Record short interest. Lock-up expirations totaling $11.5 billion. Share prices cut in half from their peaks. The market isn't just questioning MiniMax and Zhipu AI β it's actively betting against their survival.
The numbers are brutal. MiniMax's short interest sits at 20%, a level that screams institutional conviction in downside. Zhipu's stock has collapsed over 50% from its high, despite still trading 800% above its IPO price. This isn't a correction. It's a paradigm shift in how the market values pure-play AI model companies.
Speed was the only asset that didn't depreciate in this cycle. But these two companies are discovering that being fast to market means nothing when you're stuck in the middle of the AI food chain.
The Narrative Has Flipped
For two years, the market rewarded AI companies for one thing: intelligence. Model benchmarks were the currency, and every new release was supposed to trigger a repricing. That era ended abruptly.
When Kimi K3 dropped in July, the market's reaction was telling. Zhipu's stock fell 24%. MiniMax fell 18%. A flagship model release β the kind of event that used to spark rallies β became a sell signal. Why? Because investors read it as confirmation that the AI arms race is accelerating, which means one thing: more capex, thinner margins, and a longer road to profitability.
Jefferies' analysis revealed the uncomfortable truth. Zhipu's GLM-5.3 matches Kimi K3's performance while costing 19% less per task. That's a genuine engineering achievement. The market's response? Indifference. Volume tells the truth when price tries to lie, and the price action here says efficiency doesn't matter when the entire sector is being repriced on profitability timelines.
Based on my experience auditing trading pairs and liquidity dynamics, I've seen this pattern before. When a sector transitions from growth-stage pricing to margin-stage pricing, the first casualties are always the companies without a structural cost advantage or a defensible ecosystem. Technical parity becomes a liability, not an asset.
The Middle Ground Is a Graveyard
Hedgeye's assessment of MiniMax cut through the noise: "neither the smartest nor the cheapest." That's the kiss of death in this market.
Look at the competitive landscape. DeepSeek has captured the value narrative with aggressive pricing that redefined market expectations. Alibaba's Qwen and ByteDance's Doubao leverage massive ecosystems β compute, data, distribution channels β that pure-play model companies simply can't replicate. Zhipu and MiniMax are caught in a pincer movement: outgunned on technology by the top tier, outmaneuvered on price by leaner competitors.
This isn't a temporary positioning problem. It's structural.
Arbitrage isn't just about price differences β it's the market correcting its own soul. And the market is telling us these companies have no moat. Their technology is replicable, their pricing power is nil, and their path to profitability is obscured by the very arms race they must win to stay relevant.
The lock-up expirations add another layer of pressure. Zhipu's 25.68 million shares and MiniMax's 150 million shares became tradeable after the July IPO lock-up period ended. Combined, that's roughly $11.5 billion in potential selling pressure. Early investors who've held through the hype cycle have every incentive to take profits now. The question isn't whether they'll sell β it's how quickly.
The Southbound Capital Trap
Southbound capital has been buying the dip. Zhipu's holdings sit around 12%, MiniMax's around 8.1%. On paper, this looks like conviction. In practice, it's starting to look like a value trap.
Mainland investors are applying the logic that worked in previous tech cycles: buy the fear, hold for the recovery. But AI model companies don't follow the same playbook as semiconductor manufacturers or internet platforms. The cost structure is different. The competitive dynamics are different. The regulatory environment is different.
Efficiency is the price we pay for speed, and these companies paid for speed with efficiency. Their burn rates are unsustainable without either dramatic revenue growth or continued capital injections. The stock market has closed that second door. Now they're betting everything on the first.
The earnings reports will be the moment of truth. MiniMax reports on August 26, Zhipu on August 31. The market is pricing in disappointment. The question is whether these companies can surprise to the upside on the metrics that matter: revenue growth, gross margins, customer acquisition costs, and retention rates.
Here's what I'll be watching. Revenue concentration β if either company derives a disproportionate share of revenue from a handful of enterprise clients, that's a red flag. R&D efficiency β can they demonstrate that each dollar of R&D produces measurable capability gains? And pricing power β can they hold API prices stable without losing market share?
The Real Bear Case Nobody's Talking About
The short sellers are focused on the obvious: valuation, competition, profitability. But the deeper problem is technological commoditization.
When model performance converges β and it's converging faster than anyone expected β the model itself becomes a commodity. The value shifts to distribution, data moats, and application ecosystems. Zhipu and MiniMax don't have those advantages. They're selling a product that's becoming indistinguishable from what their competitors offer, at prices that don't reflect their cost structures.
Survival is a strategy, but leverage is a mindset. These companies are leveraged to a narrative that's collapsing β the idea that model quality alone can sustain premium valuations. They need to pivot toward something defensible: proprietary data sets, vertical-specific solutions, or embedded distribution channels. Without that pivot, they'll remain at the mercy of the short sellers who've already identified the weakness.
There's also the acquisition angle. At these valuations, both companies become potential targets. A strategic buyer with deep pockets and existing infrastructure β a major cloud provider, a large internet platform β could absorb the technology and talent at a discount. That's the bull case that nobody's pricing in, and it's the one scenario that could trigger a violent short squeeze.
But waiting for a white knight is not a strategy. It's hope dressed up as planning.
The Verdict Is Pending
The next two weeks will define the near-term trajectory for both stocks. The earnings reports won't just be about financial results β they'll be about credibility. Can these companies articulate a realistic path to profitability that doesn't require winning an arms race they can't afford?
We didn't get into this business to watch value destruction. But the market is sending a clear signal: pure-play model companies without structural advantages are being repriced for a reality where most of them won't survive independently.
The shorts are bold. The longs are hopeful. The data is unforgiving. The only question that matters is whether the earnings reports can change the narrative before it's too late.
In this market, speed matters. But it's not the speed of innovation that counts anymore. It's the speed of adaptation. Zhipu and MiniMax need to show they can evolve faster than the market's skepticism. The clock is ticking.