// energy

Source: The Atlantic

AI's computational demands are forcing communities to confront the physical costs of generative AI—massive energy consumption, water usage, and grid strain—that Silicon Valley had previously externalized into the background. Unlike cloud infrastructure that could hide in remote locations, AI training requires so much power that it's now competing directly with residents for reliable electricity and triggering coordinated local opposition that standard corporate lobbying struggles to overcome. This creates real constraints on where and how quickly companies can deploy next-generation models, potentially shifting competitive advantage toward firms with existing power infrastructure or those willing to negotiate serious community concessions.

Source: Financial Times

Tech giants are now competing in acquisition markets they once ignored, buying power plants, data center real estate, and fiber networks as core business assets rather than operational support. This shift creates a new M&A category where infrastructure deals command the same strategic weight as software acquisitions once did. Control over physical infrastructure—not just code—now determines competitive position in AI deployment.

Source: TechCrunch

xAI's pivot to natural gas infrastructure reflects a hard constraint: frontier AI training demands baseload power that renewable energy can't reliably supply on the required timeline. Natural gas plants scale faster than solar farms and provide uninterrupted power to GPU clusters. This exposes a gap between the clean-energy narrative around AI and what its builders actually choose when speed matters. Musk has spent a decade promoting solar as civilization's energy answer, yet is now betting on gas. SpaceX's parallel push into orbital data centers suggests the company sees the real solution not as reimagining Earth-based energy sources but as escaping the grid through space-based infrastructure.

Source: Stratechery by Ben Thompson

Data center opposition is rooted in genuine local costs—water depletion, grid strain, noise, land use—that concentrate in specific communities while benefits accrue to distant tech companies and users. Ben Thompson's conclusion is that compensation (not environmental promises or job creation) is the only mechanism that actually moves projects forward. This exposes a deeper problem: the AI infrastructure race is running ahead of any consensual settlement between corporations and the places forced to host them. Without formalizing payment structures now, data center projects will face year-long permitting battles and local vetoes that slow AI expansion.