Updated
Updated · newsletter.semianalysis.com · Jul 9
Meta Builds 5 1GW+ AI Data Centers, Assigns 3,000 Engineers After Llama 4 Failure
Updated
Updated · newsletter.semianalysis.com · Jul 9

Meta Builds 5 1GW+ AI Data Centers, Assigns 3,000 Engineers After Llama 4 Failure

1 articles · Updated · newsletter.semianalysis.com · Jul 9

Summary

  • Five 1GW-plus “titan” clusters are now central to Meta’s AI rebuild, with Prometheus partly operational and campuses in Ohio, Louisiana, Iowa, Indiana and El Paso under development.
  • About 3,000 engineers—roughly 70% of new graduates plus senior staff—have been moved into a new applied AI engineering group to create reinforcement-learning tasks and environments full time.
  • The push follows Llama 4’s failure and includes a $14.3 billion Scale AI investment, aggressive recruiting packages and employee workflow tracking to capture high-value training data.
  • Meta says its in-house infrastructure gives it unusual flexibility to scale compute, including linking campuses across up to 2,000 kilometers, as it tries to close the gap with OpenAI and Anthropic.

Insights

Meta is tracking employees to build its AI, fueling internal crisis. Can this controversial data strategy actually create a winning model?
Meta's stock is down despite its $125B AI spend. Is this a visionary pivot or another costly Metaverse-style gamble?