Sam Altman is starting to panic

TL;DR

• 1

  Uber's AI wake-up call was a $1,200 demo: Bitar says a two-hour internal presentation meant to showcase savings ended up costing more than the work it was supposed to replace, prompting Uber leadership to call it a "head exploding moment."

• 2

  The incentives were upside down: Uber reportedly tracked team token usage on a leaderboard, then had to reverse course after burning its full 2026 AI budget in just four months with little tangible output beyond summaries and specs.

• 3

  Bitar's core claim is that LLMs are structurally inefficient: He argues autoregressive models have to reread the entire context for every generated word, which makes the economics bad at a deep architectural level, not just because of temporary hype.

• 4

  He compares enterprise AI to a slot machine: The appeal comes from intermittent wins, especially in coding, where one good output keeps teams pulling the lever despite frequent confident nonsense and rising bills.

• 5

  The cost backlash is broader than Uber: Walmart's internal coding agent "Code Puppy" was reportedly curtailed, GitHub Copilot shifted customers to token-based billing with some seeing 100x price jumps, and Microsoft is described as cutting cloud licenses for similar reasons.

• 6

  This matters because OpenAI needs an IPO: Bitar says Sam Altman now faces customers openly complaining about pricing while OpenAI allegedly loses $1.22 for every $1 it makes, leaving public markets as the next source of cash.