Lenders and investors financing China's data-center buildout now hear a former central bank chief urging that rescue tools be readied before any AI losses arrive.
Zhou Xiaochuan, a former governor of the People’s Bank of China (PBOC), warned on October 20 that the scale of capital flowing into artificial intelligence infrastructure means that if part of it fails, the result could be financial turmoil and a crisis. Speaking on a panel at the 2026 Bund Summit in Shanghai, he said such an episode could require large-scale action to support the economy and rescue some financial institutions, according to the 21st Century Business Herald, which carried the most detailed account of his remarks.
The South China Morning Post quoted him in English: "if a part of it goes to failure, it causes financial turbulence." He added: "It needs a very large action to bail out our economy." His policy prescription was preparation. Regulators, he said, should study the risk, draw up contingency plans and keep suitable tools in the policy toolbox before any failures occur.
Zhou did not name any company, lender or project, and none of the reports attributes a figure to him for how much bank or market exposure China has to AI infrastructure. His warning carries weight because of who delivered it: a former central bank chief describing a possible bailout scenario in public, at an event built around technology and the global economy.
Zhou set out why some AI investments will fail
The remarks came at a roundtable titled "AI and the Productivity Revolution: Macro Shocks Under a Technological Leap" (AI与生产率革命：技术跃迁下的宏观冲击), part of a summit that Yicai Global said runs from October 19 to 22 and is co-hosted by the China Finance 40 Forum (CF40) and Tsinghua University under the theme "Technology, Geopolitics, and the Global Economy."
Zhou framed failure as a normal part of technological change. "For technological evolution, we will undoubtedly face a situation where some investments fail and some succeed," he said, in remarks that Yicai quoted in Chinese in its morning market digest. Yicai’s digest also summarized his call to prepare in advance and his view on prices.
He split the AI economy into three parts: model and computing infrastructure, AI applications, and supporting industries such as semiconductors, power and liquid cooling. Zhou Xiaochuan judged that most of AI’s productivity gains will come from the application layer, through lighter models and model distillation. Distillation trains a smaller model to imitate a larger one. On that reading, the returns would accrue to the companies that use AI, while the infrastructure builders carry the largest investment risk.
Data centers are where his doubts concentrate
The 21st Century Business Herald reported Zhou saying that market money is pouring into data center construction, while it remains uncertain how far future AI businesses will depend on large computing centers. If lighter models and distillation reduce the need for raw computing power, some of the capacity now being financed could earn less than its backers expect.
Spending of that kind is already locked into long-term contracts at China’s smaller listed compute providers. Sharetronic, for one, has a compute cloud agreement worth up to RMB37.49 billion, a commitment that rests on demand for large computing centers holding up.
On how policy should respond to the wider structural shifts AI brings, Zhou argued that price changes should be allowed to reflect shifts in global supply and demand, and that authorities should not lean too heavily on administrative measures to head off the effects. He said the effects of AI on employment, income distribution and social structure remain unclear.
A US-China mechanism is the governance thread to follow
Zhou also addressed governance. According to the 21st Century Business Herald, he said China and the United States have reached a consensus on building an AI management mechanism, but that rapid technical change makes it hard for governments to set up governance frameworks. He proposed three areas for near-term cooperation: limiting military uses of AI, combating hacking and fraud, and protecting personal privacy.
None of the reports carries a response from the central bank, and they do not show what tools he had in mind beyond economic support and rescues for some institutions.
The practical signal to watch is whether the PBOC or China’s financial regulators begin to address AI infrastructure lending in their own financial stability reporting, or whether major lenders start disclosing their exposure to data center projects. Either would show whether Zhou’s call to fill the policy toolbox early is being taken up by the officials who now hold the tools.