When Zhipu AI and MiniMax, two of China’s most prominent AI large-model startups, saw their Hong Kong-listed shares drop by over 10% in a single session, the market didn’t just blink—it flinched. The sell-off wasn’t a random tremor; it was a signal. These companies, once darlings of the AI boom, now find themselves caught in a structural repricing that mirrors the very pattern I observed during the 2022 bear market, when over-leveraged DeFi protocols collapsed under the weight of their own promises. The difference is that this time, the collateral is not code—it’s a narrative. And narratives, as any blockchain veteran knows, are the first to crack when the market demands proof of work.
The context here is critical. Zhipu AI, backed by Tsinghua University, and MiniMax, a consumer-facing AI startup, represent the second tier of China’s “AI Four Little Dragons”—a category that once commanded billion-dollar valuations based on nothing more than a few impressive demos and a slide deck about market size. Both companies likely went public via SPACs or reverse mergers, a path that often masks a fundamental misalignment: the price set by private investors is rarely the same as the price the public market will accept. I’ve seen this play out before—in 2017, during the Ethereum Classic community’s struggle to define “Code is Law,” the market punished projects that couldn’t articulate their value beyond hype. The same principle applies here. The market is now asking: “Where is the revenue? Where is the user retention? Where is the moat?”
Let’s dig into the core of this correction. The valuation of these AI companies was built on a story of technological superiority and boundless addressable markets. But the data tells a different story. Zhipu’s B2B business, reliant on API calls and government contracts, lacks the scale to justify its private-market valuation. MiniMax’s consumer apps like Talkie and Hailuo AI face churn rates that would make any SaaS investor wince. The market is now applying a discount to companies that burn cash faster than they can prove unit economics. This is not a temporary dip; it is a paradigm shift from “story-driven” to “earnings-driven” valuation. In my years of auditing L1 protocols, I learned that once a system’s fundamental utility is questioned, the price correction is rarely linear—it cascades. The same applies here. The loss of confidence in one AI stock can trigger a chain reaction, forcing other private AI companies to lower their next funding round’s valuation, creating a negative feedback loop that tightens the entire capital market.
But here is where the contrarian angle emerges. The market’s wrath against centralized AI companies is not a verdict on AI itself—it is a verdict on the structure of AI value capture. Centralized AI companies are, fundamentally, black boxes. They control the model, the data, and the distribution. Investors have no way to verify claims of usage or efficiency. They rely on quarterly reports that can be gamed, and on management teams whose incentives may not align with long-term value creation. In contrast, decentralized AI projects—like Bittensor, io.net, or Render Network—offer a different architecture. They use token incentives to reward compute providers, open-source models to ensure transparency, and on-chain governance to align stakeholders. The recent correction in centralized AI stocks may actually accelerate capital rotation into these decentralized alternatives. I recall a similar shift during the 2020 DeFi Summer: when traditional finance stocks faltered, liquidity flowed into protocols like MakerDAO, which offered transparent, programmable value. The same could happen now. The market’s loss of faith in Zhipu and MiniMax could be the catalyst that drives serious capital toward the decentralized AI stack, where every model’s inference is verifiable, and every token’s emission is auditable.
Yet, caution is warranted. Decentralized AI is not a panacea. The same speculation that inflated centralized AI valuations now threatens to contaminate decentralized AI tokens. Many of these projects have yet to demonstrate real-world usage beyond testnets and speculative trading. The risk of a “liquidity carpet pull” is real—especially in a bear market where retail capital is scarce. We must also acknowledge that decentralized governance can be slow and factional, leading to technical stagnation. The path forward is not a binary choice between centralized and decentralized, but a hybrid model where transparency and incentive alignment are prioritized. Projects that combine on-chain verification with sustainable tokenomics will survive. Those that rely solely on hype will join Zhipu and MiniMax in the graveyard of overvalued ideas.
As I write this, I am reminded of a lesson from the Ethereum Classic days: We chart the code, but the soul chooses the path. The market is now choosing. The path of centralized AI, with its opaque valuations and fragile narratives, is being pruned. The path of decentralized AI, with its promise of verifiable computation and community ownership, beckons. But it is not a smooth road. It requires patience, rigorous auditing, and a willingness to accept that the market will, once again, correct its excesses. The question is not whether AI will survive, but which architecture will bear the weight of that survival. The answer, I suspect, will be written in code, not in quarterly earnings calls.
We chart the code, but the soul chooses the path. The contract executes. The conscience judges. And in the end, the market will decide which system deserves its trust.
