The recent economic downturn has had significant repercussions for China’s ambitious plans to rapidly construct AI data centers. Initially, this expansion was stimulated by the launch of ChatGPT, which sparked a government-driven initiative to establish over 500 AI data centers across various cities and provinces. However, the reality is that a substantial number of these data centers remain underutilized, with analysts estimating that up to 80% of the new capacity is idle. This underperformance can be partly attributed to the cost and investment shifts brought about by the DeepSeek revolution, which have dramatically altered the landscape of AI production in China.
The Impact of DeepSeek and Falling GPU Costs
Underutilization of New Data Centers
Despite the initial enthusiasm for AI data center construction, many new facilities are not being used to their full potential. One of the primary reasons for this is a significant plunge in the rental costs of Nvidia GPUs. For instance, the monthly rental cost of an Nvidia #00 server with eight GPUs has dropped from RMB180,000 ($24,802) to approximately RMB75,000 ($10,317). This steep decline has not been matched by a corresponding decrease in the purchase price of these GPUs, which remains exorbitantly high. Consequently, several companies have opted to halt their data center projects and sell off their GPUs. Moreover, the high operational costs, particularly concerning energy expenses, compel numerous firms to leave their data centers idle rather than operate them at partial capacity.
Oversupply and Lack of Expertise
Another critical issue is the oversupply of data centers driven by the frenzy to build “smart data centers,” attracting entities that lack the necessary expertise or genuine interest in the sector. Many of these players were more adept at leveraging subsidized green energy and government-backed loans rather than committing to the intricate work of data center operations. This opportunistic approach has led to numerous inefficiencies and economic imbalances. For example, while these facilities were envisioned as cutting-edge, their management and operational capabilities often lag behind due to insufficient expertise. This has exacerbated the problem of underutilization and contributed to the broader economic challenges facing the sector.
Major Investments Persist Despite Challenges
Telecom Operators and Tech Giants’ Commitment
Despite the setbacks and inefficiencies, significant investments are still pouring into the top tier of the market. China’s three major telecom operators, along with leading technology giants, continue to invest heavily, albeit on a smaller scale compared to their American counterparts such as AWS and Meta. Recent financial disclosures reveal that these telecom operators plan to spend no less than RMB90 billion ($12 billion) this year alone. Additionally, Alibaba has announced its intention to invest a staggering $53 billion over the next three years in AI hardware and computing infrastructure. ByteDance, another key player, aims to allocate approximately $20 billion this year primarily towards enhancing its AI infrastructure.
Strategic Partnerships and Long-term Plans
In a move to further bolster its AI capabilities, Alibaba has entered into a new partnership with China Mobile. This collaboration will focus on joint efforts in constructing and operating AI data centers, though specific financial details of the partnership have not been disclosed. These strategic investments and partnerships underscore the commitment of leading Chinese companies to advance their AI infrastructure despite the broader market challenges. The willingness of these companies to make substantial long-term investments signals their confidence in the future potential of AI technologies, even as they navigate the current economic landscape.
Conclusions on AI Data Center Trends in China
The recent economic downturn has had profound effects on China’s ambitious plans to swiftly build AI data centers. Initially, the push for expansion was ignited by the advent and popularity of ChatGPT, which resulted in a government-led plan to create over 500 AI data centers in various cities and provinces. However, many of these data centers remain largely underutilized. Analysts estimate that as much as 80% of the new capacity is sitting idle, a stark contrast to the initial projections. One major reason behind this underperformance is the cost and investment shifts linked to the DeepSeek revolution, which have significantly changed the AI production landscape in China. Consequently, the anticipated rapid growth in AI infrastructure has faced substantial hurdles, questioning the sustainability and practicality of such expansive projects amidst ongoing economic challenges. These shifts have also prompted businesses to reassess their investment strategies, further complicating China’s AI development trajectory.