00:00Talk to us about costs, though, whether when you think about humanoid robotics in particular, whether that's something that's going to be scaled, what that timeline actually looks like.
00:10I have to say it's hard to say, but there's one question I think has been on the mind of a lot of investors, which is, is there an AI bubble?
00:19And robotics a little bit further in the future, but speaking to is there an AI bubble today, let me share with you my perspective on that.
00:26I'm going to make a nuanced answer because this is important. I think we're under-invested at the application layer.
00:33There's a lot of capex going into the AI model training, you know, beta center build-outs, but for that to work out, the applications built on top have to be even more valuable.
00:42And I see a lot of investment opportunities that are signaling room for growth at the application layer.
00:46So I'm not worried about a bubble there. A lot of concerns about the capex investment, and this is my view.
00:51What I'm seeing is that there's a lot of unmet demand in AI inference. So people train AI models that cost hundreds of millions of dollars or more, and they use the models to serve up results.
01:02That's called inference. There's a lot of unmet demand in inference. So we definitely need more data centers built out for that.
01:08I'm not worried about lack of demand there. It's possible to still have people lose money if we overbill, but there I think I feel confident about.
01:15And the only question that remains is, for all the capex going into AI training, are we going to overbill for that?
01:22That one, I think, is more of an open question, because with, say, a lot of Chinese open models being released, they're very effective.
01:29You know, what if you invest a lot in training, but the value is degraded by an open competitor?
01:34Well, I am curious about that, too, just the kind of the distinction between the way that the U.S. is kind of building out its AI capabilities, U.S. companies, I should say,
01:42and what we're seeing in China, which seems to be more of an open source type of model.
01:47And I think some people have kind of compared it to, you know, say, the Apple closed Apple ecosystem versus, say, Android.
01:53So far, we're starting to see a lot of companies switch to models from Chinese makers,
01:59because obviously it's just cheaper and in some cases even free.
02:02Does that become an impediment for U.S. progress?
02:06So I think that China's release of many very high quality open models,
02:11it gives China a lot of soft power, because if someone asks a question,
02:15those models will naturally reflect, say, China's perspective.
02:18And it does call into question why the U.S., we can't seem to get our act together to release more open models.
02:26By the way, just like 10, 15, 20 minutes ago, my friend, Yanli Kun,
02:30who was the founding director of FAIR, Facebook's AI research lab, just announced his departure.
02:36And I think for a period of time, did really good work releasing open models.
02:41But now we feel like we're a bit behind China, and that's unfortunate,
02:45because that really levels the playing field, gives researchers and developers access to tools
02:50they can then use to build amazing applications on top of.
Be the first to comment