Gerstner predicts 2024 will see more compute deployed than all previous years combined for AI.
“We're gonna launch more. We're gonna deploy more compute in 2024 than in all previous years combined.”
Gerstner predicts next year will see more compute deployed globally than in all of human history combined.
“We're going to deploy more compute on a global basis next year than all the compute ever deployed by humanity combined. That's how much compute is going to drive these training models.”
Gerstner states inference time reasoning requires way more compute than scaling pretraining.
“inference time reasoning is way more compute intensive than simply scaling pretraining. Why don't you walk us through? We have a really interesting graph here about why that's the case”
Gerstner reports DeepSeek is massively compute constrained now, with GPU differential growing versus OpenAI for o3.
“On top of that, we learned, you know, they are massively compute constrained right now. So you've seen some tweets about this, people that are, you know, getting server delay and all this stuff”
Gerstner says DeepSeek needs 10x more compute to reach o three level assuming no breakthroughs.
“in order to get to kind of the o three level now, they got a 10 x the amount of compute. Right?”
Gerstner reports X.AI targets 50 million H100 equivalent units, requiring 4 million GPUs and 11 gigawatts.
“If you look at this one out of Elon talking about the X dot AI goal is 50,000,000 in units of H100 equivalent.”
Gerstner calls $450B Anthropic 2027 exit run rate too aggressive, sees $250B as achievable base case
“Too aggressive re Anthropic. Base case of $250 B ‘27 exit run rate (same size as MSFT & Meta) seems wild but achievable. $450 B (size of Apple & Google) seems too aggressive given compute & regulatory constraints. Margins also likely lumpy as big compute comes online. 🧐📈 https://t.co/UVbtDz2CSl”