AI needs power, not just models
Big AI players are rethinking compute as an energy and infrastructure problem—OpenAI paused its Stargate UK data‑centre project citing energy costs, while discussions are underway about alternative power suppliers and Gulf compute hubs are attracting investment. (financialpost.com) (xix.ai) (ndtvprofit.com).
OpenAI just hit pause on a big United Kingdom data-center push, and the reason was not chips or software. It was electricity prices and the rules around getting enough power onto the grid. (bloomberg.com) (cnbc.com) That is a useful way to read the artificial intelligence boom in 2026: a model is only as big as the power plant behind it. A modern data center is less like an office building and more like a factory that turns electricity into answers. (openai.com) (cnbc.com) OpenAI’s Stargate project was introduced on January 21, 2025 as a plan to invest up to $500 billion over four years in artificial intelligence infrastructure, with OpenAI, SoftBank, Oracle, and MGX involved. The first promise sounded like a financing story, but the buildout quickly became a land, grid, permits, and utility story. (openai.com) (group.softbank) By January 2026, OpenAI was already trying to calm local fears with a “Stargate Community” plan that said each site would “pay its own way on energy” so nearby households would not see higher power bills. Companies do not make that promise unless they know electricity has become the political choke point. (openai.com) (finance.yahoo.com) Now the company is looking past utilities and talking directly about power supply at the source. Reuters reported on March 23 that Sam Altman stepped down from Helion Energy’s board as OpenAI and the fusion startup began exploring work together “at significant scale.” (money.usnews.com) Axios reported that the Helion talks could start at 5 gigawatts by 2030 and rise to 50 gigawatts by 2035. TechCrunch said that first tranche alone could equal about 12.5% of Helion’s projected output. (axios.com) (techcrunch.com) A gigawatt is the scale people use for power stations, not office parks. When an artificial intelligence company starts shopping in gigawatts, it is telling you that “compute capacity” now means substations, transmission lines, and generation contracts. (techcrunch.com) (openai.com) That is also why the Gulf is getting so much attention. NDTV Profit reported on April 9 that Microsoft, Google, Amazon, and other large technology firms are pouring billions into Gulf Cooperation Council countries as artificial intelligence compute hubs, because those countries can pair capital with large energy projects and fast infrastructure buildouts. (ndtvprofit.com) The Gulf story has another layer: geography now affects server strategy. A regional truce can lift technology stocks because investors see lower risk around the same campuses, cables, and power assets that artificial intelligence companies need to keep expanding. (ndtvprofit.com) (agbi.com) Put those pieces together and the race looks different from the one most people imagine. The winners will still need better models and more chips, but they will also need cheaper electrons, friendlier regulators, and places where a multi-gigawatt campus can actually get built. (bloomberg.com) (axios.com) (ndtvprofit.com)
