Short Positions in Chinese AI Model Stocks Hit Record Highs as Intensifying Competition Raises Valuation Concerns
Nashnova编辑部
Short interest in MiniMax and Zhipu AI (智谱AI) has hit all-time highs — 20% and 6% of free float respectively — as a price war and rising rivals force the market to ask how long the 'pure large-model stock' scarcity premium can last.
How high has the short interest climbed?
S&P Global data show MiniMax short interest at 20% of free float and Zhipu AI at 6% — both records since their January IPOs.
The two are the world's only pure-play large-model listed companies. Since listing, Zhipu AI is still up over 800% and MiniMax over 80%.
This means → even with hefty post-IPO gains, both stocks have more than halved from their peaks, and shorts are betting the decline is not over.
Who is squeezing their space?
Alibaba, DeepSeek, and Moonshot AI (月之暗面) are pressing from both the pricing and the technology end simultaneously.
Two events triggered the recent sell-off: MiniMax slashed M3 model pricing in June; Moonshot released Kimi K3, a direct challenger to Zhipu AI's GLM series.
In plain terms = rivals are either cheaper or stronger — the "pure large-model" scarcity label is fading.
Why didn't Zhipu AI's new model lift the stock?
Zhipu AI recently launched GLM-5.3. Jefferies analysts note it matches Kimi K3 on performance and costs roughly 19% less per task.
Yet the market has not priced in that cost edge — the stock barely moved after the release.
This reflects a market more worried about price-war margin erosion than about raw model benchmarks.
What do hedge funds and asset managers think?
Felix Wang, tech-sector head at Hedgeye Risk Management, says the price war is eroding Zhipu AI's ability to raise prices. His verdict on MiniMax: the product is "stuck in the middle — neither the smartest model nor the cheapest."
David Choa, Greater China equity head at BNP Paribas Asset Management, adds: "Low-cost Chinese open-source models are democratizing frontier AI globally — that could challenge the assumption that any single model commands scarcity value."
This means → the professional investor's core doubt is not "is the model good enough?" but where is the commercial moat?
Are mainland funds buying the dip? Is it working?
Zhipu AI joined Stock Connect less than three months ago; mainland ownership has already reached 12%. MiniMax joined just two weeks ago and mainland holdings are already at 8.1%.
Yet that buying has not driven a rebound so far.
In plain terms = mainland retail is stepping in, but not fast enough to offset short-seller and foreign institutional selling.
What are the next key moments?
MiniMax reports earnings this Wednesday; Zhipu AI follows next Monday — the first time both companies will show their fundamentals mid-price-war.
Consensus expects Zhipu AI's first-half revenue to grow roughly 153% half-on-half, but adjusted net losses are forecast to widen further.
DeepSeek surprised the market this month by raising its model pricing, though the impact on industry-wide profitability remains unclear. This means → the earnings window will test whether the market still accepts a "revenue up, profits down" story.
Content is for reference only, not financial advice.