Cloud platforms, hosting and data centers, including the power behind them: where to run it, and what it really costs.
Two different moves are being called "off the cloud": enterprises repatriating AI inference to private data centers for cost reasons, and a smaller set of workloads moving to the true edge because of latency.
Small modular reactor deals are a rounding error next to data center spending, and the one flagship project under construction won't generate power until 2030 at the earliest.
Amazon, Microsoft, Google, and Meta are building their own gas and nuclear plants for AI data centers — because grid connections now take five to eight years.
The "86% of CIOs" repatriation stat is misleading — real movement is narrow, workload-specific, and rarely announced publicly.