The Short-Seller's Verdict: Hong Kong's AI Giants and the Liquidity Mirage

The numbers arrive with the cold finality of a ledger entry. MiniMax, the ambitious large-model player that IPO'd with considerable fanfare, now carries a short ratio of 20%. Zhipu AI, its more established rival, sits at 6%. These are not arbitrary figures. A short ratio above 10% is a declaration of war by the market. At 20%, it is a rout. The bears have not merely circled; they have entrenched, and they are waiting for the August 26th and 31st earnings reports with the patience of auditors examining a suspicious balance sheet. To understand this, we must strip away the noise of the AI narrative and examine the structural mechanics at play. The proximate cause for the recent collapse in both stocks—Zhipu down 24%, MiniMax down 18%—is the release of Kimi K3 by Moonshot AI. This was not an incremental update. The market priced it as a generational leap. It is the equivalent of a new settlement layer appearing that renders the existing infrastructure obsolete. The immediate, visceral repricing of competitors is the market's way of confirming that the gap in model capability is now a canyon, not a crack. My own work on liquidity pools has shown me that capital is a coward; it flees to the safest, most certain yield. The same principle applies here. When a technology gap widens, capital doesn't wait for a turnaround; it short the laggards. Jefferies' assessment of Zhipu's GLM-5.3 as "similar performance, 19% lower cost" is a damning indictment masquerading as a silver lining. It is an admission of a follower strategy. In a hyper-competitive market, a 19% cost advantage in inference is a razor-thin moat, especially when the leader—Moonshot AI—can simply optimize its own engineering to close that gap in a single quarter. And Hedgeye's characterization of MiniMax as "neither the smartest nor the cheapest" is the most dangerous position in any market. It is the technological equivalent of being stuck in the middle of a liquidity crunch—unable to attract the premium flows and too illiquid to survive the outflows. This is not a pricing error; it is a structural deficit. This brings us to the core issue that the market is beginning to price with brutal honesty: the viability of the pure-play large-model company. In the blockchain world, we speak of the 'fat protocol' thesis. Here, we have a 'fat model' thesis, and it is failing. The fundamental unit economics are broken. The price war has decimated the potential for margin expansion, and a model that is merely 'similar' or 'adequate' has no pricing power. The 800% gain of Zhipu's stock from its IPO price is a historical anomaly, a remnant of a narrative-driven era. The market is now re-pricing these assets from a 'price-to-dreams' ratio to a 'price-to-earnings' ratio, and the earnings are not there. This is the rebalancing of the ledger. My own experience auditing liquidity in 2019 showed me that 80% of the 'value' was fleeting manipulation, not real economic flow. The same phenomenon is happening here. The Southbound capital flows—Zhipu at 12%, MiniMax at 8.1%—are not a vote of confidence; they are often a 'catch-the-falling-knife' strategy that provides liquidity to the exits of early investors. The lock-up expiry in July, releasing shares worth approximately $11.5 billion, created a supply wall that the short-sellers are using as a ladder. The shorts are not betting on poor technology; they are betting on poor economics. The high short interest also creates a potential for a 'short squeeze', a violent, unanticipated rally if the earnings are not catastrophic. But that is a trader's gamble, not an investor's thesis. The contrarian angle is that the market may be mispricing the nature of the cost advantage. My analysis suggests that Zhipu's 19% cost efficiency likely stems from engineering—quantization, speculative sampling, batch optimization—not architectural breakthroughs. This is a replicable advantage. Moonshot AI's K3, however, may be a structural, architectural edge. That difference is the key to understanding whether the gap converges or widens. If the edge is scientific, it is a defensible moat; if it is just engineering, it is a speed bump. The shorts are betting on the former, but the true signal will be in the earnings call. The question of whether pure-model companies can be profitable is not just about their current balance sheets; it is about their ability to transition from a technology narrative to a financial engineering narrative. Liquidity is a mirage; only settlement is real. The settlement date is August 26th and 31st. Until then, the market is trading on hope, and the margin is thin.