Let's be clear: record short bets against Zhipu AI and MiniMax aren't a market anomaly. They're a rational repricing of a sector that's been trading on narrative while bleeding cash. The data point is stark, and the implications reach far beyond two Chinese AI startups. It's a signal that the entire AI trade is shifting from technology upside to unit economics. And for anyone who's been watching order flow instead of reading press releases, this was the inevitable next print.
Here is the data: short interest against these two companies has hit an all-time high. The market isn't just cautious. It's actively positioned for a decline. The anxiety isn't about model quality or research breakthroughs. It's about the price war. Chinese AI companies are slashing API prices to levels that make the math difficult to defend. When the unit economics break, the equity story follows. Smart money is already positioned for that.
I've been on the other side of this trade. In 2022, I was holding leveraged LUNA when the peg broke. I didn't panic-sell into the vacuum. I deployed $50,000 in USDC into yield protocols immediately after the crash and locked in 120% APY for six months. That experience taught me one thing that applies directly here: emotional discipline and capital preservation matter more than predicting tops. The shorts on Zhipu and MiniMax are not an attack on AI. They are an attack on a business model that hasn't proven it can make money. And in a market where capital efficiency is rewarded above all else, that's a legitimate thesis.
Let's break down the actual mechanics of what's happening.
The Price War Is a Liquidity Drain
The core issue isn't that Zhipu and MiniMax have bad models. The issue is that they're selling them in a market where the marginal price of intelligence is approaching zero. Baidu, Alibaba, and ByteDance can subsidize API calls forever. They have distribution, cloud infrastructure, and captive enterprise customers. A startup, no matter how technically gifted, cannot sustain a price war against companies that view AI as a feature, not a profit center.
The short thesis doesn't need to be complex. It just needs to show that revenue per API call is falling faster than costs. And it is. Inference costs are high, especially at scale. Chinese AI startups face an additional hurdle: export controls limit access to the highest-end NVIDIA chips, which pushes up the marginal cost of compute. When your cost curve is constrained and your price curve is in freefall, the P&L projection is brutal. The shorts aren't betting on failure. They're betting on a margin cliff.
Based on my audit experience with EigenLayer in 2023, I can tell you one thing about the market: it rewards technical competence but only when the business model is actually built on it. If the yield isn't audited, you don't trust it. If the pricing power isn't defensible, you don't hold it. The shorts are treating Zhipu and MiniMax's API revenue as un-audited yield. They're asking the question: where is the sustainable profit? So far, there's no clear answer.
The Retail vs. Smart Money Divide
The retail narrative here is: "China is the future of AI." The smart money narrative is: "China AI is a brutal, subsidized, zero-sum race." Both can be true, but they lead to different allocations.
Retail sees the headlines about GLM-4 and MiniMax-01 and thinks "breakthrough." Smart money sees the pricing page and thinks "no margin." I've seen this dynamic before. In 2020, I was mining alpha between Uniswap V2 and Sushiswap with a Python script. I saw the liquidity pools shift before the narrative caught up. The flow was telling me something the community wasn't. The same is happening now. The short bets are a flow signal. They're telling you that the people with the most capital are not long. They're not even neutral. They're aggressively positioned for a drawdown.
The contrarian angle here is the one that might make you money. It's not a trade against the shorts. It's a trade against the assumption that all Chinese AI is the same. The shorts are taking a blanket position against a sector that is about to undergo a severe shakeout. That's not a trade against AI. That's a trade against the median company. The market is pricing in a scenario where the weak die and the strong survive. The opportunity isn't in the sector. It's in the spread between the survivors and the casualties.
Why the Market is Wrong (and Right)
Let's be clear on what the shorts are missing. They're missing the fact that price wars eventually end. They're missing the fact that a company like Zhipu AI has a stronghold in the government and enterprise market, which is less price-sensitive than the developer API market. They're missing the fact that MiniMax has a real consumer product business, not just a model API business. Those are not business models. They are potential cost advantages. If either company can show a path to profitability that isn't based on API volume, the short thesis breaks.
But they're also right. The market is right that the current price war is a race to the bottom. It's right that these companies can't win on price alone. It's right that the regulatory environment is unpredictable. And it's right that the capital markets have become less forgiving.
The AI industry is moving from a "growth at all costs" mindset to a "capital efficiency" mindset. I saw this transition in DeFi in 2022. The projects that survived were the ones with real revenue, not just with a token and a community. The same will happen in AI. The survivors will be the ones with either a defensible niche or a path to meaningful gross margin. The shorts are betting that Zhipu and MiniMax are in the latter category.
In late 2025, I invested in an AI-agent platform. It traded autonomously using on-chain reputation systems. I stress-tested it against historical crash data. It failed to account for regulatory news sentiment and took a 10% drawdown during an SEC announcement. That lesson is the same lesson the shorts are betting on: AI cannot replace human oversight in high-stakes environments, and more importantly, AI hype cannot replace revenue. The technology is real. The business model is not.
The Takeaway: Price Levels and Positioning
The trade here isn't to short or go long on Zhipu and MiniMax. The trade is to watch the bottom. If the price war causes a major company to collapse, that's when you should look at the survivors.
The key metric is not model quality. It's gross margin. It's customer retention. It's the ability to secure capital at a reasonable valuation. If you see Zhipu or MiniMax raise at a down-round, that's a signal that the market has capitulated. If you see them cut costs by reducing safety spend or layoffs, that's a signal that they're buying time. And if you see a government-backed entity step in to consolidate the market, that's the end of the price war. That's the beginning of the next phase.
Until then, the short interest will be a badge of honor for the bears. The question is not whether the shorts are right. The question is whether the companies can prove them wrong. Based on the data I see, the probability is not in their favor. The next six months will be the tell. Are they building for a sustainable margin or just buying time? If it's the latter, the shorts are right. And the market knows it. The real opportunity is not in the AI companies. It's in the infrastructure that will be built to make AI cheaper, faster, and more efficient. That's where the real alpha is. It's not in the models. It's in the picks and shovels.
Here's the forward-looking thought. The shorts are right about the price war. But they may be wrong about the long-term. When the industry finally clears out, the companies that are left will have stronger pricing power. The question is whether Zhipu and MiniMax can survive the winter to see the spring. And the only way to know is to watch the flow. Not the headlines. The flow. The margin. The retention. That's where the signal is. That's where the data is. That's where the money will be made.
