The era of cloud shortages is shifting from software deficits to a literal brawl for the power grid. Project Stargate is more than just the construction project of the century; it is Sam Altman’s attempt to build a physical "moat" with his new allies—one that no competitor can leap without their own money-printing press. Five new U.S. sites under the OpenAI, Oracle, and SoftBank umbrella send a clear signal: the days of playing around with APIs and model weights are over. The industry has entered a phase of consolidating 10 gigawatts of power backed by half a trillion dollars in investment. By diversifying its partner pool, OpenAI is methodically slipping out of Microsoft’s suffocating embrace. The strategy is transparent: once algorithms become a commodity, the only real asset left will be the ownership of massive, localized power capacity.
The Geography of Compute
The expansion into Texas, New Mexico, and Ohio moves the arms race into the realm of logistics, where scale is everything. According to OpenAI reports, new nodes in Shackelford, Milam, and Doña Ana counties—alongside projects in Ohio and Wisconsin—are turning the American heartland into the primary engine for training next-generation models. The Wisconsin project, developed by Oracle in partnership with Vantage, confirms that the center of gravity is shifting toward wherever there is room for reactors and cooling systems. Sam Altman and Larry Ellison have already agreed on 4.5 gigawatts of capacity as part of a deal exceeding $300 billion over five years. This alliance allows OpenAI to scale at a pace that traditional cloud giants simply cannot match.
The SoftBank-Oracle Alliance
Masayoshi Son brings more than just capital to Stargate; he brings expertise in energy and rapid-fire construction. In Lordstown, Ohio, SoftBank has already broken ground on a new type of data center scheduled to go live in 2025. Simultaneously, its subsidiary SB Energy is readying infrastructure in Texas. The alliance reviewed over 300 proposals from 30 states, selecting locations based on the immediate availability of gigawatts. The goal is ambitious: deploying 1.5 gigawatts of power within the next 18 months. This looks like a calculated move to board the physical infrastructure train while Microsoft and AWS are bogged down in bureaucracy.
This infrastructure sprint is running ahead of schedule. OpenAI claims it will secure access to the full promised 10 gigawatts by the end of 2025.
Essentially, we are witnessing the training of frontier models evolve into a matter of U.S. national industrial policy, where the entry fee to the AGI club is now $500 billion. While skeptics debate the "bubble," the Altman-Ellison-Son coalition is building a foundation that cannot be copied or replaced by code. For businesses, the takeaway is simple: if your cloud provider hasn't secured long-term energy contracts for the 2026–2027 cycle, there simply might not be a seat left for you at the table in the future of computing.