// data centers

Source: Wired

Ulanqab has become China's preferred location for new data center capacity because it offers the triumvirate that makes hyperscale infrastructure economical: coal-powered electricity at a fraction of coastal rates, vast undeveloped land, and still-reasonable latency to Beijing's tech market and regulators. AI training and inference workloads are moving compute away from expensive coastal tech hubs toward resource-rich inland provinces. This pattern mirrors but inverts Western hyperscaler logic of clustering near urban demand.

Source: The New York Times

Virginia's largest data center hub has restricted new construction after years of rapid expansion. The move reflects friction between digital economy needs and local quality of life: power grid strain, water consumption, noise, and land use changes that residents oppose. Despite massive tax contributions, data centers now face real political costs. Hyperscalers and cloud providers will need to adjust siting strategies—building in areas with less organized resistance, investing in smaller distributed facilities, or consolidating in existing mega-campuses elsewhere.

Source: The Wall Street Journal (paywall)

Caterpillar, Cummins, Eaton, and Ford are redirecting manufacturing capacity and R&D toward specialized power equipment for AI infrastructure—generators, cooling systems, electrical distribution—because data centers now consume more electricity than many countries and incumbent suppliers can't scale fast enough. This reallocation of US industrial capacity away from traditional markets (construction, automotive, utilities) toward compute infrastructure has real consequences for labor skills, supply chains, and which companies capture the economic rents of the AI buildout.

Source: SiliconANGLE

Source: The Next Web

Amazon is building its own natural gas plant to power a massive data center rather than relying on Texas's grid. The move reflects a hard constraint: AI demand is outpacing utility capacity, and hyperscalers can no longer count on public infrastructure to keep pace. By investing billions in captive power generation, Amazon is essentially opting out of the grid altogether—a signal that data center growth is now limited by energy availability, not capital or computing design. This creates a new class of industrial infrastructure operating as parallel power systems, potentially fragmenting energy markets and leaving smaller operators dependent on increasingly strained regional grids.

Source: The Verge

Amazon is contracting with a new natural gas facility in West Texas that will rank among the nation's worst polluters, gaining cheap, reliable power while passing environmental and health costs to local communities. The deal exposes a gap between Big Tech's net-zero pledges and its actual infrastructure choices: renewable energy commitments still lose to the economics of fossil fuel baseload power when utilities and regulators allow it. Data center siting decisions prioritize developer incentives and power availability over emissions accountability, a precedent other hyperscalers will likely follow.

Source: The New York Times

Amazon is bankrolling a dedicated natural gas facility to power a single off-grid AI data center in Texas—a move that would create one of the largest US emissions sources and directly contradicts the company's 2040 net-zero commitment. Current grid infrastructure cannot support the power demands of large-scale AI infrastructure, forcing hyperscalers to build independent generation rather than wait for renewable buildout or grid upgrades. Similar fossil fuel deals are likely to follow across the industry, subordinating corporate climate pledges to the immediate capital intensity of training and inference at scale.