The data is unambiguous. Short interest on MiniMax hit 20%. Zhipu AI's share price has collapsed over 50% from its peak. A combined $11.5 billion in unlocked shares is now overhanging the market. This is not a correction. This is a structural repricing of an entire business category.
Let me be precise about what I am auditing here. I am not looking at a token's liquidity pool or a smart contract's rebalancing logic. I am looking at the balance sheets of two of China's most prominent AI startups. The fundamentals are brutal.
The market is sending a clear signal: pure-play large language model companies, as a standalone business model, are facing an existential crisis of confidence. The narrative has shifted from 'technology moat' to 'profitability proof.' And the market is not buying the former without the latter.
Context: The Hang Seng's AI Reckoning
Zhipu AI and MiniMax both went public on the Hong Kong Stock Exchange to significant fanfare. They were positioned as the standard-bearers of China's AI ambitions, the local challengers to OpenAI and Anthropic. The IPO pricing reflected a 'story-driven' valuation, anchored in total addressable market projections and technological potential.
Zhipu AI's shares are still up over 800% from their IPO price. MiniMax is up over 80%. But both have been cut in half from their all-time highs. This divergence—massive gains from IPO, followed by a brutal drawdown—is the signature of a market that has repriced its expectations faster than the companies can execute.
Hedgeye, a well-known research and investment firm, has been vocal with its bearish thesis. They have labeled Zhipu AI as being under significant pressure from the ongoing price war in the AI model market. This is the core of the bear case: in a commodity market, pricing power evaporates. When you lose pricing power, your path to profitability becomes a function of cost-cutting, not revenue growth.
The lock-up expiry in July added a mechanical supply shock. Zhipu AI saw 25.68 million shares unlock, and MiniMax saw 150 million shares unlock. At prevailing prices, this represents a combined $11.5 billion in potential selling pressure. This is not abstract market anxiety. This is a concrete, quantifiable overhang that will cap any upside until it is absorbed.
Core: The Forensic Audit of the Business Model
The central question is not whether these companies have good technology. The question is whether 'good technology' in a competitive market is sufficient to generate sustainable profits.
Let's break down the order flow. The shorts are not emotional. They are data-driven. A 20% short interest on MiniMax is a staggering figure. It implies that one in every five shares available for trading has been sold short. This is a level of conviction typically reserved for companies with identified fraud or terminal business models.
The sell-side analysts are doing the math. Jefferies, in a note, pointed out that Zhipu AI's GLM-5.3 model offers similar performance to Kimi K3 at a 19% lower cost per task. On the surface, this is a bullish data point. It suggests engineering efficiency and a cost advantage in a price-sensitive market. But the market's reaction was muted. Why?
Because the market is not pricing technology. It is pricing the unit economics. A 19% cost advantage is irrelevant if the price war forces you to give that advantage away to win market share. The market sees this as a defensive metric, not an offensive weapon.
Based on my experience auditing yield strategies in DeFi, I see a direct parallel here. In the liquidity mining wars of 2020, protocols subsidized their TVL with token emissions. The APYs were high, but they were not sustainable. They were calculated subsidies, not organic growth. When the subsidies stopped, the users left. The market is now asking if Zhipu and MiniMax are doing the same with their model pricing. Are they buying market share with negative gross margins?
My audit of the situation reveals a company caught in a strategic pincer. They are simultaneously fighting a war on two fronts. On the high end, they face DeepSeek and Alibaba's Qwen, which have superior scale, ecosystem integration, and capital reserves. On the low end, they face a swarm of smaller, cheaper models that are 'good enough' for the majority of use cases. This is the 'middle ground' trap. They are neither the smartest nor the cheapest.
This is a brutal competitive position. In my trading framework, this is what I call a 'dead zone' asset. It has no clear technical edge to justify a premium, and no cost structure to win on price. It is simply squeezed from both sides.
Contrarian: The Retail vs. Smart Money Divide
The most interesting data point in this entire saga is the behavior of Southbound capital—the flow of funds from mainland Chinese investors into Hong Kong-listed stocks. They are accumulating. Zhipu AI's Southbound holdings are around 12%, and MiniMax's are around 8.1%.
This creates a classic market structure conflict. The professional short-sellers, who have done the forensic analysis of the business model, are selling. The retail-driven Southbound flows, which are often driven by thematic narratives and 'national champion' sentiment, are buying.
I have seen this movie before. In the crypto market, I saw it with the Terra/Luna collapse. There was a clear 'no algorithmic stablecoin' rule in my investment thesis, and I enforced it strictly despite massive FOMO pressure. The Southbound flows here are exhibiting similar FOMO behavior. They are buying the narrative of Chinese AI dominance, not the financials of the underlying business.
The contrarian angle is this: the shorts might be early, but they are not wrong. The Southbound capital might be providing a floor, but it is a floor made of sand. The moment the interim earnings report fails to show a clear path to profitability, that floor will wash away.
There is also a potential catalyst that the shorts may be underestimating: the possibility of consolidation. With market caps depressed, Zhipu and MiniMax could become acquisition targets. A large tech conglomerate or a state-backed entity might see value in their technology and talent. This is a real, quantifiable tail-risk for the short thesis. But it is not a tradeable thesis. It is a hope, not a strategy.
The Takeaway: The Earnings Report is the Liquidation Trigger
I do not trade on hope. I trade on data. The next major data point for this market is the interim earnings report. MiniMax reports on August 26th, and Zhipu AI reports on August 31st. This is the 'liquidation event' for this narrative.
The market is pricing in a worst-case scenario. If the earnings show better-than-feared revenue growth and a credible cost-control plan, we will see a violent short squeeze. The 20% short interest on MiniMax is a powder keg. If the earnings are even marginally positive, the covering bid will be ferocious.
However, if the earnings confirm the bear thesis—if revenue growth is decelerating, losses are widening, and there is no clear line of sight to breakeven—then the current price is not the bottom. It is simply the next rung on the ladder down. The $11.5 billion in unlocked shares will be sold into any rally.
My strategy for this is simple. Do not catch the falling knife. Let the market tell you what the earnings mean. Wait for the volatility to subside and the new trading range to establish. There is no alpha in guessing. There is only alpha in reacting to confirmed data.
The market is a discounting mechanism. It has already discounted the bad news. The question is whether the news is as bad as the discount. The earnings report is the only metric that matters. Verify the source, trust no one.
Volatility is the price of entry. But this is a volatility that can permanently impair capital. The risk-reward is only skewed to the upside if you are willing to wait for the dust to settle. Strategy beats speculation every time. Yields are calculated, not guaranteed.
I audit the code, not the charisma. And in this case, the 'code' of the business model is showing significant bugs. The smart money is betting on a crash. The retail money is betting on a story. History tells us that in the long run, the balance sheet always wins.
I will be watching the order flow on the earnings release. The liquidity dries up faster than hope. Be prepared for both scenarios. That is the only way to survive. Diversification is the only safety net.
