The Liquidity Trap Nobody Discusses
The numbers hit the terminal at 10:47 AM Hong Kong time. Zhipu, down 11.2%. MiniMax, down 10.4%. Two of China's most prominent AI "Four Little Dragons" companies, bleeding value in a single trading session on August 24th. The market didn't wait for explanations. It never does.
But here's what the flash news didn't tell you: this wasn't a technical failure. No model collapsed. No security breach surfaced. No catastrophic product launch went sideways. What we witnessed was something far more structural—a repricing of an entire asset class that had been trading on narrative rather than numbers.
I've spent two decades watching capital flow through technology cycles, from the ICO mania of 2017 to the DeFi summer of 2020 to the NFT bubble of 2021. The pattern is always the same. When liquidity contracts, the market doesn't discriminate between good stories and bad ones. It simply demands proof.
And right now, the proof isn't there.
The Valuation Mirage: When Market Cap Exceeds Addressable Logic
Let me be direct about what's happening here. Zhipu completed a funding round in early 2024 that valued the company at over 20 billion RMB. MiniMax crossed the billion-dollar valuation threshold. These are not small numbers. These are "we expect you to be a category-defining platform" numbers.
The problem? Neither company has publicly disclosed revenue figures that would justify those valuations under any traditional metric. We're not talking about high P/E ratios here—we're talking about companies where the "E" doesn't exist yet. This is the "story stock" valuation model, where investors pay for potential rather than performance.
The market has shifted from pricing dreams to pricing deliverables.
In my analysis of over 50 ICO whitepapers back in 2017, I learned a critical lesson: when you strip away the narrative, the underlying fundamentals either support the valuation or they don't. Most of the time, they don't. The same principle applies here, just with more sophisticated packaging.
The Hong Kong market is particularly unforgiving in this regard. Unlike US markets, where growth stories can run for years on promise alone, Hong Kong investors have historically demanded a clearer path to profitability. This isn't necessarily rational—it's structural. The liquidity pool is shallower, the investor base is more conservative, and the tolerance for narrative-driven valuation is correspondingly lower.
The Competitive Squeeze: Giants Don't Ask Permission
Here's what the market is actually pricing in, and it's not just valuation concerns. It's competitive reality.
Zhipu and MiniMax are caught in a pincer movement that would crush any startup, regardless of technical merit. On one side, you have the internet giants—Baidu with ERNIE, Alibaba with Tongyi Qianwen, ByteDance with Doubao. These aren't competitors; they're ecosystems. They bundle model capabilities with cloud infrastructure, distribution channels, and enterprise relationships that startups simply cannot replicate.
On the other side, you have nimble competitors like DeepSeek and Moonshot AI, who have carved out specific niches—open-source credibility for DeepSeek, long-context processing for Moonshot. The middle ground that Zhipu and MiniMax occupy is becoming increasingly untenable.
The price war that erupted in 2024 made this structural disadvantage explicit. When major players slashed API prices by over 90%, they weren't just competing—they were signaling that they could afford to operate at a loss indefinitely. For startups whose unit economics were already thin, this was existential.
I've seen this play out before. In the DeFi summer of 2020, I modeled liquidity depth across Uniswap v2 and Compound, tracking how stablecoin pegs correlated with Ethereum gas spikes. The pattern was clear: when capital is abundant, everyone looks smart. When it contracts, the structural weaknesses become visible. The same dynamic is playing out in China's AI sector right now.
The Macro Context: Liquidity Evaporates Faster Than Hype
Let me zoom out for a moment, because this isn't just a China story. It's a global liquidity story with Chinese characteristics.
The 2022 bear market taught us something crucial about the relationship between macro conditions and crypto asset valuations. When the Federal Reserve began its aggressive rate hike cycle, we didn't just see Bitcoin decline—we saw the entire risk asset complex repriced. Stablecoin minting rates collapsed. DeFi TVL followed. The causal chain was clear: tightening liquidity flows through every asset class, and the most speculative assets feel it first.
The same mechanism is at work in Hong Kong's AI stocks. Global risk appetite has been volatile throughout 2024, with investors increasingly questioning the timeline for AI returns. The "AI bubble" narrative has moved from fringe blogs to mainstream financial media. When that happens, the marginal buyer disappears, and the marginal seller becomes more aggressive.
What we're witnessing is not a company-specific problem. It's a sector-wide repricing event.
The Hong Kong market amplifies this dynamic because of its structural characteristics. Lower liquidity means larger price swings for any given volume of selling pressure. High-valuation, pre-profit companies are the most vulnerable to this dynamic because they have the least fundamental support to fall back on.
The Commercialization Gap: Where's the Revenue?
Let me be blunt about the core issue: neither Zhipu nor MiniMax has demonstrated that they can build a sustainable business at scale.
Their primary business model—API calls and B2B services—is sound in theory. The demand for AI capabilities is real and growing. Code generation, intelligent customer service, content creation—these are actual use cases with actual budgets attached. But the revenue numbers don't yet match the valuation multiples.
The market is asking a simple question: if these companies are worth 20 billion RMB, where is the revenue to support that? And the answer, based on available public information, is that it's not there yet.
This isn't necessarily fatal. Many successful technology companies took years to monetize their user bases. But in a tightening liquidity environment, patience becomes a scarce commodity. Investors want to see progress, not promises.
I've been tracking the commercialization trajectories of Chinese AI companies since the regulatory framework was established in 2023. The pattern is consistent: strong technical capabilities, weak go-to-market execution, and a growing dependence on government and state-owned enterprise contracts that are slow to close and even slower to pay.
The Contrarian Angle: This Might Be the Opportunity
Now let me offer the counter-argument, because the picture isn't entirely bearish.
The selloff may have created a genuine entry point for investors with a longer time horizon.
Here's the logic: the underlying technology is real. Zhipu's GLM series and MiniMax's abab series are legitimate technical achievements. They're not world-beating, but they're competitive. The demand for AI capabilities in China is not a bubble—it's a structural shift that will play out over the next decade.
The question is whether these companies can survive the transition from narrative-driven valuation to fundamentals-driven valuation. That's a survival test, not a death sentence.
What would change my assessment? Clear evidence of commercialization progress. API call volumes growing quarter over quarter. Enterprise customers signing multi-year contracts. Gross margins stabilizing despite the price war. These are the metrics that would justify the current valuation—or at least make it defensible.
The market is not asking these companies to be profitable tomorrow. It's asking them to show a credible path to profitability.
That's not an unreasonable demand. It's actually the market functioning as it should.
The Regulatory Angle: Hong Kong's Strategic Play
There's another dimension to this that most Western observers miss. The Hong Kong listing of these AI companies isn't just about capital formation—it's about positioning.
Hong Kong has been aggressively positioning itself as Asia's premier technology listing venue, competing directly with Singapore for the region's innovation capital. The Chinese government has signaled its support for this strategy, viewing Hong Kong as a bridge between Chinese technology and international capital.
The AI stock selloff is a test of Hong Kong's credibility as a technology listing venue.
If these companies continue to decline, it sends a signal to other Chinese tech companies considering Hong Kong listings. It suggests that the market doesn't have the depth or the appetite to support high-valuation technology companies. That's a problem for Hong Kong's ambitions.
This is why I expect to see policy support if the selloff continues. Not direct intervention in stock prices—that would be too obvious—but regulatory signals that support the AI sector's long-term prospects. The government has too much invested in both the AI industry and Hong Kong's financial center status to let this spiral.
The Technical Reality: Infrastructure Costs and the Profitability Question
Let me get into the technical weeds for a moment, because this is where the real story lies.
Training and running large language models is expensive. Really expensive. We're talking about GPU clusters that cost hundreds of millions of dollars to build and maintain. The compute costs alone are a significant drag on any AI company's financials.
For Zhipu and MiniMax, this is compounded by the US export controls that limit access to high-end chips like NVIDIA's H100 and A100. They're forced to rely on domestic alternatives like Huawei's Ascend series or downgraded versions of Western chips. This isn't just a performance issue—it's a cost issue. Domestic chips are less efficient, which means higher costs for the same compute output.
The infrastructure burden is one of the key reasons these companies are burning through cash.
I've analyzed the compute requirements for various model architectures, and the economics are brutal. The training runs alone can cost tens of millions of dollars. And that's before you factor in the ongoing inference costs of serving customers.
This is the hidden variable in the AI valuation equation. The market sees the revenue potential, but it often underestimates the cost structure required to deliver that revenue. When you factor in the infrastructure burden, the path to profitability becomes even longer and more uncertain.
The Talent War: The Silent Killer
There's another factor that doesn't show up in financial statements but is critical to long-term success: talent.
The competition for AI researchers and engineers in China is intense. The internet giants can offer compensation packages that startups simply cannot match. Stock options in a company whose valuation is declining are less attractive than guaranteed cash bonuses from a profitable tech giant.
The talent drain is a slow-moving crisis that could undermine the technical capabilities of these companies over time.
I've seen this pattern before in the crypto industry. Companies that couldn't retain their core technical talent during the 2022 bear market found themselves at a significant disadvantage when the market recovered. The same dynamic is now playing out in China's AI sector.
The Takeaway: Positioning for the Next Cycle
So where does this leave us?
The immediate picture is bearish. Zhipu and MiniMax are facing a perfect storm of valuation pressure, competitive intensity, and macro headwinds. The selloff could continue, and I wouldn't be surprised to see further declines in the coming weeks.
But the longer-term picture is more nuanced. The AI industry in China is not going away. The demand for AI capabilities is real and growing. The question is which companies will survive the current consolidation phase and emerge as the leaders in the next cycle.
The market is in the process of separating the companies with real technical capabilities and viable business models from those that were riding the narrative wave.
This is a healthy process, even if it's painful for the companies involved. The companies that emerge from this period with their technical teams intact, their cash reserves sufficient, and their commercialization strategies validated will be the ones that create long-term value.
For investors, the current selloff represents an opportunity to build positions in companies that have the potential to be category leaders. But it requires patience and a willingness to look through the short-term volatility.
Entropy is the only constant in liquid markets. The current chaos will eventually resolve into a new equilibrium. The question is who will be standing when it does.
The Signal in the Noise
Let me close with a final observation about what this selloff tells us about the broader market.
The AI stock decline in Hong Kong is not an isolated event. It's part of a broader pattern of repricing across the technology sector. The era of easy money and narrative-driven valuations is over. The market is demanding evidence of value creation, and it's punishing companies that can't provide it.
This is a good thing for the industry in the long run. It forces discipline, encourages efficiency, and separates the real innovators from the pretenders. But it's a painful process for those caught in the middle.
Fractures in the ledger reveal the truth of value. The current selloff is exposing the gap between what these companies are worth and what the market was willing to pay for them. The correction is uncomfortable, but it's necessary.
The companies that will thrive in the next cycle are those that can demonstrate real revenue growth, sustainable unit economics, and a clear path to profitability. The rest will fade into obscurity, remembered only as footnotes in the history of the AI boom.
The market is not rational; it is resistant. But eventually, it finds its level. And when it does, we'll know which companies were building real value and which were just building castles in the air.