In the chaos of summer, we found our winter soul. On August 24, the Hong Kong market watched two of China’s most celebrated AI large-model companies—Zhipu and MINIMAX—plunge over 11% and 10% respectively. Headlines screamed “AI bubble burst,” but as a DAO governance architect who has spent years auditing the trust assumptions of decentralized protocols, I saw something else: a quiet confirmation of a structural flaw that no amount of hype can patch. These companies are not failing because their models are weak; they are failing because their governance is built on sand.
Let me set the context. Zhipu (GLM series) and MINIMAX (abab series) are the poster children of China’s AI race, each valued at billions of dollars. They rely on centralized cloud infrastructure, proprietary training data, and closed ecosystems. Unlike decentralized protocols that distribute power across token holders, these companies operate as traditional corporations: a board of directors, a CEO, and a shareholder-driven model. The market’s sell-off is not a judgment on their technical capabilities—both have released impressive models—but a recognition that their value proposition is brittle. In a world where trust is the only asset that matters, centralized AI companies are vulnerable to the same forces that topple any centralized system: information asymmetry, governance opacity, and loyalty to capital over community.
The core insight here is that the market is finally pricing in the cost of centralization. During my 2020 work with the LendFlow protocol, I saw how a community’s trust could weather a liquidity scare when governance was transparent and inclusive. When we introduced quadratic voting to CivicChain in 2024, participation from non-whale addresses surged by 40%—not because the technology was better, but because the community felt heard. Zhipu and MINIMAX lack this social layer. Their users are customers, not stakeholders. When ByteDance or Alibaba slashes API prices, Zhipu cannot turn to its community for a governance vote; it must cut costs or find new investors. The market interprets this vulnerability as a discount. Indeed, the price drop is not a correction but a repricing of risk: the risk that a centralized entity can be outcompeted by rivals with deeper pockets or faster execution. Code is law, but conscience is the compiler—and these companies have no conscience encoded in their smart contracts.
Now, the contrarian angle. Many will argue that the sell-off is overblown, that AI is a long-term trend and today’s dip is a buying opportunity. They point to the real demand for AI in coding, customer service, and content creation. But I see a deeper blind spot: the market is conflating technological value with organizational resilience. A great model does not guarantee survival if the organization behind it is centralized and opaque. Consider the parallel to DeFi’s oracle problem. Chainlink, despite its dominance, remains a single point of failure because its nodes are run by a small set of centralized entities. Similarly, Zhipu and MINIMAX rely on centralized GPU clusters, centralized data pipelines, and centralized decision-making. When the market panics, there is no decentralized governance buffer to absorb the shock. The silence in the bear market is where truth compiles—and right now, the truth is that these companies are structurally fragile.
What does this mean for the future? Governance is not a vote, it is a vigil. The next wave of AI companies will not be those that simply build better models, but those that embed community ownership, transparent governance, and decentralized infrastructure into their DNA. Imagine a Zhipu that tokenizes its compute resources, allowing users to stake tokens for model training decisions. Imagine a MINIMAX that holds liquid governance votes on API pricing, aligning incentives with its developers. This is not a utopian dream—it is the logical evolution of value creation in a world where trust is scarce. The market has just issued a clear warning: centralized AI is a relic of the past. The future belongs to those who weave nets of trust, not walls of control.
We do not build walls, we weave nets of trust. The question is not whether Zhipu and MINIMAX will recover, but whether they will learn to code conscience into their governance. Until then, watch the charts, but remember: the real signal is not the price—it is the architecture of power.