“It seems like we’re witnessing the start of a bubble.”
According to Bloomberg and other international news outlets, Alibaba Group Chairman Daniel Zhang made this comment at the recent HSBC Global Investment Summit in Hong Kong. He warned that the competition to build data centers, vital for artificial intelligence (AI) development, is overheating, with supply surpassing actual demand.
On Sunday, the tech industry reported growing concerns over overheated AI infrastructure investments, following Zhang’s remarks and Microsoft’s decision to cancel its data center lease agreements.
This sentiment is also evident in China. Since the debut of ChatGPT, the country has focused on expanding its AI infrastructure, but demand is failing to keep up with supply.
MIT Technology Review reports that China launched over 500 new data center projects between 2023 and 2024, with at least 150 centers expected to be completed by the end of 2024. However, the publication states that 80% of these new data centers remain idle.
Chinese AI companies are also showing signs of change. In 2024, over 144 Chinese firms developed their own large language models (LLMs). Yet, MIT Technology Review notes that only 10% of these companies still invest in LLMs.
In this environment, GPU rental prices have also dropped. At one point, renting a server with eight H100 GPUs cost around 180,000 CNY (approximately $24,800), but now it has fallen to about 75,000 CNY (approximately $10,330). Despite the price reduction, demand remains low. Some data center operators are abandoning rentals altogether, as these prices make operations unprofitable.
This shift in Chinese AI companies is closely linked to the ChatGPT phenomenon. Since its emergence, most AI startups, except for a few well-capitalized ones, have abandoned plans to train their own AI models using data centers. Many are now pivoting to open-source models.
The Financial Times reports that Chinese AI startup 01.ai has decided to stop training its own LLM and instead focus on selling AI solutions. Their strategy is to create products that leverage multiple models rather than develop a single large model.
It highlights that ChatGPT has reshaped the competitive landscape in China’s AI sector, forcing companies to either compete head-to-head with ChatGPT or explore new markets using open-source models.