AWS hitting physical limits
Amazon is running into a new bottleneck: demand for AI compute is outpacing the rate at which data-centres can be built, so AWS is experimenting with modular construction and new site launches to speed capacity expansion. Reports name “Project Houdini” for modular build-outs and note new U.S. sites moving into early construction, which turns regional capacity into an operational constraint for latency-sensitive AI services. (businessinsider.com) (vicksburgpost.com)
Amazon Web Services is discovering that the limit on artificial intelligence is no longer just chips. It is concrete, steel, electricians, and how fast a new building can come online. (businessinsider.com) Inside Amazon, a project called “Project Houdini” is trying to move part of data-center construction off the job site and into factories. Business Insider reported that the plan is to preassemble large server-room modules so Amazon Web Services can install them faster once the site is ready. (businessinsider.com) The immediate problem is speed. Business Insider reported that Amazon expects the modular approach to cut weeks from construction schedules and save tens of thousands of labor hours on some builds. (businessinsider.com) This is happening because artificial-intelligence demand is arriving faster than the old build process was designed for. Amazon chief executive Andy Jassy said on April 9, 2026 that Amazon expects about $200 billion in capital spending in 2026, with most of it aimed at artificial-intelligence infrastructure. (msn.com) Jassy also said Amazon Web Services had already reached an artificial-intelligence revenue run rate above $15 billion in the first quarter of 2026. That means Amazon is not building for a hypothetical boom anymore; it is trying to catch up with one that is already paying. (msn.com) A cloud region is not one giant warehouse. Amazon Web Services says a region is a geographic area made up of at least three separate Availability Zones, which are isolated groups of data centers designed so one failure does not take down everything. (aws.amazon.com) That geography matters for artificial intelligence because some workloads are sensitive to delay. Amazon Web Services says its global network uses nearly 20 million kilometers of fiber, but even with that backbone, a customer in one region cannot ignore where the nearest computing capacity actually exists. (aws.amazon.com) You can see the physical expansion in Mississippi. The Vicksburg Post reported on April 10, 2026 that Amazon Web Services’ “U.S. 61 South Warren” site in Warren County had entered early construction, with the main entrance road planned off Old Cain Ridge Road. (vicksburgpost.com) That Vicksburg project is part of a larger $3 billion campus Amazon Web Services announced in November 2025. Data Center Dynamics reported the campus would begin construction in 2026, adding another piece to Amazon’s push for more regional capacity in the United States. (datacenterdynamics.com) Project Houdini does not remove every bottleneck. Reports on the plan note that even if server rooms arrive like prefabricated house sections, Amazon still has to secure land, utility hookups, and enough power for buildings packed with artificial-intelligence hardware. (businessinsider.com) So the story is not that Amazon Web Services wants more data centers. The story is that in 2026, cloud growth is starting to look like shipbuilding or highway construction: the winners are the companies that can line up factories, labor, power, and local permits fast enough to turn demand into working capacity. (businessinsider.com)
