Datacentres are becoming factories

Amazon is trying to build data centers more like cars than buildings. Inside Amazon Web Services, a project called “Houdini” turns the main server room into large factory-built modules so crews can assemble capacity on site faster. (businessinsider.com) That is a change in what counts as an advantage in the artificial intelligence race. For the last two years, investors focused on who could buy the most Nvidia chips, but Amazon is now treating construction speed and labor availability as bottlenecks too. (businessinsider.com) A data center is not one big empty warehouse with computers shoved inside. The expensive part is the dense core with servers, power gear, cooling equipment, and miles of cabling that all have to fit together with very little room for error. (businessinsider.com) Amazon’s answer is to move more of that work off the construction site and into factories. Business Insider reported that Project Houdini is designed to save weeks of build time and tens of thousands of labor hours by shipping preassembled sections instead of building everything piece by piece in the field. (businessinsider.com) This factory logic fits the scale Amazon is chasing. A company trying to open a handful of buildings can tolerate custom work, but a cloud provider adding artificial intelligence capacity across multiple campuses needs something repeatable enough to copy dozens of times. (businessinsider.com) Even if the building goes up faster, the next constraint is electricity. Reporting from North Carolina this week described hyperscale projects moving into rural counties while utilities and local officials scramble to handle the surge in power demand. (wunc.org) North Carolina residents are already seeing what that looks like on the ground. WUNC reported on April 10 that a single data center can require as much energy as an entire power plant produces in a year, and Duke Energy has proposed a huge capital plan tied to expected demand from these facilities. (wunc.org) The fight is no longer just over land and substations. North Carolina regulators have been pushed to examine how Duke Energy should integrate “large load” customers above 100 megawatts, which shows how data centers are starting to look like industrial plants in utility planning. (nelsonmullins.com) That is why Amazon’s construction tweak is bigger than one company’s internal project. When power hookups can take years and local opposition can slow approvals, the winner is not just the company with the best chips, but the one that can standardize, site, and energize physical capacity fastest. (businessinsider.com) (wunc.org) In older cloud booms, software scale was the magic trick. In the artificial intelligence boom of 2026, the magic trick is pouring concrete, finding electricians, reserving transformers, and getting megawatts to the fence line before your rivals do. (businessinsider.com) (wunc.org)