The Fight Over Data Centers Is a Fight Over Who Owns the Compute
Widespread opposition to data center development is increasing across the US, leading to numerous project delays and blockages due to concerns over water, electricity, and local impacts. The article suggests that advancements in distributed AI training could reduce the need for massive, concentrated facilities, while policy changes are proposed to make large data center loads more accountable for infrastructure costs.
The article highlights a growing national backlash against hyperscale data center development across the United States. Residents in cities like Tucson, Reno, Phoenix, and Tooele are actively opposing projects, citing concerns over excessive groundwater consumption, strain on regional electricity grids, increased utility costs for ratepayers, and local impacts such as noise and truck traffic. This widespread opposition has led to a significant number of project delays and blockages, with 75 U.S. projects worth an estimated $130 billion stalled in the first quarter of 2026 alone.
A core argument presented by Product Science's Anastasia Matveeva is that the concentration of AI compute in enormous data centers, while traditionally necessary for training frontier models requiring ultra-low latency, is becoming less critical due to advancements in distributed training. Recent research, including work by Google DeepMind in 2023 and projects like Templar in 2024, demonstrates that AI models can be trained effectively across geographically dispersed hardware, significantly reducing the need for massive, single-site facilities. Nvidia has also acknowledged the physical limits of individual sites, introducing solutions to stitch sites together over long distances.
These technical shifts open possibilities for more decentralized and locally owned data center infrastructure. The article advocates for policy changes that would address the current imbalance, suggesting that large data center loads should bear the full cost of required grid upgrades. It also points to Oregon's creation of a dedicated data center rate class as a potential model and recommends that public agencies prioritize distributed and locally owned compute supply when procurement allows. The underlying debate, the article concludes, extends beyond physical inputs to questions of compute ownership and equitable participation in the AI economy, suggesting a better-designed build-out could alleviate community concerns.