U.S. data centers are expected to more than double their electricity consumption by the end of the decade, a surge driven largely by the insatiable power needs of artificial intelligence. The projected jump — from 167 terawatt-hours in 2023 to roughly 376 TWh by 2030 — is already rattling markets and hitting individual AI companies. On the prediction platform Polymarket, the probability of AI startup Anthropic fell to 92.85%, a drop tied directly to the data-center power crunch.
Why electricity demand is climbing
Grid constraints are the main bottleneck. New data centers require enormous, round-the-clock power, and utilities are struggling to connect them fast enough. The 209 TWh increase over seven years is roughly equivalent to adding the entire current electricity consumption of France. Most of that growth will come from AI training and inference workloads, which use far more energy per query than traditional cloud computing.
The power crunch’s impact on AI companies
Anthropic, the company behind the Claude chatbot, isn’t alone in feeling the heat. Every major AI lab needs guaranteed power to run its models. When grid capacity tightens, it delays new data-center builds and raises operational costs. The Polymarket probability drop — from a higher level not specified in available data — reflects traders’ belief that the power shortage could slow Anthropic’s rollout plans or increase expenses enough to hurt its valuation.
Polymarket is a decentralized prediction market where users bet on real-world outcomes. A 92.85% probability still implies the market thinks Anthropic will succeed, but the decline suggests growing concern about the power constraint.
What the numbers mean for the grid
The U.S. Energy Information Administration and grid operators have warned that the pace of data-center construction is outstripping transmission upgrades. Some regions, like northern Virginia, have already imposed moratoriums on new connections. Without faster permitting and investment in generation, the 376 TWh figure could become a floor rather than a ceiling.
Regulators haven’t yet announced new rules. But the Federal Energy Regulatory Commission has signaled it will examine how data centers pay for grid upgrades. That review could lead to higher electricity rates for the biggest users — including AI companies.
For now, Anthropic’s odds on Polymarket offer a real-time barometer of how investors view the power problem. The next data point to watch is the company’s own disclosure of its data-center contracts, or a broader grid reliability report due from the North American Electric Reliability Corporation later this year.
