06-reference

stratechery interview power for ai

Wed Jan 07 2026 19:00:00 GMT-0500 (Eastern Standard Time) ·reference ·source: Stratechery ·by Ben Thompson
ai-infrastructureenergydata-centersgrid

“Interview: Jeremie Eliahou Ontiveros and Ajey Pandey About Building Power for AI” — Ben Thompson

Why this is in the vault

Covers how AI labs are solving the power constraint by building on-site generation — relevant background for understanding AI infrastructure scaling and cost trajectories.

The core argument

Interview with two SemiAnalysis researchers (Eliahou Ontiveros, Head of Datacenter & Energy Infrastructure Research, and Pandey, energy infrastructure analyst) about how AI labs and hyperscalers are increasingly building their own power generation on-site, inspired by Elon Musk/xAI’s approach. Key topics include the unique electrical load profile of AI workloads, grid challenges, behind-the-meter generation, and what this means for US electrical infrastructure. Thompson’s takeaway is optimistic: AI demand is forcing long-overdue innovation in US electrical grid infrastructure, connecting to his thesis that “bubbles really do have benefits.”

Mapping against Ray Data Co

No direct connection to RDCO’s current workstreams. Useful background context for understanding AI cost curves and infrastructure constraints that affect the tools we build on.