Stargate: The Megaproject

In January 2025, OpenAI, Oracle, and SoftBank announced Stargate, a roughly $500 billion bet to build out US AI data centers, about ten gigawatts of capacity, the output of ten nuclear plants, more than the inflation-adjusted cost of the interstate highway system. By late 2025 the partners claimed the flagship Abilene, Texas site was online and running ahead of schedule. It is not a product launch. It is a wager that demand for AI compute grows fast enough, for long enough, to justify pouring the physical layer before the demand is fully proven.

For founders and operators, the interest isn't the dollar figure, it's the structure of the risk underneath it. When this much capital runs through this few decisions, the margin for error collapses and there's no diversification to fall back on. The decision it sharpens is naming the single load-bearing assumption your own biggest bet depends on, and spotting the earliest signal that it's wrong. What separates the megaprojects that survive from the ones that strand is the thread the app has you pull.