Core Scientific is trading coins for tokens, revealing plans on Monday to convert a 300-megawatt bitcoin mining operation in Pecos, Texas, to an 1.5 gigawatt AI datacenter campus.<\/p>\n
Core Scientific is one of several companies racing to switch from crypto mining to token farming before the bubble bursts, and recently announced plans to sell $3.3 billion of junk bonds<\/a> to help facilitate the move. CoreWeave, which has previously contracted with Core Scientific<\/a> to use capacity at its datacenter in Denton, Texas, and Crusoe are two of the most recognizable converts in the field.<\/p>\n The bitcoin miner turned AI bit barn builder began construction of the new facility earlier this year, with the first new data hall expected to come online in early 2027.<\/p>\n When complete, Core Scientific estimates<\/a> that the facility will offer about a gigawatt of leasable capacity, roughly the equivalent to the output of a large nuclear reactor. But all of that’s assuming the company can secure adequate power.<\/p>\n Core Scientific says that it’s already secured an additional 300 megawatts of power from the local utility and plans to supplement that with some kind of “scalable behind-the-meter solution.”<\/p>\n What exactly that will entail Core Scientific hasn’t said. The company certainly wouldn’t be the first bit barn builder to embrace on-site power generation to keep the lights on and the GPUs churning out tokens.<\/p>\n Over the past year, all of the major hyperscalers have embraced some kind of non-traditional energy storage or generation tech, some more exotic than others. Google, Oracle, AWS, and others are all betting<\/a> on small modular reactors (SMRs), tiny nuclear power plants, that can be deployed on site to fuel their AI ambitions.\u00a0<\/p>\n Meanwhile Meta this week signed<\/a> an agreement with Overview Energy to beam a gigawatt of solar power down from orbit, just as soon as they can lob the arrays into orbit. But, just like SMRs, that won’t happen until at least 2030.<\/p>\n Power constraints have become such a limiting factor that major model builders like AWS, Google, and xAI are now talking about building orbital datacenters. However, the economics of such a deployment remain dubious<\/a> to say the least.<\/p>\n