Hook: The Signal in the Sell-Off
On a trading day that will likely be dissected in boardrooms from Beijing to Menlo Park, shares of Zhipu AI and MiniMax—two of China's most prominent large language model startups—plummeted more than 11% on the Hong Kong Stock Exchange. The drop wasn't a blip. It wasn't a technical correction. It was a verdict.
For years, the narrative around Chinese AI has been one of relentless ascent: massive funding rounds, ambitious technical roadmaps, and the promise of a "China ChatGPT" that would reshape the digital economy. But the Hong Kong market, with its notoriously unforgiving attitude toward unprofitable tech companies, just delivered a different kind of message. It's a message that echoes what I've seen repeatedly in my years covering both crypto and emerging technology: the gap between what private markets believe a company is worth and what public markets are willing to pay is where the truth lives.
This isn't just a story about two companies. It's a story about the end of a valuation era—and the beginning of a much more uncomfortable one.
Context: The Second Tier and the Hong Kong Gauntlet
To understand what's happening, we need to step back and look at the landscape. China's AI "Big Four"—Zhipu, MiniMax, Moonshot AI (Kimi), and Baichuan—have been the darlings of the domestic tech scene. They've raised billions, hired the best talent from Tsinghua and other elite institutions, and released models that, on paper, compete with the best America has to offer.
But there's a hierarchy. The first tier belongs to the giants: Baidu with Ernie Bot, Alibaba with Tongyi Qianwen, ByteDance with Doubao. These companies have distribution, data, and the ability to subsidize AI development with profits from their core businesses. The second tier—where Zhipu and MiniMax sit—has none of those luxuries. They're pure-play AI companies, which means they're burning cash in a race where the finish line keeps moving.
Zhipu, with its GLM series, has positioned itself around open-source models and government/enterprise (B2G/B2B) deployments. It's a sensible strategy on paper—the Chinese government is a massive potential customer, and there's a premium on domestic AI sovereignty. But the reality is that this market is fiercely competitive, with Alibaba and Baidu fighting for the same contracts, often with deeper pockets and more comprehensive cloud ecosystems.
MiniMax, on the other hand, has bet on consumer-facing products like Talkie and Hailuo AI. The "AI + social" thesis is compelling—if you can build a companion app that people genuinely love, the network effects could be enormous. But here's the uncomfortable truth that the market is now pricing in: consumer AI retention and monetization remain unsolved problems globally. Even OpenAI, with ChatGPT's massive user base, has struggled to convert engagement into sustainable, high-margin revenue. The idea that a Chinese startup can crack this code while competing against ByteDance's distribution machine is, at best, an unproven hypothesis.
The choice of Hong Kong as a listing venue is itself revealing. For Chinese tech companies, Hong Kong has become the default option when US listings are complicated by audit disputes and geopolitical tensions. But Hong Kong is not the NASDAQ. Its investors are more conservative, more focused on fundamentals, and less willing to tolerate "story stocks" that don't generate profits. This is the same market that watched SenseTime—the so-called "first AI stock in Hong Kong"—lose over 70% of its value since its 2021 IPO. The precedent is not encouraging.
Core: The Valuation Paradigm Shift and the SPAC Trap
Let me be direct about what I believe is happening here, based on my years of auditing whitepapers and analyzing market structures: this sell-off is the concentrated release of a private-market AI valuation bubble into the unforgiving arena of public markets.
The math is brutal. In 2023 and 2024, private investors were throwing money at AI companies based on a simple narrative: "AI is the future, and these are the leaders." Zhipu reportedly reached a valuation of around 20 billion RMB (roughly $2.8 billion) in 2024. MiniMax similarly commanded eye-watering numbers. These valuations were based on potential, not performance. They were based on the assumption that China's AI market would grow fast enough to justify the burn rates.
But here's what the private markets missed, and what the public markets are now ruthlessly correcting: the path from "promising model" to "profitable business" is much longer and more expensive than anyone wanted to admit.
The evidence is in the price action. A drop of 11% isn't a normal fluctuation. It suggests institutional investors are exiting positions in size, not retail panic. It suggests that the lock-up periods for early investors and cornerstone backers may be expiring, and those investors are choosing to take whatever liquidity they can get rather than hold for a better price. This is the classic "valuation arbitrage" problem: private investors got in at one price, public markets are offering a significantly lower one, and the smart money is cutting its losses.
There's also a strong possibility that these companies went public via SPAC (Special Purpose Acquisition Company) mergers. I've seen this pattern before, both in crypto and in tech. SPACs offer a fast track to listing, but they come with a hidden cost: they often embed inflated valuations that the market immediately corrects. Historical data shows that SPAC-listed companies, on average, underperform significantly in the 12 months following their merger. If Zhipu and MiniMax took this route, the current decline might not be the bottom—it might be the beginning of a longer, more painful normalization.
Let me put this in a framework I've used throughout my career, from auditing ICO whitepapers in 2017 to analyzing DeFi protocols in 2020: when a token or stock is priced on narrative rather than cash flow, the correction is not a bug—it's a feature. The market is a truth-telling machine, and it's telling us that the gap between private-market expectations and public-market reality is enormous.
The comparison to SenseTime is instructive. SenseTime went public in December 2021 at a valuation that seemed to justify its position as China's leading AI vision company. Within two years, it had lost more than 70% of its value. The reasons were complex—regulatory crackdowns, geopolitical tensions, and a fundamental mismatch between its revenue growth and its valuation. But the lesson was clear: Hong Kong is not a place for AI companies to raise money at inflated valuations. It's a place where they go to be judged.
And the judgment is harsh. Hong Kong investors want to see revenue growth, gross margins, and a clear path to profitability. They don't care about your model's benchmark scores on MMLU or your impressive paper at NeurIPS. They care about whether you can sell something to someone at a price that covers your costs. This is a fundamentally different evaluation framework than what private markets use, and it's catching Chinese AI companies completely off guard.
Contrarian: The Opportunity in the Rubble
Now, let me offer a perspective that might seem counterintuitive, especially given the doom-and-gloom narrative. This sell-off might be creating the first genuinely attractive entry point for AI exposure in Hong Kong since the sector's inception.
Here's my reasoning. The market is now pricing these companies for failure. The 11% drop, combined with the broader sector weakness, suggests that investors have swung from irrational exuberance to irrational pessimism. The truth, as always, lies somewhere in between.
Zhipu and MiniMax are not bad companies. They have real technology, real teams, and real products. Zhipu's GLM models are genuinely competitive, and the company has a credible strategy for the government and enterprise market. MiniMax's consumer products have shown flashes of product-market fit, even if the monetization path is unclear. The question isn't whether these companies have value—it's whether that value is closer to the private-market valuation or the current public-market price.
If the current price reflects a genuine belief that these companies will fail, then there's an opportunity. If it reflects a temporary mismatch between supply (investors exiting lock-up positions) and demand (institutional investors waiting for clarity), then the downside might be limited.
I've seen this pattern before. In the crypto market, I've watched projects with real technology get crushed to valuations that made no sense—only to recover when the market realized it had overcorrected. The key is to distinguish between companies with fundamental problems and companies that are simply caught in a sector-wide downdraft.
For Zhipu, the government and enterprise market is a real opportunity. The Chinese government is committed to AI sovereignty, and it will pay for domestic models. The question is whether Zhipu can win enough of that market to justify its current valuation. For MiniMax, the consumer AI market is a lottery ticket—but it's a lottery ticket with a real product and a real team.
The broader opportunity, though, might be in the AI supply chain. When model companies' valuations compress, it creates pressure on the entire ecosystem. GPU cloud providers, data services, and application developers all face repricing. But this repricing also creates opportunities for companies with real revenue and real profits. The AI industry is maturing, and maturity means that the "picks and shovels" companies—the ones that provide infrastructure and services to AI developers—might be better investments than the model companies themselves.
This is a lesson I learned in the crypto market. During the 2022 bear market, the projects that survived were the ones with real usage and real revenue, not the ones with the most impressive technology demos. The same principle applies here. The AI industry is entering its "DeFi Summer" moment—the transition from speculative narrative to practical utility. And in that transition, the winners will be the companies that can demonstrate actual economic value, not just technical prowess.
Takeaway: The Signal for the Next Narrative
So what does this mean for investors, for the AI industry, and for the broader technology landscape?
First, it means that the era of "story-driven" AI valuations is over. From now on, AI companies will be judged on the same metrics as any other business: revenue, margins, customer retention, and cash flow. This is a healthy development, even if it's painful for the companies and investors who benefited from the old regime.
Second, it means that the Chinese AI market is entering a consolidation phase. The "hundred models war" is ending, and the survivors will be the ones with the strongest distribution, the deepest pockets, or the most defensible technology. Zhipu and MiniMax are not guaranteed to be among the survivors, but they're not guaranteed to be among the casualties either. The next 12-18 months will be decisive.
Third, it means that the AI investment thesis is shifting from "who has the best model" to "who has the best business model." This is a shift I've seen before in the crypto market, and it's always painful for the early-stage investors who bought into the hype. But it's also the moment when the real value creation begins.
The question I keep coming back to is this: Are we witnessing the death of the Chinese AI dream, or the birth of a more mature, more sustainable industry? My instinct, based on years of watching technology cycles, is that it's the latter. The hype is dying, but the technology is real. The companies that survive this correction will be stronger, more focused, and more likely to build lasting value.
But here's the uncomfortable truth that the market is teaching us: trust is the only currency that matters. And trust, in the public markets, is earned through performance, not promises. Zhipu and MiniMax have a lot of work to do to earn that trust. The market has given them a clear signal: show us the receipts, or we'll keep marking you down.
The next few quarters will be telling. If these companies can show accelerating revenue, improving margins, and a clear path to profitability, the current sell-off will look like a buying opportunity. If they can't, the decline will continue, and the broader Chinese AI ecosystem will feel the pain.
In the meantime, I'm watching the signals. I'm watching for quarterly reports, for customer announcements, for any sign that these companies are making the transition from "AI story" to "AI business." And I'm reminding myself of a lesson I've learned over and over in my career: noise filtered. Signal preserved. The noise is the daily price action, the panic, the doom-scrolling. The signal is the underlying fundamentals, the technology, the team, and the market opportunity.
The Hong Kong market has just delivered a powerful signal. The question is whether anyone is listening.