AI datacentre limits show up
Big AI datacentre projects are running into money, people and energy constraints as they scale physical infrastructure. Oracle’s large AI build‑out has been linked to cash strain and layoffs, while the Stargate programme has already started construction in Texas even as several top OpenAI compute leaders are reported to be leaving for Meta ( ). OpenAI has also paused its planned UK Stargate datacentre, citing high energy costs and regulatory uncertainty ( ).
Artificial intelligence datacentres are hitting old-economy limits: power, cash and staffing are now slowing projects that were pitched as open-ended buildouts. (cnbc.com) A datacentre is a warehouse full of servers, networking gear and cooling equipment, and the newest artificial intelligence sites need electricity on the scale of a small city. Bloomberg reported the Abilene, Texas, Stargate campus under construction for OpenAI could draw about 1.2 gigawatts, roughly enough to power 1 million U.S. homes. (bloomberg.com) That physical scale is now colliding with corporate budgets. Bloomberg reported in March that Oracle was planning thousands of job cuts and slowing some hiring as it dealt with a cash crunch tied to heavy spending on artificial intelligence datacentres; Mint later reported layoffs affecting Oracle Cloud Infrastructure and other teams. (livemint.com, livemint.com) OpenAI is still building in Texas. Infobae reported on April 11 that construction is under way in Abilene on what OpenAI, Oracle and SoftBank have described as one of the world’s largest artificial intelligence datacentre projects. (infobae.com) But the same programme has already been pared back in other places. CNBC and Bloomberg reported on April 9 that OpenAI paused its planned Stargate project in the United Kingdom, citing high energy prices and regulatory uncertainty. (cnbc.com, bloomberg.com) Texas has limits too. Bloomberg reported in March, cited by U.S. News and Data Center Dynamics, that Oracle and OpenAI dropped plans to expand the Abilene site beyond about 1.2 gigawatts to 2 gigawatts, even as their broader 4.5 gigawatt partnership remained in place. (money.usnews.com, datacenterdynamics.com, openai.com) The staffing squeeze is showing up alongside the construction squeeze. Bloomberg reported on April 11 that three key OpenAI infrastructure leaders, including Peter Hoeschele, were leaving for Meta as Mark Zuckerberg pushed a separate datacentre expansion. (bloomberg.com) OpenAI has kept describing Stargate as a long-term buildout, and its February community post said each site would be tailored to local power, workforce and permitting conditions. That language now matches the facts on the ground more closely than the early “build everywhere” rhetoric did. (openai.com) The pattern across Oracle, OpenAI and Meta is that artificial intelligence infrastructure is no longer just a software race. It is a race for substations, financing and the small pool of people who know how to assemble a gigawatt-scale datacentre on schedule. (bloomberg.com, livemint.com, bloomberg.com)
