

Aseem
10.2K posts

@aseemchandra
Building https://t.co/qvDBh3ga2O to connect AI to your enterprise data | startups, investing and AI bottle-necking









Leopold Aschenbrenner, manager of Situation Awareness with roughly $10B AUM, said AI training clusters could scale from roughly 25k $NVDA H100s during the GPT-4 era to potentially “100M H100 equivalents” by 2030, requiring over 20% of total U.S. electricity production. More importantly, that trajectory continues reinforcing the long-term infrastructure demand tied to GPUs and ASICs from $MRVL and $VVGO alongside the broader power, datacenter $CRWV $IREN, and hyperscaler buildout happening across companies like $AMZN and $GOOGL as AI systems continue scaling globally.



