The Chinese Bedrock of American AI

The high-tech rivalry between the US and China has forged a complex web of mutual dependencies that both nations are now desperate to sever. However, while the US maintains a strategic edge in software and chip design, the landscape shifts dramatically at the hardware level. The generative AI boom demands a massive expansion of physical infrastructure, revealing a critical vulnerability: building modern data centers (DCs) in the US is virtually impossible without components from China.

The issue is not high-end processors, but the foundational elements of power systems and connectivity. Analysts point out that the American data center industry is critically dependent on Chinese power transformers, networking and switching equipment—including fiber optics—and batteries for uninterruptible power supplies (UPS). This situation is further strained by political pressure; executive orders granting the US Department of Energy the authority to block supplies deemed a threat to national security could paralyze construction projects at any moment. Power systems saturated with Chinese components have become the primary target of such restrictions.

The scale of this dependency is staggering. In the power transformer segment, Chinese suppliers control up to 30% of the market, while their share of backup power batteries reaches 40%. When descending further into the raw material supply chain required to manufacture this equipment, the US dependence on China becomes almost absolute.

A similar pattern emerges in optical data transmission—the "nervous system" of any data center. In several key categories, Chinese vendors control approximately two-thirds of the global market. Despite Washington's drive to reduce this reliance and curb imports, the reality is that Western manufacturers would need at least two years of intensive effort just to compensate for the resulting supply gaps. Furthermore, transitioning to local alternatives will inevitably drive up construction costs, exerting additional financial pressure on the industry.

The current trajectory of capacity expansion renders decarbonization and production localization nearly impossible in the short term. According to S&P Global forecasts, total data center capacity in the US is projected to grow from 62 GW in March 2026 to 152 GW by 2030. Given this pace of expansion, attempts to rapidly decouple from Chinese equipment appear utopian. Localizing the production of all necessary components will take years, while the market is already grappling with an acute equipment shortage.

The problem is exacerbated by a grid connectivity crisis. Even in developed regions, integrating a new data center into the power grid can take several years, despite businesses expecting a launch within six months. In some locations, the wait for connection stretches to eight years. Transformer suppliers are already booked out three years in advance, and the most forward-thinking players are attempting to secure deliveries as far out as 2030.

Parallel to the growth in computing power is an increasing thermal footprint. The demand for electricity and efficient cooling has become the dominant factor in development. By the end of the decade, up to 70% of data centers are expected to be equipped with liquid cooling systems. The need to simultaneously solve power supply and heat dissipation issues is forcing the industry to explore radical infrastructure deployment strategies. The focus is shifting toward extreme environments: from deep mountain ranges and underwater sites to the full-scale deployment of computing power in space.