AI scaling now runs on three compounding scaling laws — pre-training, post-training (reinforcement learning practice), and inference-time thinking — rather than the single pre-training scaling law people previously assumed drove progress.
Jensen lays out the mechanism behind the current AI compute boom: pre-training, post-training (RL 'practicing'), and inference-time 'thinking' are three separate, stacking scaling laws, which is why compute demand keeps compounding. ✦ AI generated
Jensen Huang · BG2 Pod · 2025-09-26 · original ↗
starts at this moment · 1:54
I estimated we now have three scaling laws, right? We have pre-training scaling law. We have post-training scaling law. Post-training is basically like AI practicing... in order to do that, you've got to do inference. So now training and inference are now integrated in reinforcement learning... And then the third is inference... the new way of doing inference, which we appreciate, is thinking. So think before you answer.
verbatim transcript · starts at 1:54
1:54practicing. Yes, >> practicing a skill until it gets it right. And so it tries a whole bunch of different ways and and uh in order to do that, yeah, you've got to do inference. So now training and inference are now integrated in reinforcement learning. >> Really complicated. And so that's called post training. And then the third >> is inference. The old way of doing inference was one shot,
2:17>> right? >> But the new way of doing inference, which we appreciate, is thinking. So think before you answer. >> Yeah. And so now you have three scaling laws. The the longer you think, the better the quality answer you get. While you're thinking, you do research, you go check on some ground truth. And you you learn some things, you think some more, you go learn some more, and then you
2:38generate an answer. Don't just generate right off the bat. >> And so thinking, post-training, pre-training, we now have three scaling laws, not one. You knew that last year, but is your level of confidence this year in the inferences going to 1 billionx and where that will take the levels of intelligence is it higher? Are you more confident this year than you were a year ago? >> I'm more confident this year and the
3:00reason for that is because look at the agent systems now >> and AI is no longer a language model and AI is a system of language models and they're all running concurrently maybe using tools. Some of us using tools, some of us doing research and yeah, there's a whole bunch of stuff and it's all multimodality and look at all the video that's being generated. I mean, it's just crazy stuff. Yeah.
3:23>> It really brings us to, you know, kind of the seminal moment this week that everybody's talking about the massive deal. You announced a couple days ago with OpenAI Stargate where that you're going to be a preferred partner, invest hundred billion dollars in the company over a period of time. they're going to build 10 gigs and if they used Nvidia for those 10 gigs that could be upwards
3:42of 400 billion in revenue to Nvidia. So help us understand you just tell us a little bit about that partnership what it means to you right and why that investment makes so much sense for Nvidia. >> So first of all that I'll answer that last question first and then I'll come back and present my way through. I think that OpenAI is likely going to be the next multi- trillion dollar