Cerebras CEO Andrew Feldman presents a unique perspective on the AI competition between the United States and China, focusing on energy consumption as a decisive factor. He argues that by 2025, AI training could consume up to 20% of the world’s electricity, creating a significant bottleneck in AI development. The article emphasizes that access to power infrastructure and energy resources will become crucial determinants in AI leadership. Feldman highlights that China’s advantage in manufacturing could be offset by the US’s superior energy infrastructure and more reliable power grid. He points out that while China leads in solar panel production and rare earth minerals, the US maintains an edge in advanced semiconductor technology and stable power distribution. The CEO suggests that future AI development will be constrained not by chip availability or algorithmic advances, but by access to reliable, abundant electricity. The article also discusses how this energy requirement could reshape global AI development strategies, with countries needing to balance their AI ambitions with energy capacity. Feldman predicts that energy efficiency will become a primary focus in AI system design, potentially leading to new innovations in power-efficient computing. The conclusion emphasizes that success in the AI race will increasingly depend on a nation’s ability to provide and manage massive amounts of reliable energy for AI training and deployment.
Source: https://www.businessinsider.com/cerebras-ceo-andrew-feldman-china-us-ai-race-energy-2025-10