The cost of electricity on America's largest power grid, operated by the PJM Interconnection, has surged by 76% over the past year. This unprecedented spike has put the spotlight on a clear culprit: the insatiable hunger for power of modern data centers, especially those optimized for artificial intelligence. An industry watchdog, the Public Utility Law Project, has pointed fingers at grid management, accusing operators of failing to properly plan for the explosion in energy loads.
This price surge is not an isolated event. It represents the first symptom of an infrastructural crisis that could reshape tech companies' location strategies. With generative AI requiring ever more powerful GPU clusters, power consumption per rack has soared, often exceeding 30-40 kilowatts. The PJM grid, serving roughly 65 million people from the Midwest to the Atlantic Coast, is paying the price for demand growth that development plans did not anticipate. The result is that local utilities are passing costs onto consumer tariffs, causing an average 76% increase that hits households and small businesses alike.
The AI energy paradox
The situation highlights a crucial paradox. While the tech industry pushes for greater computational efficiency through advanced chips, the sheer volume of computation needed for training and inference of AI models is growing exponentially. In recent months, companies like Apple have begun testing chips on Intel 18A processes, as reported in a recent deep dive on Apple's new chip course, but per-watt efficiency gains cannot compensate for global demand. Hyperscale data centers, those housing thousands of accelerators, are becoming the new energy hogs. US regulators are starting to demand transparency on power supply contracts signed by tech giants, contracts that often bypass normal grid procedures and worsen the burden for other users.
Impact on VC and cloud strategy
Rising energy costs are also influencing venture capital decisions. Investing in AI infrastructure no longer means just buying GPUs; it also means securing access to affordable energy. As analyzed in our article on Capital, Data and Second Chances, venture capital is redefining the entire AI ecosystem, and electricity cost is becoming a key risk factor in startup balance sheets. Some companies are exploring local solutions to reduce dependence on centralized cloud computing, shifting inference workloads to edge devices. This approach emerged in our coverage of AI between local and cloud, where privacy and security intertwine with the need to contain energy costs. However, for training large models, cloud computing remains essential, and the power grid must be upgraded.
Future outlook and systemic risks
If this trend is not reversed, AI expansion risks being hampered not by computational limits but by infrastructure constraints. PJM has already announced a long-term plan to increase generation and transmission capacity, but construction timelines take years, while demand grows month by month. The watchdog has suggested differentiated tariffs for large industrial consumers to incentivize efficiency and plant siting near renewable sources. Furthermore, the US situation could foreshadow similar dynamics in Europe and Asia, where new data center construction is already facing opposition. For more context on PJM's regulatory framework, see the Wikipedia page for PJM Interconnection. The AI energy bill has arrived, and it will be steep for everyone.
Sponsored Protocol
