Tom Arnold

Tech crowd keeps leaning into this off grid is the only way thesis and I disagree with this notion. The exhibit A example of xAI colossus even overstates what’s actually happening at Colossus. xAI did not build a permanently “off-grid” AI campus because the grid was unusable. Colossus already has 300 MW of TVA interconnect approved across two substations xAI funded, plus contractual obligations to curtail during periods of grid stress. That is a grid-tied architecture with supplemental on-site generation, not an islanded power system. The turbines solved a time-to-power problem. They accelerated deployment while substations and transmission infrastructure were being built. That’s very different from “do this all off grid.” And economically, most large-scale AI load will still want to be grid tied long term. I have made this point several times but running your own generation fleet at hyperscale is expensive and operationally complex: fuel logistics, maintenance, N+1 redundancy, permitting, emissions compliance, balancing load/supply continuously, multi-year equipment lead times, etc. You don’t just figure that out or want to do that overnight. Amin Vahdat, chief technologist of Google AI straight up confirmed this on a podcast and they literally bought a hybrid/island power developer. BTM generation absolutely matters. It’s becoming a critical bridge solution during a period where utility interconnect timelines are badly lagging demand growth. But I’d strongly push back on the idea that the steady-state future is fully off-grid hyperscale campuses. The industry is converging toward hybrid architectures: grid interconnect + captive generation + storage + demand response.

Today I announced new legislation to un-rig the process for San Francisco & other cities to break up with PG&E & form their own publicly owned utilities. Over the years, PG&E has managed to insert provisions into state law to make it harder to break away. SB 875 reverses that.🧵

WILDDDDD DATA