// data centers

Source: Works in Progress

Data center power delivery has become the hard constraint on AI expansion, not semiconductor manufacturing. Abilene's $20 billion Lantana power project typifies how utilities and grid infrastructure now limit GPU cluster placement. The bottleneck has shifted from Silicon Valley's chip design cycle to Texas utility politics and transmission line permitting, where 18-month Environmental Impact Statements matter more than TSMC's fab capacity. Cloud giants are scrambling to secure nuclear power contracts. Grid operators, not OpenAI, effectively control the pace of model training.

Source: NYT > Business (paywall)

Data center noise—particularly the low-frequency vibration from cooling systems and generators—is an externality of AI infrastructure expansion that tech companies have largely ignored in their siting decisions. Unlike previous tech booms concentrated in urban centers with existing zoning frameworks, the computational demands of large language models are driving facility construction in rural and suburban areas where residents have fewer legal protections and noise ordinances weren't designed for industrial-scale acoustic pollution. As companies optimize for land cost and power availability, they're externalizing health and quality-of-life costs onto communities with limited recourse or political leverage.

Source: The Prof G Pod

The electricity infrastructure powering AI clusters is hitting physical limits faster than semiconductor production, a constraint that alters both the timeline and geography of AI deployment. Shah's framing shifts the bottleneck from vendor control (NVIDIA) to physics and regulatory approval—data centers need grid connections that take years to secure, meaning capital and permits now matter more than wafer starts. Energy policy and utility reform move from peripheral concerns to competitive advantage for countries and companies able to solve the grid problem.

Source: Popular Information

While grassroots movements mount resistance to data centers over environmental and infrastructure concerns, state governments are actively competing to attract these facilities through massive tax incentives. This creates a structural misalignment: local communities bear the costs (water consumption, grid strain, noise) while state treasuries absorb revenue loss. Governors are bidding against each other for installations that generate immediate job claims but uncertain long-term fiscal returns, effectively outsourcing AI infrastructure buildout costs to the public sector.

Source: The Verge

Amazon disclosed its water consumption at 2.5 billion gallons as Seattle imposed a one-year moratorium on new data center construction. The timing exposes a direct collision: AI infrastructure scaling meets municipal resource constraints. Cooling systems for compute-dense facilities account for most of that consumption. The disclosure suggests Amazon's expansion plans now face friction from local water stress, particularly in water-scarce regions where hyperscalers are concentrating builds. This creates a hard infrastructure ceiling that neither voluntary sustainability commitments nor geographic arbitrage easily circumvent. AI's computational demands are hitting regional environmental capacity limits.