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.
"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.
Meanwhile, under China's 'Eastern Data, Western Computing' initiative in the early 2020s, numerous startups constructed large AI and cloud data centers across western regions of China, where electricity costs are lower, with the goal of serving demand from economically stronger eastern provinces. While the strategy reduced power expenses, it turned out that longer distances increased latency and made these facilities less attractive for many latency-sensitive applications, which limited actual usage.
Also, many projects were developed on the expectation that state-owned enterprises and government institutions would become primary customers for computing capacity. In practice, demand failed to meet projections, and many of these facilities were either idle or operating at only 20% to 30% utilization, far below designed capacity. Despite weak utilization rates, investment continued in 2024 and even 2025, according to a Reuters report, which made investors wonder about the long-term sustainability and economic return of these large-scale data center projects, whereas the government is imposing restrictions to prevent overbuilding.
At the same time, 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. Yet, developing such a network will be exceedingly hard as data centers rely on different hardware and software stacks with different capabilities.
Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.
