Hyperscale vs. control layer
Coverage of OpenAI’s Stargate and similar hyperscale projects highlights a split: giant compute campuses are clustering where land, power and political backing are easiest to secure, while leadership, R&D and orchestration functions cluster in talent hubs. The pattern suggests compute goes where infrastructure is cheapest and permissive, and management and product teams keep locating near engineering talent. (motherjones.com) (techweez.com)
The artificial intelligence boom is splitting in two: giant server campuses are moving to power-rich regions, while the people directing them stay in talent hubs. (openai.com) OpenAI said in January 2025 that Stargate would invest $500 billion over four years in United States artificial intelligence infrastructure, with SoftBank, OpenAI, Oracle, and MGX as initial equity funders. In October 2025, OpenAI said Stargate’s planned capacity had reached nearly 7 gigawatts across Abilene, Texas, five new U.S. sites, and related CoreWeave projects. (openai.com 1) (openai.com 2) Oracle’s Abilene campus is tied directly to the Texas grid, where the company says it can tap a large wind supply and use natural-gas turbines for backup power. Techweez reported on April 17 that a modern hyperscale data center can draw 20 to 100 megawatts, with cooling alone taking roughly 40 percent of that load. (oracle.com) (techweez.com) OpenAI’s own hiring shows the other half of the map. Current job listings for software infrastructure, communications, and user operations roles are based in San Francisco and use a hybrid in-office schedule. (openai.com 1) (openai.com 2) (openai.com 3) That division tracks the economics of artificial intelligence. Land, transmission lines, water, turbines, and permits decide where the machines go; model design, product decisions, and orchestration software stay near dense pools of engineers and executives. (openai.com) (oracle.com) (openai.com) OpenAI’s San Francisco infrastructure roles describe work on caching, metadata stores, file storage, and workflow orchestration, the software layer that routes jobs across hardware. The physical hardware itself is increasingly being bought in bulk from outside providers and campus operators. (openai.com) (coreweave.com) CoreWeave said in March 2025 that its first OpenAI infrastructure contract was worth up to $11.9 billion, and in September 2025 it said later expansions brought the total to about $22.4 billion. That arrangement let OpenAI add compute without putting every chip inside facilities it owns or runs directly. (coreweave.com 1) (coreweave.com 2) The same pattern is spreading outside the United States. OpenAI launched Stargate United Arab Emirates in May 2025, and MGX said this month that it is backing Stargate UAE and a 1.4-gigawatt artificial intelligence campus near Paris expected to break ground in 2026. (openai.com) (mgx.ae) Mother Jones described the result this week as a hyperscale buildout tied to political influence and control over power, land, and financing. The map now looks less like one company in one city than a chain: compute in permissive infrastructure zones, command in places like San Francisco. (motherjones.com) (openai.com)
