Hook: The Numbers That Broke the Narrative
Over the past 48 hours, two of China's most prominent AI large-model companies—Zhipu AI and MiniMax—have watched their Hong Kong-listed shares shed more than 11% of their value in a single trading session. The market didn't blink. There was no dramatic headline, no regulatory bombshell, no catastrophic product failure. Just the quiet, relentless recalibration of expectations.
I've spent the better part of a decade watching valuation narratives form, inflate, and eventually rupture. The pattern is always the same: first the story, then the numbers, then the silence when the two refuse to align. What we're witnessing in Hong Kong right now isn't a market correction. It's a paradigm shift disguised as a sell-off.
The ledger remembers every trembling hand that bought the narrative at the top. And right now, that ledger is bleeding red.
Context: The Road to Hong Kong
Let me set the stage properly, because the context here matters more than the price action itself.
Zhipu AI, born from Tsinghua University's prestigious computer science program, has long positioned itself as China's answer to OpenAI. Its GLM series of large language models has earned genuine respect in technical circles. The company's strategy has been methodical: B2B API access, private deployments for enterprise clients, and a steady stream of government partnerships. It's the kind of business that sounds solid on paper—until you ask about the revenue figures.
MiniMax took a different path. The company bet on consumer-facing products, launching Talkie and the Hailuo AI assistant. The "AI + social" thesis was compelling in theory: if you could crack user retention and monetization in the consumer AI space, the upside was enormous. But consumer AI products globally have struggled with a fundamental problem—users try them, marvel at the novelty, and then drift away. The retention curves tell a story that PowerPoint decks rarely capture.
Both companies chose Hong Kong over the United States. That decision alone speaks volumes. Whether it was driven by audit regulatory constraints, geopolitical risk avoidance, or the increasingly hostile environment for Chinese tech listings in New York, the result is the same: they now face the scrutiny of Hong Kong investors, a group historically far less forgiving of unprofitable growth stories than their American counterparts.
The deeper context here is the structural mismatch between how private markets valued these companies and what public markets are willing to pay. In the private markets of 2023-2024, AI was the only game in town. Every fund wanted exposure to the narrative. Zhipu reportedly reached valuations around 20 billion RMB in private rounds. The story was everything: technological leadership, market potential, the China AI exceptionalism thesis.
Then these companies hit the public markets, and the story collided with something far less romantic: actual financial statements.
Core: The Valuation Reckoning
Let me be precise about what's happening, because the surface narrative—"AI stocks are falling"—obscures the mechanical reality underneath.
The core issue is that Hong Kong's public market is forcing a paradigm shift from story-driven valuation to performance-driven valuation, and the transition is proving brutal.
Here's what I mean by that. In the private markets, AI companies were valued on a combination of technological capability, market size, and narrative momentum. The question investors asked was "what could this become?" rather than "what is this right now?" This is standard for early-stage investing, but the scale of the valuations involved created a dangerous assumption: that public market investors would share the same forward-looking perspective.
They don't. And they never have.
Hong Kong's market structure amplifies this problem. The city's exchange has a long history of punishing unprofitable tech companies, with SenseTime serving as the cautionary tale that every AI founder should study. Since its 2021 IPO, SenseTime has watched its market capitalization evaporate by over 70% from its peak. The pattern is so consistent that it's almost become a law of Hong Kong markets: if you're an AI company without clear profitability, expect your stock to bleed.
The mechanics of this decline are worth understanding. When a stock falls 11% in a day, it's rarely retail investors driving the move. It's institutional repositioning. It's funds that received their allocations at IPO or through SPAC mergers, finally hitting their lock-up expiration dates and deciding that the risk-reward no longer justifies holding. It's the quiet, systematic unwinding of positions that were never based on conviction, but on the hope that someone else would buy higher.
I've audited enough on-chain data and market microstructure to recognize the signature of forced selling versus genuine capitulation. The volume patterns in this sell-off suggest institutional distribution, not panic. Someone is methodically reducing exposure, and the market is absorbing the supply at increasingly lower prices.
The SPAC angle deserves particular attention. If these companies went public through special purpose acquisition companies—and the evidence strongly suggests they did—they've walked into a statistical trap. Historical data shows that SPAC-listed companies average declines of over 50% within 12 months of their merger completion. The structure itself creates perverse incentives: sponsors need to find targets, targets need to hit aggressive projections, and the resulting pressure to overpromise creates a predictable correction when reality sets in.
The fundamental problem is that both companies are burning cash at a rate that their current revenue cannot support, and the public markets have finally started pricing that reality.
Let me break down the commercialization challenges that the market is now scrutinizing.
Zhipu's enterprise-focused model sounds robust until you examine the competitive landscape. In China's government and enterprise AI procurement, they're competing against Alibaba's Tongyi Qianwen, Baidu's Ernie Bot, and ByteDance's Doubao. These aren't startups with limited resources—they're divisions of some of the largest technology companies in the world, with virtually unlimited capital to subsidize their AI offerings. When Alibaba decides to price its API access at a loss to capture market share, Zhipu has to either match the pricing and destroy its margins, or maintain pricing and lose the enterprise deals.
MiniMax faces a different but equally challenging dynamic. Their consumer-focused strategy puts them in direct competition with ByteDance, which has essentially unlimited distribution through Douyin (TikTok's Chinese counterpart) and Jianying (CapCut). The traffic advantage that ByteDance enjoys is not something that can be overcome through superior technology alone. Consumer AI is a distribution game, and ByteDance holds all the cards.
The market is beginning to understand these dynamics, and the understanding is showing up in the price action.
Contrarian: The Unreported Angle
Here's where I diverge from the consensus narrative. Everyone is framing this as a story about AI companies failing to meet expectations. I think it's something more interesting: this is the first real test of whether China's AI industry can survive the transition from subsidized ambition to economic reality.
The conventional wisdom says that the decline of Zhipu and MiniMax is bad news for China's AI ecosystem. I'd argue the opposite. This correction is the market doing its job—forcing a reckoning that was inevitable and necessary.
Consider what happens next. The valuation compression at the public market level will transmit directly to the private markets. The next time Zhipu or MiniMax—or their peers like Moonshot AI or Baichuan—go out to raise capital, they'll face significantly lower valuation expectations. This is painful for existing shareholders, but it's healthy for the industry. It forces capital allocation toward companies with genuine revenue traction rather than those with the most compelling narrative.
The second contrarian angle: this sell-off may be creating the first genuine value opportunity in Chinese AI. If the fundamentals haven't deteriorated—if the technology is still strong, if the enterprise pipeline is still growing, if the cost structure is improving—then the current price may represent a significant discount to intrinsic value. The market is pricing these companies for failure, but the reality is more nuanced.
I've seen this pattern before. In the crypto markets, the same dynamic plays out repeatedly: the narrative drives the price to unsustainable levels, the correction overshoots to the downside, and the companies with real fundamentals eventually recover while the pretenders disappear. The key is identifying which is which.
The silence in the market right now is the only honest metadata. Neither company has issued a statement explaining the decline. No analyst calls, no investor presentations, no reassuring press releases. That silence tells me that the companies themselves are uncertain about how to respond—which suggests they don't have good news to share.
But here's what the silence also tells me: the companies are conserving their ammunition. They're not going to waste their best arguments on a market that's in full risk-off mode. They're waiting for the selling to exhaust itself, and then they'll make their case.
Takeaway: What to Watch Next
The next 90 days will determine whether this is a buying opportunity or the beginning of a longer decline. Here's what I'm watching:
First, the volume patterns. If the selling continues on increasing volume, there's more downside ahead. If volume dries up and the price stabilizes, the selling pressure has been absorbed.
Second, any announcements from the companies themselves. New enterprise contracts, revenue updates, or technology milestones would signal that the fundamentals are improving despite the market's skepticism. Silence, on the other hand, will confirm the market's worst fears.
Third, the broader Hong Kong tech sector. If the Hang Seng Tech Index is also declining, this is a systemic issue rather than a company-specific problem. If Zhipu and MiniMax are underperforming the index significantly, the market is making a specific statement about their business models.
Fourth, the private market response. If we see other Chinese AI companies postponing their IPO plans or accepting significantly lower valuations in private rounds, that confirms the systemic nature of this repricing.
The deeper question that nobody is asking: what happens to China's AI ambitions if the public markets refuse to fund them? The answer will shape the industry's trajectory for years to come.
Speed wins the trade, but clarity wins the war. Right now, the market is moving fast, but the clarity is still emerging. The ledger is still recording every transaction, every position unwound, every hand that trembled at the sell button. The truth will emerge from the data, as it always does.
The question isn't whether Zhipu and MiniMax can survive this correction. The question is whether they can thrive in a world where the narrative no longer protects them from the numbers. That's the test that will separate the real AI companies from the stories dressed up as businesses.
Infinite leverage, finite patience. The market has run out of patience. Now we'll see what's actually left.