Key facts
- Chinese AI firms achieve 90% of US performance at 10% of the cost, according to Mehrdad Emadi.
- China's centralized 800-1,100 kV UHV DC power grid allows seamless energy transmission over 3,000 km with near-zero loss.
- The US grid is split into three isolated interconnections (Eastern, Western, Texas/ERCOT) using 345-500 kV HV AC lines.
- US data center capacity demand is forecast to reach 118-134 GW by 2030 and 194 GW by 2035.
- US AI firms have hundreds of billions of debt attached, with costs running five to ten times their revenues, Emadi said.
- Gas turbine manufacturers GE Vernova, Siemens, and Mitsubishi Heavy Industries have order backlogs sold out for up to seven years.
Concerns are mounting over the potential for a significant downturn in US artificial intelligence (AI) stock valuations, driven by a structural cost disadvantage compared to Chinese competitors, according to analysts. Mehrdad Emadi, head of risk analysis at Betamatrix, stated that major Chinese AI players achieve approximately 90% of the performance of their US counterparts but at roughly 10% of the cost. This discrepancy is largely attributed to the fundamental differences in electricity grid infrastructure.
Steve Keen, an honorary professor at University College London, highlighted that China benefits from a centralized power grid designed by engineers, enabling efficient transmission of power across vast distances with minimal loss. This grid utilizes 800-1,100 kilovolt (kV) Ultra-High Voltage Direct Current lines, capable of moving up to 12 gigawatts of power over 3,000 kilometers. In contrast, the US grid is fragmented into three isolated interconnections (Eastern, Western, and Texas/ERCOT) relying on standard 345-500 kV High-Voltage Alternating Current lines, which are less efficient and more costly for transmitting large amounts of energy.
Emadi warned that the high electricity costs, which constitute up to half of US AI firms' expenses, are likely to increase, while Chinese firms' costs may remain stable or decrease. This, coupled with the US grid's physical limitations on scaling AI clusters, presents a significant challenge for American companies. The demand for power for US data centers is projected to surge, with Goldman Sachs Research forecasting capacity demand to climb from 42 GW to 118-134 GW by 2030, and BloombergNEF projecting a rise to 194 GW by 2035.
Furthermore, the supply chain for essential equipment like gas turbines, dominated by GE Vernova, Siemens, and Mitsubishi Heavy Industries, faces long waiting times of up to seven years, with prices expected to triple. Emadi also noted the heavy reliance on private credit by US AI firms, accumulating hundreds of billions in debt, which could exacerbate losses if these companies face financial distress.