The number hit my screen like a cold splash of water. MiniMax at 20% short interest. Not 5%. Not 10%. Twenty percent of the float, bet against the machine. Zhipu AI wasn't far behind, with the narrative turning as fast as a flash crash. This isn't a dip. This is a coordinated market verdict. The AI gold rush narrative has hit the wall of arithmetic, and the market is now doing the math out loud.
Over the past few weeks, we've seen the dream of the "pure-play AI model" company get priced for perfection and then thrown into the discount bin. The numbers are brutal. Zhipu and MiniMax have both more than halved from their peaks. This isn't a correction; it's a repricing. The question isn't just "Are these stocks cheap?" The real question is, "Does the business model even work?"
You need to understand the backdrop here. For the last 24 months, the game in AI was simple: raise money, buy GPUs, train a bigger model, and watch the valuation multiple expand. It worked. Until it didn't. The shift happened when the market realized that the moat is not the model; the moat is the distribution, the data, and the cost of serving the inference. When a junior analyst at a short fund like Hedgeye can publish a note that moves the price more than a model launch, you know the technical narrative is dead.
Let's get into the meat of it. We have two companies, both in that awkward "middle ground" of the AI landscape. Not the smartest. Not the cheapest. They are in the dead zone. The market is not kind to middle children. I've seen this playbook before in crypto. The layer-one wars. The liquidity wars. The same dynamic: when the tide goes out, you see who's wearing the swimming trunks. Here, the tide is the short wave.
The Governance of Profit
The core technical narrative that used to work is now broken. Let's look at the data. In July, Kimi K3 dropped. It was a solid upgrade, but the immediate market reaction wasn't to bid up the leaders; it was to hammer the followers. Zhipu's GLM-5.3 is roughly on par with K3, and Jefferies even pointed out that Zhipu's cost per task is 19% lower than Kimi's. That's a technical efficiency win. Yet, the stock price didn't just stay flat—it got wrecked.
Why? Because in a bear market, the market reads "efficiency" as "price war." If you are cheaper, the market doesn't think you are a winner; it thinks you are going to have to cut margins to survive. The concept of technical edge is now a liability. Speed is the only currency that never inflates. But in this cycle, speed of execution isn't enough. You need speed of revenue.
I don't predict the market; I ride its heartbeat. And the heartbeat right now is saying that the AI inference market is becoming a commodity. Commodities are driven by the lowest-cost producer, not the best engineer.
The Unlocking of the Bag
Let's talk about the elephant in the room: the lockup expiries. In July, the IPO lockups expired for both Zhipu and MiniMax. We are talking about a combined 115 billion USD in shares that can now be sold. That is not a trickle; it's a dam. The logic is simple: early investors and founders have been waiting for months to take profits off the table. Why hold a paper asset in a declining market when you can dump it on the secondary market and diversify? That's the psychological game of the short seller.
I've seen this exact playbook in crypto—the token unlocks that crush prices for months. The difference here is that it's not a decentralized protocol; it's a centralized business. The insiders have the keys. And the shorts know it. They are betting on the inevitable, the overhang of supply.
The shorts are also reading the flow. The Southbound capital is buying. Zhipu is up about 12% of Southbound holdings, and MiniMax is about 8.1%. But it's not helping. When the local capital cannot halt the slide, it tells you that the outflows from the big institutions are more significant. The whales are exiting the pool.
The Contrarian Play: The Cost of Being the "Cheap"
Here is the angle that most media is missing. The market is not just betting on the business failing. They are betting on the fragmentation of the AI ecosystem. The idea of a "pure-play" model is dying. The model layer is being embedded into every product, every app, and every OS. The application is the token, and the model is the gas.
But I think this is a manufactured narrative. Just like the liquidity fragmentation in DeFi, the market is scared of a problem that is, in reality, a feature. The fragmentation is the edge. Let me break it down. In DeFi, we spent years thinking that having liquidity on one chain was the only way to win. Then we realized that aggregators made the fragmentation work. The same is true here. The AI ecosystem is not about the model; it's about the orchestration.
If you are a dev, you don't care if it's Kimi or GLM; you care about the API latency and the cost. The API is the new trading venue. The aggregator is the new bridge. The current market is pricing Zhipu and MiniMax as if they are isolated islands. That's the blind spot.
Look at the cost advantage. GLM-5.3 costs 19% less per task. In a bear market, the cheap winner takes all. The high-end performance is a luxury that companies cut in Q3. The cheap tier is a necessity. If the market doesn't see this, then the market is wrong. But the market can stay wrong longer than you can stay solvent. That's the reality of the short.
The Trial of the August Earnings
So, we are entering the most critical phase. The earnings are coming in late August. On August 26, MiniMax drops its report; Zhipu follows on August 31. This is the Super Bowl. This is the moment where the narrative meets the data. The market is not asking for a 50% growth rate; the market is asking for a path to profitability. If they show a 40% growth but a widening loss, the market will punish. If they show 20% growth but a 10% reduction in losses, the market might bid it up.
I'm going to watch the gross margin. In the AI inference, the gross margin is the current heartbeat. If the gross margin is declining because of the price war, the stock will continue to bleed. If it's flat, that's a signal that the market is the leader. The revenue quality is also important. If the revenue is coming from a few enterprise deals, that's a sign of weakness. If it's coming from many small developers, that's a sign of platform strength.
The shorts have a clear thesis. They believe the revenue is not recurring. They believe the customer churn is high. They believe the model is not differentiated. The long thesis is that the market is throwing the baby out with the bathwater. The unlock is a one-time event. The shorts are over-exposed. A squeeze is possible.
But let's be honest. The headwinds are real. The funding for the AI arms race is slowing. The venture money is not flowing into the pure-play as much. The cost of training is not falling fast enough. The GPU is still the bottleneck, and the price is not coming down.
The price war is the core. We saw in the crypto world that the game is about the layer 2. The layer 2 is about the cost per transaction. In AI, the cost per token is the new unit. If the cost per token drops too fast, the revenue can't keep up. If the cost per token drops because of efficiency, that's a win. The short sellers are betting that the cost drop is not the engineering efficiency but the desperation.
The Takeaway
The short is the market's way of saying "Show me the money." The AI model layer is entering the "B2B SaaS" era, and that is a brutal market. The narrative is no longer the star; the unit economics is the king.
I don't predict the market; I ride its heartbeat. And the heartbeat is telling me to wait for the August 26 data. If the data shows a positive gross margin trend, we have a classic "trap." If the data shows a blow-out loss, we have a confirmation of the short thesis.
The real story is not the short. The real story is the refocus of the value. The value is shifting from the model itself to the infrastructure around it. The GPU cloud, the data pipeline, the enterprise distribution. Those are the true winners in the bear market.
As for the losers, they will be the ones who can't pivot. Pivot or perish. The market doesn't wait. The next few weeks will tell us which one is which. Keep your eye on the volume. The whisper is turning into a roar. The question is: which side is going to make the move?
