Zhao Haijun, co-chief executive of Semiconductor Manufacturing International Co., has warned that AI data centers being built at a pace never seen before across the world could remain idle, just like data centers built in China's suburbs in the early 2020s, many of which have yet to find tenants.
China's Ministry of Industry and Information Technology (MIIT) is considering a centralized cloud platform designed to pool idle computing resources nationwide and distribute computing capacity as a service through a unified national network.
"Companies would love to build 10 years' worth of data center capacity within one or two years," Bloomberg cites Zhao Haijun as saying during the most recent earnings call with financial analysts and investors. "As for what exactly these data centers will do, that has not been fully thought through.
Artificial intelligence is expected to impact the vast majority of industries and businesses in the coming years. Arguably, the biggest question is how much time it will take for AI to reach different sectors of the economy and the magnitude of this impact. Unfortunately, nobody knows the correct answers to these questions, which is why there are many talks aboutan AI bubble around the industry.
Zhao compared the current buildout to constructing high-speed rail networks or highways ahead of traffic growth, highlighting that infrastructure is being completed in anticipation of future usage rather than immediate necessity. Developers of frontier AI models, such as Alphabet, Meta, OpenAI, and xAI, would argue that they can consume virtually all resources that are given to them. However, they are not the only companies that invest billions in AI infrastructure in anticipation of future growth.
According to Moody's Ratings, spending on AI-related infrastructure could surpass $3 trillion over the next five years. In 2026 alone, capital expenditures by Alphabet, Amazon Web Services, Meta Platforms, and Microsoft are expected to reach approximately $650 billion as these companies continue to expand their AI capabilities. China-based Alibaba, Tencent, and ByteDance are also aggressively investing in AI infrastructure.
In the early 2020s, China's 'Eastern Data, Western Computing' initiative led to many startups building large AI and cloud data centers in the western parts of the country, where electricity is cheaper. The aim was to meet the demand from the economically stronger eastern provinces. Although this strategy helped cut down on power costs, the increased distance resulted in higher latency, making these centers less appealing for applications that are sensitive to delays, which limited their actual use.
Additionally, many projects were based on the assumption that state-owned companies and government agencies would be the main customers for computing power. However, the actual demand fell short of expectations, leaving many of these centers either unused or running at only 20% to 30% of their capacity, which is much lower than what they were designed for.Even with these low utilization rates, investments kept flowing in 2024 and 2025, as reported by Reuters. This raised questions among investors about the long-term viability and financial returns of these large data center projects, especially since the government is putting restrictions in place to avoid overbuilding.