Latest / Minds, Bodies, and Terawatts / Microsoft Routes GitHub Through AWS as AI Coding Agents Overwhelm Infrastructure

Transcript

Microsoft Routes GitHub Through AWS as AI Coding Agents Overwhelm Infrastructure · Minds, Bodies, and Terawatts

• 0:00
  [Host] This is Minds, Bodies, and Terawatts, where the machines call for backup. On Monday, multiple outlets reported that Microsoft is routing GitHub traffic through Amazon Web Services because AI coding agents have overwhelmed its own Azure cloud.
• 0:14
  [Guest] GitHub COO Kyle Daigle posted on X in April that commits were on pace for fourteen billion this year. In all of 2025, the total was one billion. Pull requests opened by AI agents jumped from four million last September to seventeen million by March.
• 0:32
  [Host] Fourteen times last year's commit volume. But this isn't just a demand spike. GitHub was mid-migration from its own data centers to Azure, deadline 2027. The agent surge hit while the house was half-renovated.
• 0:45
  [Guest] Less than half the platform traffic had moved to Azure by May. The agents don't follow human usage curves — Cursor, Claude Code, Copilot, Devin run continuously through the API, never logging off, never taking weekends. GitHub's capacity planning was built around developers who do both.
• 1:04
  [Host] Nine service incidents in May. June availability dropped to roughly eighty-eight percent — nearly three hours of downtime every day on a platform that over a hundred million developers depend on.
• 1:16
  [Guest] Banks deploying security patches, retailers pushing app updates — all stalled. Microsoft called its biggest cloud competitor for help rather than wait for the concrete to dry on new Azure capacity.
• 1:28
  [Host] Here's what complicates the more-servers story. Former GitHub CEO Thomas Dohmke didn't just watch this happen. He left and started a company called Entire, a platform designed from scratch for agents. His bet is the architecture itself is wrong.
• 1:44
  [Guest] He may be right about the shape of the problem. A human developer opens a few pull requests a day. An agent fleet opens hundreds, and each one triggers builds, tests, and deployments through the same pipeline. The workload profile changed faster than any data center buildout can follow.
• 2:03
  [Host] The four biggest cloud companies are spending over six hundred billion dollars on data centers this year, and demand is still outpacing supply.
• 2:12
  [Guest] That's the feedback loop. Agents write code that needs infrastructure, and building that infrastructure requires more agents. Whether capacity catches up before this becomes a reliability crisis for every company shipping software through GitHub is the open question.
• 2:27
  [Host] Patrick Deglon calls this the Agentic phase in Unscarcity — when AI stops generating text and starts executing tasks nonstop. GitHub just showed what happens when machines that never sleep hit infrastructure that still thinks in business hours.
• 2:42
  [Guest] For more on how AI is reshaping the infrastructure it runs on, head to unscarcity dot ai. The book is on Amazon, the audiobook is on Spotify. This has been Minds, Bodies, and Terawatts.