Hong Kong's AI Giants Under Siege: Record Short Interest Exposes the 'Capability-Cost' Trap

The protocol does not lie; the interface does. In Hong Kong's equity markets, the interface is the price chart, and it has been speaking with unusual clarity about China's pure-play AI model companies. Record short selling has hit two of the most prominent names: MiniMax and Zhipu AI. MiniMax's short ratio has reached 20%, a level that signals not merely skepticism, but outright conviction in a thesis of decline. Zhipu AI sits at approximately 6%, still elevated, still telling. To own the chain is to own the history. To own the narrative is to understand the asymmetry between perception and fundamentals. The trigger for this bearish surge is not a scandal or a regulatory crackdown. It is a single event: the release of Kimi K3. The market watched the launch, read the benchmarks, and then re-priced the entire competitive landscape of China's pure-play large language model (LLM) sector. The response was violent. Zhipu AI's stock fell approximately 24% in the days following. MiniMax dropped 18%. A single product launch moved billions in market capitalization. That is the speed at which technical reality is now being discounted. The core issue is not short-term earnings. It is the viability of the pure-play model company itself as a business species. Kimi K3 is the catalyst, but the infection runs deeper. The market is finally asking a question that has been deferred for three years: Can a company whose sole output is a general-purpose artificial intelligence model achieve sustainable unit economics in a market defined by price wars and rapid technological iteration? The answer, based on current price action and short-seller conviction, is increasingly a resounding no. The data from S&P Global highlights that short interest is not just high; it is concentrated. MiniMax's 20% short ratio is exceptional. It is a level reserved for companies expected to miss, to falter, or to fail. This is not a hedge against a normal pullback; it is an outright position against the company's ability to deliver on its interim financial results. The shorts are building their positions ahead of the earnings reports, which are scheduled for late August. The signal is unambiguous: the market expects the financials to expose a business model that spends aggressively on compute and sales, yet cannot command pricing power for its API output. Zhipu AI's strategy, as articulated by its positioning and third-party analysis from Jefferies, is to differentiate on cost. Their GLM-5.3 model is positioned as having "similar performance" to the market leader but at a 19% lower cost per task. This is a classic follower's strategy. It admits, tacitly, that the company cannot win a pure capability race against Kimi K3. So, it retreats to the cost curve. The problem is that in a market driven by technical moats, a cost advantage based on engineering is not a durable moat. It is a target. If Kimi K3's architecture is fundamentally more efficient, or if the leading company simply optimizes its inference stack, Zhipu AI's 19% cost advantage evaporates within a quarter. MiniMax faces a more existential crisis. Hedgeye's assessment was brutally concise: MiniMax is "neither the smartest nor the cheapest." That is the worst possible position in a technology market. It is the "stuck in the middle" trap. The company lacks the technology premium to justify higher prices, and it lacks the scale or efficiency to win on price. Its product is a commodity that is worse in both dimensions. This is a strategic dead end. The short sellers understand this. They see a company with no clear path to differentiation and a high burn rate. The situation is dire. The market's collective skepticism has also put the broader valuation logic under scrutiny. Both companies are down over 50% from their peaks, yet Zhipu AI's stock price remains roughly 800% above its initial public offering price. The IPO priced in a future that now appears uncertain. The unlock of lock-up periods in July, with approximately 1.5 billion shares for MiniMax and 25.68 million for Zhipu AI, created an overhang of roughly 115 billion USD of potential selling pressure. Early investors, despite the drawdown, are still sitting on enormous paper gains and have every incentive to realize them. The southbound capital flows have provided some support, with mainland investors holding about 12% of Zhipu AI and 8.1% of MiniMax. Yet, this has not been able to stem the tide. This suggests that the buying is likely strategic positioning or "bottom-fishing" that is not based on an improvement in fundamentals. It is catching a falling knife. The selling pressure is simply overwhelming the buying. Certainty is a bug in a stochastic world. The narrative of China's AI race has a nuance that is often lost in the narrative. The arrival of Kimi K3 is not just an incremental improvement. It is a potential "generational leap" that repositions the entire value chain. This creates a clear tiering in the market. The capability gap is now a pricing signal. The market is betting that this gap will not be closed in the near term. A contrarian angle lies in the structure of the short interest itself. A 20% short ratio is a loaded spring. If the upcoming earnings reports do not confirm the bearish thesis, the subsequent short squeeze could be dramatic. The stock could rally sharply, as shorts are forced to cover. However, this is a trading opportunity, not an investment thesis. It is a bet on the timing of a reversal, not the sustainability of a business model. The real issue is the cost of the compute. The price war is not just about how many parameters a model can handle. The cost of serving a single query is the bottom line. The chip restrictions, supply chain constraints, and the fundamental high cost of compute in China mean that every percentage point of cost reduction is hard-won. Zhipu AI's efficiency may be a temporary engineering advantage. The question is whether these companies can solve the algorithm to be efficient. If they cannot, they will be perpetually at the mercy of the market's pricing power. We build in the dark to light the public square. The public square of AI is currently lit by the red glow of sell orders. The interim results on August 26 and August 31 will not just determine the trajectory of these two companies. They will validate or invalidate the entire concept of a pure-play AI model company. The market is pricing in a failure of the current business model. The only question is what the future looks like. The only way to survive is to transition from a model provider to a comprehensive solution provider. That is a strategic pivot of a decade's length. The shorts are betting they cannot do it. The earnings will be the first proof of who is right.