The Hong Kong AI Correction: When Narrative Liquidity Meets Structural Reality
Over the past 72 hours, two of China's most prominent AI large-model companies—Zhipu AI and MiniMax—have shed more than 11% of their market value on the Hong Kong Stock Exchange. The sell-off wasn't isolated. It rippled through the entire AI concept sector, dragging down sentiment across a market that had, until recently, priced these companies as the vanguard of a new technological era.
I don't need to tell you that this feels like a familiar pattern. I've watched this movie before—in 2021, when DeFi protocols with zero revenue were commanding billion-dollar valuations, and in 2022, when the modular blockchain thesis was being sold to anyone with a venture capital email address. The characters change. The narrative structure doesn't.
What's happening in Hong Kong right now isn't a story about Chinese AI companies failing. It's a story about what happens when a narrative built on potential collides with a market that demands proof. And for anyone who's been paying attention to how capital flows actually work, the signals were there long before the price charts confirmed them.
The Context: Two Companies, One Structural Problem
Zhipu AI, backed by Tsinghua University's technical pedigree, has built its commercial strategy around the GLM series of large models. Their revenue model leans heavily on B-end API calls, private deployment, and government-enterprise partnerships. MiniMax, by contrast, has staked its claim on consumer-facing products—Talkie, Hailuo AI—with monetization built around subscriptions and advertising.
On paper, these are complementary approaches to the same problem: how to turn artificial intelligence into a sustainable business. In practice, they share a critical vulnerability. Neither company has demonstrated the kind of revenue growth, gross margin profile, or customer retention metrics that Hong Kong's institutional investors require before they'll assign a premium multiple.
This isn't a China-specific problem. OpenAI, with annualized revenue now in the tens of billions, still faces questions about its path to profitability. But there's a crucial difference: US public markets have shown remarkable tolerance for AI companies that can tell a compelling growth story. Hong Kong's investors, scarred by years of underperforming tech listings, demand evidence.
The Core: Valuation Paradigm Shift and the SPAC Trap
Let me be direct about what I think is happening here. This sell-off represents the concentrated release of a primary market AI valuation bubble into the secondary market. It marks a paradigm shift from story-driven valuation to performance-driven valuation—and it was entirely predictable.
Based on my audit experience across both crypto and traditional tech markets, I've developed a framework for understanding these transitions. It has three components.
First, the valuation logic that dominated 2023-2024 was built on narrative potential. Zhipu's reported valuation of approximately 20 billion RMB in 2024 was justified by technical leadership and market size projections—not by revenue multiples or cash flow analysis. This is the same logic that drove DeFi valuations in 2021, and it carries the same structural weakness: narratives can sustain valuations only as long as the next round of capital is willing to accept them.
Second, the SPAC listing path—which I strongly suspect both companies used, given the timing and structure of their listings—creates a specific valuation trap. Historical data shows that SPAC-listed companies experience an average decline of over 50% within 6-12 months of listing. The mechanism is straightforward: SPACs allow companies to go public with less scrutiny, which means the initial valuation often reflects promotional optimism rather than market reality. The correction isn't a bug. It's a feature of the structure.
Third, the market environment in Hong Kong is fundamentally hostile to unprofitable growth companies. Liquidity is thin. Southbound capital flows have shifted from speculative enthusiasm to cautious selectivity. And the global AI investment cycle has entered a cooling phase—US AI stocks have shown increased volatility, and investors are demanding clearer paths to monetization.
The Contrarian Angle: What the Market Is Missing
Here's where I diverge from the consensus narrative. The market is treating this as a crisis of Chinese AI. I think it's actually a crisis of narrative construction—and that distinction matters.
Consider the comparison to SenseTime, Hong Kong's first AI stock, which has lost over 70% of its value since its 2021 IPO. The market reads this as evidence that Hong Kong can't support AI valuations. I read it differently. SenseTime's decline wasn't a failure of the AI thesis. It was a failure of the company to transition from research excellence to commercial execution. The market didn't lose faith in AI. It lost faith in SenseTime's ability to monetize it.
The same logic applies to Zhipu and MiniMax. Their decline doesn't necessarily mean their underlying businesses are deteriorating. It means the market is demanding evidence of commercial viability that these companies haven't yet provided. This is a liquidity problem, not a solvency problem.
There's another angle that most analysts are missing. The compression of AI model company valuations will inevitably cascade through the entire value chain. GPU cloud providers, data services, downstream application developers—all of these will face valuation resets. But this creates opportunity. Companies with real revenue and genuine profitability will become relatively more attractive as the speculative froth is stripped away.
I've seen this pattern before. In 2022, when modular blockchain infrastructure was being written off as a bear market casualty, the companies that survived were the ones with actual usage metrics and sustainable revenue models. The narrative reset didn't kill the sector. It separated the signal from the noise.
The Takeaway: What Comes Next
Let me offer a framework for what I'm watching over the next 6-18 months.
First, watch for the quarterly disclosures. If Zhipu and MiniMax can demonstrate accelerating revenue growth, improving gross margins, and narrowing losses, the current valuation levels will look like a buying opportunity in hindsight. If they can't, the decline will continue—and it will accelerate.
Second, watch the primary market. If other Chinese AI companies—Moonshot AI, Baichuan, Zero One Everything—begin delaying their funding rounds or accepting lower valuations, that confirms the systemic nature of this correction. If they continue raising at previous levels, this is company-specific rather than sector-wide.
Third, watch the regulatory environment. China's AI policies around model registration and data security will shape the competitive landscape. Companies that can navigate regulatory requirements while maintaining innovation velocity will emerge stronger.
The deeper question is whether Chinese AI companies can transition from the narrative of technological potential to the reality of commercial execution. The market is no longer willing to fund potential alone. It wants evidence.
I don't think this is the end of the Chinese AI story. I think it's the end of the beginning. The companies that survive this correction will be the ones that built real businesses underneath the narrative. The ones that don't will become case studies in the difference between storytelling and value creation.
In crypto, we have a saying: narrative liquidity is more important than technical liquidity. The same principle applies here. The narrative that carried Zhipu and MiniMax to public markets has been exhausted. The next narrative—the one that will drive the next phase of value creation—will be built on metrics, not promises.
The question isn't whether Chinese AI will succeed. It's which companies will be around to define that success. And that answer will be determined not by the quality of their technology, but by the discipline of their execution.
Follow the structure, not the hype. The market is always telling you what it believes. The only question is whether you're listening.
