Hot Summer, Short Fall: Increased Power Shortfall, Increased BTM Clarity, and Addressing Rubin Rack kW Misconceptions

We raise our 2026-28 net US power shortfall (including US DC spillover into APAC) from 33GW to 38GW, driven by higher data center power demand, partly offset by greater front-of-the-meter grid capacity (20GW vs 15GW in 2026-28) and behind-the-meter (BTM) solutions. Power demand from chips sold in 2026-28 rises from 97GW to 116GW, lifting the shortfall before time-to-power solutions from 57GW to 70GW, or 38GW net after probability-weighted solutions, still roughly 33% of demand. The increase reflects higher historical demand, higher rack-level chip power (including storage, networking and other components) and proper squaring of a 75% load factor. We address four areas of pushback: BTM upside is gated by labor (the existing labor base supports only 10-20GW of annual gas additions); ~15GW of US hyperscaler demand may spill into APAC through 2030, with 6GW in 2026-28; grid-enhancing technologies and data center curtailment face utility willingness and regulatory constraints;