AI Radar

Noteworthy AI conversations and writing, summarized from publisher notes.

Behind-The-Meter Power For Datacenters: Part 1

· SemiAnalysis

The report highlights the growing trend of AI labs and hyperscalers adopting behind-the-meter (BTM) power solutions to overcome power constraints for AI compute, tracking 75GW of firm, binding orders. Microsoft has signed over 5GW of BTM capacity in 2026, while Google is deploying 930MW of off-grid turbines and 900MW of fuel cells. Anthropic and Meta have signed 300-500MW deals, and OpenAI is set to operate a 1.4GW off-grid campus. The article explains that the high value of megawatts for AI inference revenue makes paying more for faster deployment a viable strategy, as the grid struggles to keep up with demand. It also discusses the challenges and definitions of BTM power, including permitting delays, pipeline issues, and labor shortages, and explores the future of these power plants.

AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.