Utilities are spending billions on the data center boom. What are the risks?

Core Insights - The GPU depreciation issue poses a significant risk to neocloud companies like CoreWeave, which provide GPU capacity to hyperscalers, indicating a potential threat to the entire sector [1] - The future of the inference services industry remains uncertain due to the presence of competing companies offering similar services, making it difficult to predict demand [2] - The electric power industry operates on long-term planning horizons, while the surge in data center demand driven by AI technologies has occurred rapidly since late 2022, creating challenges for utilities [4] Industry Dynamics - The demand for electricity from data centers is unprecedented, leading utilities to race to build new generation and grid infrastructure [5] - Neocloud companies are at a higher risk during market corrections, especially compared to established cloud service providers like Google and Amazon [6] - The credit risk associated with large load customers is a significant concern for utilities, as the financial stability of these customers is crucial for recovering infrastructure investments [7][8] Market Trends - There are reasonable concerns about a potential market correction, similar to the dot-com bubble, due to the interconnected nature of investments in the AI and data center sectors [9][10] - Utilities are increasingly adopting large load tariffs and long-term contracts to manage risks associated with connecting data centers to the grid [20] - The Northern Virginia Electric Cooperative anticipates that data center customers will account for over 95% of its energy sales by 2032, raising concerns about dependency on a single customer segment [26] Technological Considerations - The energy demands of AI data centers are fundamentally different from traditional data centers, complicating future power needs predictions [3] - GPUs, essential for AI workloads, can draw significant power and generate unpredictable energy spikes, posing challenges for energy management [14][15] - The focus on securing power for large data centers has overshadowed efforts to improve energy efficiency, although there is a push for cleaner power solutions [16][19] Regulatory and Structural Factors - Utilities are implementing new tariff structures to rationalize demand and manage risks associated with data center connections [22][28] - The structure of utilities, whether vertically integrated or not, influences the effectiveness of models like bring-your-own-generation for meeting data center demand [28] - The potential acquisition of the Northern Virginia Electric Cooperative by Dominion Energy highlights the interconnected nature of utilities and data center operations in the region [26]