The $3 trillion AI infrastructure boom is spawning a new class of “all-in on AI” mega-projects. The most prominent is “Stargate,” a $500bn joint venture between OpenAI, Oracle, and SoftBank, which aims to build a network of AI-dedicated datacenters across the US and UK.
This project is part of a wave of specialized, high-performance facilities. Microsoft is building the “world’s most powerful AI datacentre” in Wisconsin and backing another AI-dedicated site in Essex. Elon Musk’s xAI is constructing its “colossus” project in Memphis.
These facilities are distinct from the “general purpose” datacenters that Microsoft is also building in places like Newport, Wales, which handle email and Zoom calls. These AI-dedicated sites are built specifically for the “exponential demand” of AI workloads, which are far more intense.
This specialization is part of what is driving the “incredible” $750bn spending spree by “hyperscalers.” However, it’s also what worries “bubble” theorists. These “all-in on AI” projects have less flexibility. If the “lofty revenue expectations” for generative AI don’t materialize—a risk, given the MIT study showing 95% of pilots have “zero return”—these specialized, expensive assets could be left high and dry.
The $500bn “Stargate” is the ultimate bet that AI is not just a feature, but the future. Its success or failure will be a key indicator of whether the $3tn boom is a revolution or a “speculative” bubble.