An essay arguing that AI should be understood as infrastructure rather than a novelty, comparing it to electricity and the internet. The author contends that while AI's language capabilities invite anthropomorphization, the real challenges involve regulation, equitable access, and efficient distribution of computational resources rather than model ownership.
Google has agreed to fund upgrades at Georgia Power's Vogtle and Hatch nuclear plants to increase electricity capacity for its AI data centers, as local opposition to AI facilities mounts over concerns about strain on power grids and rising utility costs. The deal reflects a broader trend of AI companies turning to nuclear power for infrastructure projects, partly due to the Trump Administration's opposition to solar and wind energy development.
This post develops a seasonal ARIMA forecast for household electricity consumption using four years of minute-by-minute meter data from UCI, achieving 0.2774 kW mean absolute error. The analysis covers data cleaning, stationarity testing, multiple modeling approaches including GARCH and state space models, and backtesting procedures to predict daily power use.
Anthropic has committed to 15 GW of additional computing power over the next few years at an estimated cost of $500 billion, while OpenAI increased its data center expense projection to $750 billion through 2030. Major concerns include whether these companies can generate sufficient revenue to offset massive spending commitments and whether data center construction delays—with only 50% of planned AI capacity expected to come online by 2028—will create significant compute supply shortages.
EV charging involves unavoidable losses as electricity is transmitted and converted from grid AC to battery DC. Testing shows AC Level 1/2 chargers typically lose 14–20 percent of electricity to heat, conversion, and thermal management, while DC fast chargers lose approximately 10 percent. Using dedicated wall-mounted chargers significantly reduces losses compared to household outlets.
Oregon has 111 operational data centers consuming about 23% of the state's retail electricity in 2025, with 32 more facilities planned or under construction, according to a new report from ECOnorthwest and the University of Virginia. The facilities directly employ 2,630 workers, mostly concentrated in Eastern Oregon, though data centers represent only 0.2% of the state's workforce. The report highlights gaps in available data on water use and community impacts as data centers become an increasingly contentious issue in Oregon.
India's coal power generation remained flat from mid-2024 to mid-2026, marking the first two-year period without growth in over 50 years despite rising electricity demand. The stagnation coincided with record non-fossil power generation increases, though India continues investing in new coal capacity and fossil fuel infrastructure.