Why Don't Data Centers Use More Green Energy? (www.nytimes.com)

Big AI data-center builds from Nvidia, OpenAI (with SoftBank and Oracle) and other cloud giants are accelerating power demand just as global investments in A.I. infrastructure surge — Nvidia pledged $100 billion to support OpenAI’s expansion, OpenAI’s Stargate project was pegged at about $500 billion, and together major firms are spending more than $325 billion on new facilities this year. Those facilities run 24/7 on extremely high power densities to feed advanced GPUs and dense cooling systems, so in practice much of that new demand is being met by readily dispatchable fossil fuels: in the U.S. more than half of the incremental power still comes from coal, natural gas and oil. The significance for AI/ML is twofold: operational constraints (continuous uptime, latency and power-density needs) and grid limitations make intermittent renewables and current storage solutions hard to rely on exclusively, meaning rapid AI scale-up can increase carbon emissions unless procurement changes. That creates pressure and opportunity for technical and market fixes — larger PPAs, on-site clean generation, microgrids, utility-scale storage, green hydrogen and grid upgrades — to decarbonize compute-heavy workloads. For researchers and infrastructure teams, the takeaway is that hardware and model scaling choices now have direct energy-system implications, and integrating energy-aware design and procurement will be essential to align AI growth with climate goals.