Two stories this week, separated by an ocean, share the same skeleton. SoftBank announced up to €75 billion to build data centers in France, targeting 5 gigawatts of capacity. Meanwhile in Utah, Kevin O'Leary wants to build Stratos, a 7.5-gigawatt AI infrastructure project across 10,000 acres, and he wants you to think it can be beautiful. Together, these projects represent something beyond normal capital expenditure. They are territorial claims made in the name of compute.
Infrastructure as Aesthetic and Political Argument
O'Leary's insistence that a data center campus can be "beautiful" is not naive. It is strategic. The framing borrows from the playbook of sustainable architecture and civic infrastructure. If you can make people feel proud of a power-hungry slab of servers, you reduce regulatory friction. SoftBank's French announcement plays a similar game, arriving conveniently as European AI sovereignty anxiety peaks. Both moves dress infrastructure colonialism in the language of partnership and progress. The land being used, the water being consumed, the energy being routed: these are political decisions with permanent consequences, sold as technical ones.
The Scale Problem and the Capital Behind It
A 2023 paper in Nature Energy by Masanet et al. found that data center energy consumption projections routinely underestimate demand growth by factors of two to three, a trend only accelerating with generative AI workloads. The capital flows here are staggering and largely opaque to public scrutiny. TurboFund's live VC intelligence shows infrastructure and AI compute plays among the fastest-growing investment categories in 2026. Meanwhile Fast Company's parallel story on documenting disappearing tropical glaciers sits in quiet, devastating counterpoint: the infrastructure being built to power AI is operating in direct ecological tension with the world the models are supposedly helping us understand.