The Short Sellers' Verdict: Why Hong Kong's AI Darlings Are Facing a Crisis of Faith
We are told that AI is the new oil, the engine of the next industrial revolution. But what if the market is telling us a different story? In the heart of Hong Kong's financial district, a quiet but devastating signal has emerged: record short interest against two of China's most prominent AI pure-plays, MiniMax and Zhipu AI. This isn't just a market correction; it's a philosophical referendum on whether a pure-play AI model company can survive the brutal economics of the attention economy.
Let's start with the numbers, because they are stark. MiniMax, the company behind the popular chatbot, is facing a short interest of 20%—a level that screams institutional distrust. Zhipu AI, the 'Tsinghua AI dragon,' isn't far behind. Both stocks have been cut in half from their peaks, even as the broader market celebrates the AI boom. The catalyst? A perfect storm of lock-up expiries, a brutal price war, and a growing realization that 'being smart' is no longer a moat.
I've spent the last decade in the trenches of decentralized protocols, watching similar narratives play out in crypto. The pattern is eerily familiar: a technology with immense potential, a market that overprices the 'first mover,' and then the inevitable reckoning when the hype meets the P&L statement. The difference here is that the 'token' is a stock, and the 'community' is a group of short sellers like Hedgeye, who are betting against the very concept of the standalone model provider.
The core issue isn't the technology. Jefferies, a respected broker, notes that Zhipu's GLM-5.3 is on par with MiniMax's Kimi K3, and at a 19% lower cost per task. That's a technical win. Yet, the stock didn't rally. It fell. Why? Because in a market where every model is 'good enough,' the marginal value of a 5% performance boost is zero. The market is no longer paying for intelligence; it's paying for distribution, for enterprise contracts, for a path to profitability that doesn't involve burning cash on GPU clusters.
This is the 'middle ground' trap. MiniMax is neither the smartest (that's DeepSeek or the big labs) nor the cheapest (that's a race to the bottom). Zhipu has a cost advantage, but it's a double-edged sword—it signals to the market that you're competing on price, which is a death knell for margins. The short sellers aren't just betting on a price drop; they're betting on a structural failure of the business model. They're asking the question that every founder dreads: If your only differentiator is a slightly better benchmark score, and your costs are spiraling, what is your actual value?
Let's talk about the elephant in the room: the lock-up expiries. In July, Zhipu and MiniMax saw a combined $11.5 billion in shares unlock. That's not just a technical event; it's a signal. Early investors, the VCs who funded the 'AI for good' narrative, are now looking at their exit. The fact that they're selling into a falling market tells you everything about their internal conviction. They don't believe the current valuation is sustainable, and they have the data to prove it.
Now, here's where my contrarian nature kicks in. The narrative is overwhelmingly bearish, and the data supports it. But I can't help but wonder if we're missing the forest for the trees. The short sellers are right about the near-term pain, but they might be wrong about the long-term trajectory. The 'pure-play' model is under pressure, yes, but that doesn't mean the technology is worthless. It means the business model needs to evolve.
Consider the 'Ghost Protocol' I built during the 2022 bear market. It was a framework for privacy, but it taught me a lesson about resilience. The companies that survive a bear market aren't the ones with the best tech; they're the ones that adapt their narrative. MiniMax and Zhipu need to stop selling 'AI models' and start selling 'solutions.' They need to become the 'AWS of AI'—the infrastructure layer that others build on, not the flashy app that gets replaced in six months.
The Southbound capital flows are interesting here. Mainland investors are buying the dip, holding roughly 12% of Zhipu and 8% of MiniMax. They're playing a different game—a long-term bet on China's AI sovereignty. They're not looking at the P&L; they're looking at the strategic importance. This creates a fascinating tension: Western short sellers betting on financial reality, and Eastern long-termists betting on geopolitical necessity. In the short term, the shorts will win. In the long term, the state might just bail them out.
But let's be honest about the risks. The upcoming earnings reports in late August are the 'judgment day.' If MiniMax and Zhipu report widening losses and no clear path to profitability, the shorts will be vindicated, and the bloodbath will continue. If they surprise with strong revenue growth and a credible cost-cutting plan, we could see a violent short squeeze. The asymmetry is brutal.
I've audited enough protocols to know that when a token's price decouples from its utility, you get a correction. The same applies here. The 'utility' of these AI companies is their ability to generate cash flow. Right now, they're generating buzz, not cash. The market is finally waking up to that distinction.
Decentralization is a verb, not a noun. The same applies to AI. It's not about having the model; it's about what you do with it. The companies that will win this cycle are the ones that treat AI as a utility, not a religion. They'll embed it into workflows, they'll sell outcomes, not APIs. They'll become invisible, and that's when they'll become indispensable.
So, what's the takeaway? This isn't the death of AI; it's the death of AI hype. The short sellers are doing the dirty work of clearing out the froth. For the survivors, the path forward is clear: diversify revenue, focus on verticals, and stop trying to out-benchmark the giants. The market is telling you that 'smart' is cheap. 'Useful' is what pays the bills.
As I watch this unfold from my Seattle office, I'm reminded of the DeFi Summer of 2020. We had a thousand yield farms, all claiming to be the future of finance. Most of them died. The ones that survived were the ones that built real infrastructure, not just speculative tokens. The AI model market is no different. The 'yield farms' are the model providers with no business model. The 'infrastructure' is the layer that makes AI accessible, affordable, and accountable.
The question isn't whether MiniMax and Zhipu will survive. It's whether they'll pivot fast enough to become infrastructure, or whether they'll remain speculative tokens in a market that has lost its appetite for speculation. The next six months will tell us everything. I'm not betting against the technology; I'm betting on the evolution of the business. And in that evolution, the short sellers might just be the catalysts for a much-needed reset.