Hook: The Signal in the Spread
MiniMax short ratio hits 20%. That's not a trade signal; it's a liquidity event. For context, anything above 10% is a red flag. At 20%, the market is not just betting against the stock – it's front-running a structural thesis. The timing is deliberate: shorts piled on ahead of the August 26 earnings report. The bet is not that the company will miss revenue; it's that the business model itself is broken. Smart money doesn't trade the headline; it trades the block time. And the block time here is the earnings release.
But this isn't just about MiniMax. Zhipu AI, the other Chinese AI darling, sees a short ratio of 6% – still elevated for a stock that's already down 50% from its peak. The two companies together represent the pure-play AI model layer in Hong Kong. And the market is telling you they are toxic.
Context: The Kimi Shockwave
In July, Moonshot AI's Kimi K3 dropped. The model didn't just improve – it jumped a generation. Within days, Zhipu AI lost 24% of its market cap. MiniMax lost 18%. The market priced in a technology gap that no sell-side analyst had predicted. Jefferies later confirmed the gap: Zhipu's GLM-5.3 offers "similar performance at 19% lower cost" – a classic follower's playbook. But the market didn't care. The stock didn't bounce on the GLM-5.3 release. That's the first red flag: when a company's own product launch fails to move the needle, the narrative is already priced.
Hedgeye's assessment was brutal: "MiniMax is neither the smartest nor the cheapest." That's the death sentence for a pure-play model company. You either lead on capability or dominate on cost. If you're stuck in the middle, you're not a company – you're a commodity with a ticker.
Core: The Order Flow Tells the Real Story
Let's look at the capital flows. Southbound money (mainland Chinese investors buying via Hong Kong Stock Connect) has been steadily accumulating: Zhipu AI now has ~12% held by Southbound, MiniMax ~8.1%. This looks like a vote of confidence. But the stock keeps falling. That means the selling pressure is overwhelming the buying. The shorts are not just betting against the company – they are betting against the "dumb money" that thinks it's catching a falling knife.
I've seen this pattern before. In DeFi Summer 2020, yield farmers piled into protocols with no moat. The insiders unlocked their tokens, the price collapsed, and the retail buyers who thought they were "accumulating" were actually providing exit liquidity. The same dynamic is playing out here. The IPO lock-up period ended in July, releasing 25.68 million shares of Zhipu AI and 150 million shares of MiniMax – a combined $11.5 billion in potential selling pressure. The shorts are positioning ahead of that wave.
But the real story is the unit economics. The article's analysis is spot-on: these companies are burning cash on compute, while API prices are in a freefall. The cost of inference is the single biggest line item. Zhipu AI claims a 19% cost advantage – but that's from engineering optimizations, not architectural innovation. Those optimizations can be replicated within months. Moonshot AI can match that cost with their next release. Then what? You're back to a race to zero, with no pricing power.
Contrarian: The Shorts Might Be Right, But They're Late
Here's the contrarian angle: the shorts are correct about the thesis, but the stock is already pricing in a lot of bad news. Zhipu AI is still up 800% from its IPO price. That's not a floor – that's a roof. But the decline from peak is 50%. The question is how much of the "model-as-a-business" risk is already discounted.
Sentiment buys the dip; data fills the position. The data says the market is still pricing these companies as if they will figure out monetization. The revenue multiples are not yet compressed to traditional SaaS levels. If the earnings report shows a widening loss, the next leg down could be 30-40%. But if – and it's a big if – either company shows a path to gross margin expansion, the short ratio becomes a bomb. 20% short interest on a stock that beats estimates can trigger a 50% rally in days. That's the symmetric risk the shorts are taking.
But I'm not betting on that miracle. The structural problem is that pure-play model companies have no defensible moat. The technology is open-source or quickly replicable. The distribution is controlled by big tech (Alibaba, Tencent, ByteDance) who can afford to subsidize their own models. The only escape is verticalization – build a killer application on top of your own model. But neither Zhipu nor MiniMax has shown that capability.
Takeaway: The Only Levels That Matter
I'm watching the August 26 and August 31 earnings like a hawk. The key metric is not revenue – it's gross margin and cash burn. If gross margin is below 30%, the business is unsustainable. If cash burn exceeds $200 million per quarter, the runway is less than 12 months. The shorts will be right. But if gross margin surprises to the upside, the short squeeze will be violent.
For now, the smart money is short. The data is clear. The question is not whether these companies are overvalued – it's whether the market has already priced in the worst. I don't think it has. Not yet. The lock-up overhang alone is enough to keep the pressure on through Q4. The takeaway is simple: don't catch this knife. Let the shorts fight it out. When the dust settles, the survivors will be the ones that built a moat, not a model.