Google is Short of AI Compute And Delayed an AI Model Release

Bloomberg reports $GOOG engineers are hitting capacity constraints when they try to use AI internally. At a company guiding $180B to $190B of capex this year.

Gemini 3.5 Pro is months behind schedule. Engineers are now required to use AI to write code and there isn’t enough compute to go around. Q1 capex was $35.7B, more than double a year ago, and the CFO already said 2027 will significantly increase from there.

Grok thinks there are high odds that Google expands the AI compute deal with SpaceXAI. Anthropic Fable is less certain on Google but thinks someone big announces a deal in Q3 and more deals in Q4. Anthropic think some new major tenant for SpaceXAI deal (anyone) is announced by end of Q3 is 70–80%. SpaceXAI has signed three deals in ~six weeks and Musk has said he’s actively seeking customers. The base rate here is high. The question is who, not whether. Google odds are as high as 60%, 35-40% for Anthropic and then others like Microsoft.

Screenshot
Screenshot
Screenshot

2 thoughts on “Google is Short of AI Compute And Delayed an AI Model Release”

  1. It makes no sense that they did not reserve capacity for their own engineers. Also, I just tried Gemini without issue in thinking mode.

Comments are closed.