AI Exchanges: Power Problems?

In this episode of Goldman Sachs Exchanges, hosts Allison Nathan and George Lee welcome Brian Singer, head of GSUSD and Goldman Sachs Research, to explore the growing power demands driven by the AI boom. The conversation centers on the massive capital investment—over $300 billion in 2026–2027—by hyperscalers in data centers and AI infrastructure, which is fueling unprecedented growth in global data center power demand. Singer highlights a projected 220% increase in AI and non-AI data center power consumption by 2030 compared to 2023, equivalent to adding another top-10 energy-consuming country. Despite this surge, supply remains tight due to a critical shortage of skilled labor—particularly in transmission and distribution—making 'people' the primary constraint, not power pricing. The discussion also covers the rise of behind-the-meter natural gas solutions, the role of policy and public sentiment, and the long-term shift toward green energy, including nuclear, solar, and battery storage. While hyperscalers can absorb higher energy costs without significantly impacting profitability, utilities face greater financial and regulatory hurdles. The episode concludes with a nuanced view of a 'yes-and' future where multiple energy sources coexist, driven by innovation, policy, and time. Key takeaways include: 1) The AI infrastructure buildout is creating a structural power demand crisis, not just a temporary spike; 2) The biggest bottleneck isn't energy supply but skilled labor in grid infrastructure; 3) Hyperscalers are financially resilient and willing to pay more for green energy to avoid public backlash; 4) Behind-the-meter solutions are a near-term necessity, even if less efficient; 5) Nuclear and renewables are poised to play a major role by the 2030s, but deployment timelines are critical; 6) Policy and public acceptance will shape energy sourcing more than price; 7) The AI power challenge is a multi-dimensional problem requiring coordinated investment across energy, labor, and regulation; 8) The market underestimates the human capital intensity of the AI transition.