock rise after unlock while MiniMax’s plunged?
The core differences are threefold: In unlock scale, Zhipu unlocked only 5.76% of shares, while MiniMax unlocked 48.9%. In shareholder structure, Zhipu’s unlocked shares were mostly held by industrial capital and long-term institutions, while MiniMax’s included many financial investors. In financing methods, Zhipu’s placement price of HKD 1,588 was fully subscribed by institutions, while MiniMax’s placement plus convertible bonds raised concerns about dilution.
Q3: What is Zhipu’s "Touch High" initiative?
On July 11, 2026, Zhipu founder Tang Jie issued an internal letter announcing the launch of the "Touch High" initiative—strategically investing in AGI fundamental research for the next two years, without pursuing short-term application monetization. The plan focuses on four technical directions: long-range tasks, autonomous intelligent systems, fully self-training, and extreme safety governance. The company simultaneously completed a HKD 31.4 billion placement to fund the initiative.
Q4: How does MiniMax’s business model differ from Zhipu’s?
MiniMax focuses on multimodal AI and consumer applications, with products like Talkie/Xingye and Hailuo AI serving over 200 million users globally. Zhipu is centered on enterprise AI services, serving over 5 million enterprise users and developers, and building a domestic large model ecosystem. They represent the consumer (C-end) and enterprise (B-end) commercialization paths of AI, respectively.
Q5: What will be the core of future valuations for Hong Kong-listed AI large model companies?
The market’s valuation anchor is shifting from "user numbers + traffic" to "model capability + API revenue + agent ecosystem + enterprise service revenue." The ability to turn technical capabilities into sustainable commercial income, build ecosystem moats in the agent era, and achieve positive returns on computing power investment will become the key variables determining valuation.
