CNN: AI data center pipeline trimmed

CNN reported on August 6 that the U.S. has a 565-gigawatt pipeline of proposed data-center power demand, but Columbia Business School professor Stijn Van Nieuwerburgh said he expects only about 180 gigawatts to be built over the next decade. Even that smaller figure would still imply an enormous construction program, centered less on server halls alone than on the electrical systems needed to run them. POWER Magazine reported that data-center demand could exceed available U.S. grid supply by more than 50 gigawatts by 2030, while Bloomberg, cited by Yahoo Finance, said AI’s volatile electricity use is already damaging equipment inside facilities. Slate reported that local opposition in Maryland has already paused at least one project. ### Why are so few of the proposed projects expected to get built? CNN said the gap starts with the difference between announced demand and buildable demand. Van Nieuwerburgh told CNN that only a fraction of the 565-gigawatt pipeline is likely to make it through financing, utility interconnection, land use review and physical construction over the next 10 years. The 180-gigawatt figure still points to a massive buildout. (cnn.com) CNN said Van Nieuwerburgh estimated that level of deployment could represent roughly $10 trillion of investment over a decade, which helps explain why developers, utilities and equipment suppliers continue to treat the sector as a major source of future work even as many proposals fall away. ### Where is the bottleneck now — buildings or power? (cnn.com) POWER Magazine said the main constraint is increasingly the grid, not developer appetite. Its report said data-center demand could outstrip U.S. grid supply by more than 50 gigawatts by 2030, leaving developers to rely more heavily on on-site generation, storage, transmission additions and substation work. (cnn.com) That changes the physical scope of a “data-center project.” Instead of just acquiring land and erecting buildings, developers now have to secure high-voltage service, backup power, cooling infrastructure and utility upgrades that can take years to permit and complete, according to POWER’s account of the sector’s constraints. ### What does AI’s power profile do to the equipment inside these campuses? (powermag.com) Bloomberg reported, via Yahoo Finance, that AI workloads can create rapid swings in electricity demand that are stressing equipment inside data centers. The report said those load fluctuations can damage essential systems, adding an operational problem to the broader question of how much power utilities can deliver. (powermag.com) Those reports suggest the engineering challenge is not only total megawatts but power quality and stability. In practice, that pushes more attention toward power conditioning, backup systems and equipment designed to handle abrupt changes in load, an inference supported by the reported equipment stress and grid-capacity concerns. ### Why are local fights becoming part of the buildout story? (finance.yahoo.com) Slate reported on August 6 that activists in Maryland succeeded in pausing an AI data-center project, at least for now. The article said opposition to data centers has become strong enough in some places to shape local politics and force concessions from developers. That means site selection is becoming more contested alongside power procurement. (powermag.com) A project may have capital and demand behind it, but still face delays if residents and local officials object to energy use, water consumption, noise, land use or transmission-related infrastructure, according to Slate’s account of the Maryland fight. ### So what gets built first as this pipeline gets trimmed? (slate.com) The likely near-term work is the enabling infrastructure. CNN’s lower build estimate, POWER’s grid warning and Bloomberg’s report on equipment stress all point toward earlier spending on substations, backup generation, storage, transmission upgrades and other electrical hardening before many headline data-center campuses fully materialize. That is an inference from the reported constraints, not a separate forecast from one named source. (slate.com) The next evidence to watch will come from utility interconnection queues, local permitting decisions and announced power contracts. Those are the places where the difference between a 565-gigawatt proposal list and a 180-gigawatt built market will become visible first. (cnn.com)