August 7, 2025. Hong Kong's closing auction prints a number that shouldn't exist in a sideways market.
MiniMax-W (00100.HK) up 24.8 percent. Zhipu AI (02513.HK) up 17.2 percent. Two of China's most-watched AI model companies. Same session. Same direction. No earnings call. No product launch. No government decree. Just two vertical lines ripping through trading ranges that had been dead space for weeks.
We didn't get a single press release from either company explaining the move. We didn't get a regulatory filing. We didn't get an official confirmation from the Hong Kong exchange. What we did get was a market data feed from Bitget โ a cryptocurrency exchange โ reporting the surge to its global user base. That's not how institutional capital validates a thesis. That's how a story gets distributed.
Let me be precise about what this tells us. In a consolidating market where the Hang Seng Tech Index has been chopping sideways for a month, investors were willing to pay a 25 percent premium in one session for exposure to China's AI ecosystem. But the why behind this move has not been independently verified. And in my experience โ years of auditing protocol launches and watching price action outrun fundamentals โ the day after a 25 percent move, the question everyone avoids is simple: Did we just witness a re-rating, or did we witness a rumor?
Here's the context that matters.
Both companies are part of what the Chinese market calls the AI Big Six Tigers โ the six most prominent large language model startups that emerged from the post-ChatGPT funding wave alongside Moonshot AI, Baichuan Intelligence, 01.AI, and StepFun. These are the names building China's answer to OpenAI, Anthropic, and Google DeepMind. And two of them just made the leap from the private market to a public one.
MiniMax and Zhipu listed on the Hong Kong Stock Exchange under Chapter 18C โ the special regime designed for pre-profit special technology companies. The venue was explicitly created to attract capital-hungry AI and biotech names that couldn't pass traditional profitability tests. That's the structural reason these companies can trade at all. But the structural reason they surged on August 7 is far murkier.
The two companies couldn't be more different in practice.
MiniMax runs a consumer-first playbook. Its products include Hailuo AI, a content generation tool, and Talkie, an AI emotional companionship app that goes head-to-head with Character.AI in overseas markets. This is a product-driven growth story, monetizing consumers, not enterprises. It's the 'China's Character.AI plus Midjourney' narrative wrapped into one stock.
Zhipu runs the opposite play. Born from Tsinghua University lineage, its GLM series has iterated from GLM-130B to GLM-4, building a complete foundation-model-plus-alignment research chain. Its revenue comes from API calls, private deployments, and government and enterprise contracts across finance, education, healthcare, and public sectors. This is a land-and-expand enterprise story. High-ticket contracts. Long sales cycles. Deep customization costs.
One consumer. One enterprise. Two different routes to the same destination. And on August 7, the market treated them as one interchangeable basket.
That's the first signal I want to flag. When two companies with fundamentally different business models, customer bases, and unit economics move in identical double-digit percentages on the same day, the market isn't pricing their individual fundamentals. It's pricing a sector thesis. It's buying a label: China AI Core Assets.
Now let's talk about what that actually means and why the rally deserves more skepticism than enthusiasm.
The Tape: What the Numbers Actually Tell Us
Let's start with the numbers, because they're the only hard facts in this story.
MiniMax-W, trading under ticker 00100.HK, closed up roughly 25 percent. Zhipu AI, under 02513.HK, closed up north of 17 percent. Both companies are recent listings. Both have relatively small public floats โ a structural feature of Chapter 18C listings, where a limited number of shares are available for public trading while the majority remain locked up with pre-IPO investors.
Small floats create amplified moves. A $50 million order into a stock with $200 million of free-floating shares moves the price far more than the same order into a $2 billion float. This isn't speculation; it's basic market microstructure. Newly listed AI companies in Hong Kong routinely exhibit this behavior. The question is whether the move on August 7 was large enough to be a fundamental repricing or just a thin-market aberration.
Here's what I can say from watching similar patterns in crypto markets: a 25 percent single-day move in a low-float asset is often the beginning of a move, or the end of one. There's no middle ground. In crypto, when a fresh listing on a major exchange rips 25 percent on no news, the pattern is typically driven by one of three forces: a strategic accumulator taking a large position, a market maker unwinding a hedge, or a coordinated narrative campaign designed to attract attention. The same forces operate in equities. And just like in crypto, the follow-through matters more than the first candle.
The absence of a company announcement is remarkable. In normal circumstances, a 25 percent move triggers a 'price-sensitive information' inquiry from the exchange. Companies often issue a 'no knowledge' statement clarifying that there are no undisclosed corporate developments. Neither MiniMax nor Zhipu did so publicly within the initial reporting window. Silence in the face of a 25 percent move is either a sign that the companies genuinely don't know what happened, or that they know exactly what happened and aren't allowed to talk about it. Both scenarios are more interesting than a clean fundamentals story.
Two Roads, One Valuation
Now let's dig into the companies themselves, because the divergence in their business models is exactly what the market's synchronized pricing is ignoring.
MiniMax represents the consumer-internationalist bet. It's a company that effectively said: China's domestic enterprise market is too slow; let's build consumer products for the world. Talkie is the flagship โ an AI companionship app that competes in the same arena as Character.AI. The product targets emotional engagement, and in 2025, that's a massive market. The global AI companionship space has seen explosive user growth, with monetization through subscriptions, virtual gifts, and premium features.
Hailuo AI, meanwhile, targets AI media generation โ the image-and-video creation space dominated in the West by companies like Runway. If Talkie is the emotional hook, Hailuo is the utility play. Both products share a common architecture: they lean on MiniMax's proprietary trillion-parameter MoE (Mixture of Experts) models, which the company claims to have trained on domestic Ascend 910B chip clusters. That's a significant technical claim. It suggests MiniMax has gone deep on domestic AI infrastructure, building training pipelines that don't depend on Nvidia's most advanced GPUs. In a geopolitically constrained environment, that's a genuine moat โ or a genuine liability, depending on how Ascend cluster reliability and software maturity hold up at scale.
Zhipu's technical foundation is different. The GLM series has been a research-first project since its inception. GLM-4, the latest major iteration, has established a credible reputation in Chinese-language understanding and tool calling. Zhipu has also released open-source variants like GLM-4-9B, building a developer ecosystem that functions as a low-cost acquisition channel. The 'open core' strategy is straightforward: open-source models attract developers; the company monetizes enterprises that need guaranteed performance and compliance through closed-source APIs and private deployments.
These two companies have completely different revenue profiles. MiniMax's consumer revenue is high-volatility, high-upside, driven by user acquisition costs and overseas payments infrastructure. Zhipu's enterprise revenue is predictable but slow, with long sales cycles and heavy customization burdens. The market's decision to price both stocks identically on August 7 suggests one of two things: either investors are making a broad macro bet on China AI rather than analyzing the underlying companies, or the market is rationalizing post-hoc that 'AI momentum will lift all boats.'
Based on my audit experience โ where I've watched investors conflate ETF narratives with protocol fundamentals โ the first explanation is far more likely.
The Liquidity Illusion
The next lens is liquidity, and this is where the rally gets uncomfortable.
Hong Kong's new-economy listing regime has a well-documented pattern: low float, high volatility, then a slow bleed as lockups expire. The first few weeks after listing often produce dramatic price swings precisely because the supply of shares is so constrained. The August 7 surge fits that pattern perfectly. With a limited public float, a concentrated buy order from a few funds โ or even a single large accumulator โ can move the price by double digits.
Who might be buying? Here's where the Bitget data feed becomes important. The fact that this story circulated through a cryptocurrency exchange suggests that crypto-native investors are watching these listings. That's not coincidental. We're in a market where crypto capital increasingly looks for AI exposure, and Chinese AI stocks offer a liquid, regulated avenue to bet on the AI narrative without touching tokens. The overlap between crypto communities and AI-equity speculation has been growing all year.
But this capital behaves differently from long-term institutional capital. Crypto-native money is fast. It indexes to narratives, not fundamentals. It chases momentum and abandons positions just as quickly. If the marginal buyer on August 7 was a crypto fund rotating into Chinese AI equities, the move carries a very different risk profile than if the marginal buyer was a Hong Kong pension fund accumulating for the long run.
We don't know who bought. The Hong Kong exchange doesn't publish real-time beneficiary ownership data during trading. We'll only see the footprint when ownership disclosures and clearing data become available in the coming weeks. Until then, the identity of the buying force remains a mystery โ and a 25 percent move built by anonymous, potentially momentum-driven capital is a fragile thing.
Capital Flow Mechanics: Southbound vs. Foreign Money
There's another angle that institutional traders are already parsing: southbound flows.
If the surge was driven by mainland Chinese investors routing capital through the Stock Connect program, it would reflect domestic policy optimism โ a bet that Beijing will continue supporting the AI sector with directives, subsidies, and favorable procurement. If the surge was driven by international investors, it would indicate that foreign capital is reassessing China AI as an investable theme despite geopolitical tensions.
The two carry completely different implications.
Southbound money tends to be influenced by mainland market narratives, often moving in episodes of sentiment-driven trading. Foreign institutional money, in contrast, generally requires a more visible catalyst โ an earnings report, a strategic partnership, a valuation reset โ before committing large positions. Given the absence of a catalyst, the southbound hypothesis feels more consistent with the move. But without daily southbound holding data broken down by stock โ and we're unlikely to get granular data for weeks โ this remains an inference, not a conclusion.
The equity derivatives market might offer a clue. If the surge came with unusual call volume or a spike in single-stock futures activity, that would point to sophisticated positioning. If it came on plain cash-market buying, it smells more like narrative-driven retail or momentum money. I haven't seen the derivatives tape for August 7. And that gap in information alone should temper anyone's confidence in explaining this move.
The Index Inclusion Speculation
One more mechanically important factor: index inclusion.
MiniMax and Zhipu are candidates for inclusion in the Hang Seng Composite Index and, eventually, the Southbound Stock Connect eligible list. Inclusion would trigger passive buying from exchange-traded funds and index trackers. The market frequently front-runs these announcements. A surge on no news, days before a quarterly index review window, is exactly the kind of pattern that sparks inclusion speculation.
If that's what's happening, the rally has a semi-rational foundation โ but it also means the near-term upside is capped by the review schedule. Once the index decision lands, the speculative premium evaporates. The stock either gets included, rewarding the front-runners, or gets excluded, punishing them. This is a binary event with a binary market reaction, and neither outcome supports a sustainable re-rating narrative.
The Contrarian Angle: The Rally Nobody Verified
Now here's the part the mainstream coverage is entirely missing.
The August 7 surge is not evidence of strength. It's evidence of a vacuum โ a vacuum of information, of fundamental catalysts, and of meaningful liquidity depth. We're treating a low-float upward price movement as a signal of China AI's commercial viability when it might just be the result of a small, anonymous group of traders positioning ahead of a mechanical event.
Let me be direct about the uncomfortable comparison. This looks like one of those early crypto altcoin listings: limited supply, concentrated holders, fast market cycles, and a chart that moves on narratives rather than revenue. The Bitget distribution is the tell. The crypto ecosystem's attention machine arrived before the institutional research coverage did. When a crypto exchange's ticker feed is the first to report a major move in a regulatory equity, you're looking at an audience crossover that says more about the liquidity class than the company fundamentals.
There's also a structural danger here that nobody's discussing: valuation anchoring for the private market.
MiniMax and Zhipu's public valuations now serve as anchors for the private companies still waiting to list โ Moonshot AI, Baichuan, 01.AI, and StepFun. If the public market is pricing these two at 25 percent above their recent private market levels, the four unlisted tigers will use these numbers in their next funding round negotiations. That pushes private valuations up. But private valuations don't always reflect fundamental improvement โ they reflect comparison sets. And if the public surge turns out to be a liquidity mirage, the 'anchor' that looked like a blessing becomes a curse. Unlisted AI companies will demand higher prices, close fewer rounds, and face a down-round correction when the anchor resets.
That's not my speculation. That's a pattern I've seen repeat across both crypto and tech markets. When the reference price rises on thin volume, the last people to benefit are the founders who raised at the inflated comparison.
And then there's the elephant in the room: DeepSeek. The open-source reasoning model that crushed China's commercial AI pricing power in 2025. MiniMax and Zhipu both operate in the shadow of DeepSeek's near-zero marginal cost for high-quality inference. If DeepSeek's models remain open and competitive, the headroom for closed-model monetization narrows. Consumer apps can still win on interface and experience. Enterprise deals can still win on compliance and customization. But the pricing power that supports a premium valuation narrative is eroding every quarter. The stock market's synchronized surge on August 7 contained no analysis of this competitive pressure.
There's one more uncomfortable angle. Regulation didn't drive this move. There was no new policy announcement, no model approval, no subsidy package. And that absence matters. In a China AI market that has historically moved on policy signals, a surge without a policy backdrop suggests the move is more about market mechanics than about the industry's operating environment. If Beijing's next statement is cautious โ or worse, if it announces new compliance requirements โ the same market mechanics that drove prices up could reverse them just as violently.
The Verification Checklist
So what would actually justify the surge? Here's the evidence trail I'd need before taking the move at face value.
First, volume confirmation. A 25 percent move on steadily expanding volume over 5-10 trading sessions has meaning. A single-day spike on thin volume is noise until proven otherwise. I'd want a clear picture of that day's volume relative to the listing-period average.
Second, disclosure. If either company issues a price-sensitive announcement in the coming days โ a contract win, a strategic investment, a new model release โ the surge gains a fundamental anchor. If we get a 'no knowledge' statement, watch out.
Third, an earnings beat. Both companies will eventually report their first public earnings. For MiniMax, the metric that matters is paid conversion and user retention on Talkie and Hailuo. For Zhipu, it's contract value and customer renewal rates in the government and enterprise segment. Until those numbers land, a 25 percent rise is a vote of confidence, not a confirmation.
Fourth, sector breadth. Did other Hong Kong AI names โ like SenseTime or 4Paradigm โ move in sympathy on August 7? If yes, this is genuinely a sector rotation, possibly a sign of sustained capital interest. If no, it's an isolated event and likely an idiosynchratic market structure effect rather than an industry signal.
Fifth, the derivative market. I'd check the options and single-stock futures activity for that session. Positioning data tells you whether sophisticated capital was involved or whether this was purely cash-market momentum.
The Takeaway: What to Watch, Not What to Believe
I came into this analysis expecting to find an obvious catalyst โ a new model announcement, a government deal, a major investor's position increase. Instead, the evidence points to a market mechanism event: a low-float, newly listed AI stock catching a wave of attention in a sideways market with nothing to anchor it.
Here's what I'm actually watching now.
Short-term, the next five to ten trading sessions will separate signal from noise. If prices hold above the surge levels on average or better volume, the market is telling us something real. If they fade back to pre-surge levels, we'll know the August 7 liquidity spike was exactly what it looked like: an anomaly.
Mid-term, the first earnings reports will be the true test. MiniMax's consumer business and Zhipu's enterprise business are both building toward their first public financial disclosures. Revenue growth, gross margin, cash burn, and customer counts will reveal whether the AI commercial stories that supported these listings are real.
Long-term, there's a bigger question. Can Hong Kong sustain a genuine AI sector? The city has been searching for a growth narrative since the tech crackdown years. MiniMax and Zhipu are the first wave of China AI names to provide one. They're joined by a pipeline of waiting companies โ Moonshot and the others โ whose listing plans will move faster if these two stocks hold their gains.
The concern is circularity. If the sector forms because stocks rise, and the stocks rise because the sector forms, nothing actually needs to be true underneath to sustain the loop โ until it stops working. And when the loop breaks in a low-liquidity, high-narrative market, it breaks fast.
In my eleven years watching markets โ across crypto, DeFi, and now AI equities โ I've learned that the most expensive mistake is treating price movement as evidence. Price movement is just information. Evidence is what survives scrutiny.
August 7 delivered a 25 percent price movement. It has not yet delivered the scrutiny. We didn't get an answer that day. But we did get a question worth holding onto: Are we pricing China's AI future โ or just the first ripple of capital trying to find a home in a market that has no other growth story?
The next few weeks will tell us. And for once, I'm happy to wait for the data.
But make no mistake โ in a sideways market, the first whale to move is always the most dangerous. And on August 7, the whale was moving before anyone had a chance to ask why.