The record short selling against Hong Kong-listed AI giants MiniMax and Zhipu AI is not a market anomaly—it is a structural audit of a business model that has yet to prove its unit economics. MiniMax’s short ratio hit 20%, a level typically reserved for companies on the brink of insolvency, while Zhipu AI’s 6% ratio signals a deep, systemic skepticism. This is not a flash crash; it is a slow, deliberate unwinding of a narrative that conflated technological promise with financial viability.
Liquidity is a mirage; only settlement is real. The market is now asking: Can these pure AI model companies settle their debts to reality?
Context: The AI Hype Machine Meets Its First Real Test
MiniMax and Zhipu AI went public in Hong Kong during the peak of the AI narrative frenzy in late 2024. Their IPOs were oversubscribed, riding on the coattails of OpenAI’s success and China’s national AI ambitions. But beneath the surface, their business models were eerily similar to the DeFi protocols I audited in 2019: high total value locked (TVL) in user attention, but zero sustainable revenue mechanisms.
Both companies are pure-play large language model (LLM) providers. They generate revenue primarily through API calls and enterprise subscriptions. Yet the market is flooded with competing models—from Moonlight’s Kimi K3 to ByteDance’s Doubao—each offering similar capabilities at ever-lower prices. The result is a price war that erodes margins faster than any technological improvement can offset.
In July 2025, Moonlight released Kimi K3, a model that the market interpreted as a generational leap. Zhipu AI’s stock dropped 24% in a single week; MiniMax fell 18%. This was not a normal competitive response. It was a revaluation of the entire sector’s technological moat. The message was clear: the gap between first and second movers is widening, and being a “me too” model is no longer sufficient.
Core: The Structural Fragility of Pure LLM Economics
Based on my experience auditing DeFi liquidity pools during the 2019 bear market, I recognize a familiar pattern: the illusion of organic growth masking a fragile cost structure. The short sellers are not just betting on a price decline; they are betting on a fundamental failure of unit economics.
1. The Cost of Intelligence is Unsustainable
Zhipu AI’s GLM-5.3, released in response to Kimi K3, positions itself as “similar performance at 19% lower cost.” This is a classic follower strategy—acknowledging technological inferiority and pivoting to price competition. But in the AI world, cost advantages are fragile. Most inference optimizations (quantization, speculative decoding, batch processing) are engineering problems, not scientific breakthroughs. They can be replicated within months.
MiniMax faces an even worse fate: it is neither the smartest nor the cheapest. This “stuck in the middle” position is the death zone of competitive strategy. In my 2021 DeFi analysis, I identified the same pattern with yield farming protocols that offered neither the highest returns nor the lowest risk—they were systematically abandoned.
2. Lock-up Expiration: The Liquidity Tsunami
In July 2025, lock-up periods expired for both companies: 25.68 million shares for Zhipu AI and 150 million shares for MiniMax, totaling approximately $11.5 billion in market value at the time. Early investors—who bought at IPO prices still 800% below current levels—have every incentive to cash out. The stock price decline of over 50% from peak has not deterred selling; it has only accelerated the rush to exit before further losses.
This is a classic liquidity drain. The market is absorbing the supply, but at a steep discount. Southbound capital (mainland Chinese investors) has been buying the dip, increasing holdings to 12% and 8.1% respectively, but their buying power is insufficient to offset the institutional selling wave. This is reminiscent of the 2022 Terra collapse, where “buy the dip” mentality masked a fundamental insolvency.
3. The Earnings Report Catalyst
Both companies will report their first-half 2025 earnings on August 26 (MiniMax) and August 31 (Zhipu AI). Short sellers are increasing their positions ahead of these releases, expecting the numbers to reveal the grim reality: revenue growth slowing, gross margins contracting, and cash burn accelerating.
Based on my analysis of comparable AI companies, the path to profitability for pure LLM providers requires either: - A 10x improvement in model efficiency (to lower inference costs), or - A 5x increase in API pricing power (which is impossible in a price war).
Neither is happening. The short thesis is rational.
Contrarian: The Decoupling That Could Surprise Everyone
But here is where the narrative gets interesting. The market is pricing these companies as if they are doomed to fail. However, the short ratio of 20% for MiniMax also creates a classic squeeze risk. If the earnings report shows even a modest beat—say, revenue growth above expectations or a credible path to profitability—the shorts may be forced to cover, triggering a sharp rally.
More importantly, the AI sector is not monolithic. The current bearishness may be over-discounting the long-term value of AI infrastructure. Just as the crypto market in 2023 was prematurely written off before the ETF-driven rally, the AI market may be experiencing a similar cycle of narrative fatigue.
Moreover, the decoupling thesis is worth considering: AI companies are now trading on their own fundamentals, not on macro liquidity. This is a healthy correction, not a death spiral. The market is demanding that companies prove they can generate cash flow, not just buzz. This is the same transition that the crypto industry underwent in 2022-2023, when projects with real utility survived and those with only hype perished.
Another blind spot: the short sellers may be underestimating the strategic value of these companies to Chinese state-backed initiatives. The government’s push for AI sovereignty could lead to preferential procurement contracts, subsidized compute, or even a merger with state-owned enterprises. Such a scenario would invalidate the pure financial analysis.
Takeaway: The Settlement is Coming
The final verdict on MiniMax and Zhipu AI will not come from the short sellers or the bulls. It will come from the earnings reports. Those reports will settle the question of whether these companies are building real businesses or just riding a narrative wave.
As I wrote in my 2024 paper on institutional friction in crypto markets: “Price is a conversation; settlement is a conclusion.” The market is now holding its breath, waiting for the conclusion.
For investors, the lesson is clear: do not confuse liquidity with value. The short sellers are not the enemy; they are the auditors. Listen to them, but also watch for the moment when the audited entity finally reveals its true financial health. That moment, for AI giants, is now.