We didn’t expect the reckoning to come from a Hong Kong trading floor, but it did. On August 24, two of China’s most-hyped AI startups—Zhipu and MINIMAX—saw their shares (or their shadow stocks) tumble over 10% in a single session. The headlines read like a bear market déjà vu, and for a moment, I felt the same chill I did back in 2018 when the ICO bubble burst. But this time, the crash wasn’t about a stolen private key or a failed smart contract. It was about something far more mundane: the market finally asking, “Where’s the revenue?”
As someone who spent years in the crypto trenches—from auditing genesis blocks to watching my own savings vaporize in a DeFi exploit—I’ve learned that the market’s greatest lies are often the ones we tell ourselves. The AI stock drop isn’t just a story about overvalued chatbots. It’s a mirror held up to the crypto space, reflecting our own unresolved tensions between hype and substance, centralization and trust, and the uncomfortable truth that code alone doesn’t guarantee value.
Let me set the stage. Zhipu (developer of the GLM series) and MINIMAX (creator of the abab models) are part of China’s “AI Dragon” quartet, each valued at billions of dollars. Their stock decline, reported by Bitget market data, came amid a broader tech selloff, but the magnitude—11% and 10% respectively—signaled something deeper. The news was sparse, offering no technical failures or regulatory crackdowns. Just price action. In crypto, we call that a “rug pull” without the rug. But here, the rug was woven from expectations, not code.
Truth in blockchain isn’t found in whitepapers or Medium posts; it’s etched into smart contracts and transaction histories. The AI stock crash, in contrast, happened behind closed doors. We don’t know the exact sell orders, the margin calls, or the institutional whispers. All we see is the aftermath. This opacity is the exact disease that blockchain was invented to cure—yet here we are, watching the same pattern play out in the “real” economy.
The core of the problem is valuation divorced from on-chain reality. In crypto, we have a crude but effective balancer: token price reflects both speculation and utility, but at least we can track wallet activity, TVL, and daily active addresses. For AI startups, the metrics are softer—API calls behind corporate firewalls, contract wins that may never materialize, and a pricing model that’s being crushed by giants like Baidu and ByteDance. The 2024 AI price war, where some models dropped costs by over 90%, is eerily reminiscent of the DeFi Summer fee compression. Only here, the startups don’t have a token to incentivize liquidity or lock in users. They only have equity, and equity is a blunt instrument.
I’ve seen this movie before. In 2020, I poured my savings into a yield farming protocol that promised 1,000% APY. The smart contract was unaudited, the team was pseudonymous, and the code was a fork of a fork. It took 48 hours for the exploit to drain everything. Looking back, the warning signs were all there—but I was blinded by the narrative. The AI stock crash feels the same. The narrative is “AI will change everything,” and it might, but the market is now asking: “Will it change this quarter?” And the answer is a sobering “not yet.”
From a technical perspective, both AI and crypto suffer from centralization of control. Zhipu’s model architecture is closed-source; MINIMAX’s MoE design is proprietary. The code is locked behind corporate walls. In blockchain, we fight this with open-source ethos and decentralized governance, but we’re not innocent either. Layer2 sequencers, as I’ve written before, are essentially single points of failure. The “decentralized sequencing” promised in 2022 is still a PowerPoint slide in 2024. The same irony applies: AI startups promise to democratize intelligence, but their own infrastructure is a black box. When the market turns, investors have no way to verify the underlying value, so they flee.
Commercialization is the bleeding edge. Both companies are burning cash on GPU clusters—training a single large model costs millions. The unit economics are brutal. In crypto, we have token emissions to subsidize early adoption, but that’s a double-edged sword (ask any LUNA holder). The AI startups can’t print money; they can only dilute equity. And in a rising interest rate environment, dilution is a death sentence. The price drop is a signal that the market believes the runway is shorter than the narrative.
Competition is a knife fight. Tech giants like Alibaba and Tencent have infinite compute and can afford to give away API calls for free. The AI startups are caught between a rock and a hard place: differentiate or die. But differentiation is expensive, and the market is punishing them for it. In crypto, we see the same dynamic with Ethereum L2s competing for sequencer fees and user attention. The difference? At least the L2s have tokens that can be programmed to reward loyalty. The AI startups have nothing but promises.
Here’s where the contrarian in me surfaces. The AI stock crash doesn’t prove that crypto is superior; it proves that both ecosystems are subject to the same human folly. The hype cycle is universal. In 2017, we believed ICOs would fund the next internet. In 2021, we believed NFTs would democratize art. Now, we believe AI will replace jobs. Each time, the market overshoots, and reality pulls it back. The only difference is that crypto’s failures are more public—you can audit the code, trace the hack, and see the wallet go to zero. AI’s failures are hidden in quarterly earnings calls and layoff announcements.
But I’ve also witnessed something beautiful in crypto: the ability to learn from failure. After my yield farming disaster, I spent three months reverse-engineering the exploit and publishing the findings. That transparency, while painful, allowed the ecosystem to improve. The AI industry lacks this mechanism. When an AI model fails—say, it produces biased outputs or hallucinates facts—the fix is often a closed PR, not a community discussion. The stock market serves as a crude feedback loop, but it’s slow and full of noise.
The takeaway isn’t about shorting AI or going all-in on crypto. It’s about recognizing that value is a function of transparency, not narrative. The AI stock crash is a reminder that we need better tools for verifying claims—tools that blockchain can provide, but only if we stop treating it as a magic wand. Tokenization of AI compute? Yes, but only if the underlying hardware is auditable. Decentralized model training? Yes, but only if we solve the coordination problem without sacrificing performance.
I’m not saying that crypto will save AI. I’m saying that the same forces that drove Zhipu and MINIMAX down will drive many crypto projects down too. The market is a ruthless teacher, and it doesn’t care about your ideology. It cares about cash flows, user adoption, and real-world utility. The sooner we internalize that, the better.
So here’s my forward-looking thought: The next bull run won’t be built on promises alone. It will be built on systems that combine the verifiability of on-chain data with the intelligence of open models. We didn’t build this to be a casino, but that’s what it became when we forgot the fundamentals. The AI stock crash is a gift—if we’re willing to see it as a mirror, not a monster.