Meta发布Muse Agent，AI应用转向消费端与推理需求

Last week, we spent a lot of time thinking about pacing the frontier, whether that had to happen. This week, we seemed to be entirely much more positive about the AI narrative. Is that because actually, it's not really about the frontier? The valuations are about the frontier. They're about inference, and they're about companies using AI to drive productivity and profit. Yes, I think it is the tug of war between how to regulate the worries around what does this technology mean for humanity?

Um, at the same time, um, you know, continuing to participate in the earnings benefits of this technology, we're moving to the first kind of iteration of AI for consumers, right eye to solve everyday problems. And if you look at some of the social media, uh, use cases of the muse agent, it's to reboot hotels or, uh, you know, do online shopping. And this moves away from this productivity or coding kind of based, uh, um, consumer chain right at this minute.

The question was whether the training is going to slow down and inference is going to increase. But again, as these new models are coming out, it seems that even training is not going to slow down aggressively in the near term. So I think the demand for GPUs, the demand for CPUs, the demand for memory will likely continue. This is kind of practical uses about the practical solutions that will kind of you become my assistant.

You actually do shopping. Ultimately, I think the problem I think probably are right. Earlier this year, I saw some eye agents coming along the cloud core work, and now we are coming to muse, right. So there are some stocks to benefit and to lose out. Do you want to position for a slowdown in CapEx on I you know, fundamentally, you still see a positive growth trajectory for CapEx. Maybe it's not double digits forever and absolutely shouldn't be.

But given that that medium term outlook is good, other than that bit of turbulence, we don't think it really changes the trajectory. We're seeing just this robust consumer demand that's really adding to that one question around demand. But we're seeing this rapid adoption. So that suggests more inference, more compute and more semiconductors. At the end of the day, at a time when the market is already exceptionally tight.

We think about how much chips come into data centers, and I think that's really going to continue to provide this pillar of support.