- 3 months ago
Category
🗞
NewsTranscript
00:00All right, Stephanie, so break it down for us.
00:02What are the big trends you're seeing driving this investment and why it's still a good prospect?
00:06It's been a huge inflection in the demand towards AI that has really come with this new era of agentic
00:13AI.
00:13But what we're seeing in the markets today, despite the fact that the stock market performance has been remarkable,
00:19the semiconductor complex is up 80% since the end of March, which is just remarkable.
00:25The exceptionalism has been in earnings.
00:28And this is really due to the fact that as everyone is getting ramped up around their agentic processes,
00:35they are racing to secure compute.
00:38But the thing about this investment today, relative to the investment we saw a year or two ago,
00:43all of that investment was really around training models.
00:45And we didn't know what the ROI was going to be around training models.
00:49But now the demand and the investment is around inference.
00:52And we've been proven with agentic coding tools, like Cloud Code, that these tools are going to be very productivity
01:00enhancing.
01:01So now this CapEx has a lot more runway.
01:04And that, you know, we're now suddenly looking at CapEx potentially approaching $1 trillion as early as next year.
01:11How do you know we're going to see the ROI when it comes to the investment for inference?
01:14Not everyone's going to have the same ROI.
01:16And I think that's definitely going to be evident.
01:19You know, there won't be all winners and losers.
01:21But the fact that you have, I mean, the agency that has been achieved around coding in and of itself
01:27is transformative.
01:28And it's not the only thing that is going to be automated by AI.
01:33It is just the first real proof point.
01:35We have these tech firms essentially writing no code themselves.
01:38I'm Vibe coding myself.
01:39I avoided every Python class they tried to throw at me.
01:42And you've been successful in Vibe coding?
01:44Yeah.
01:45I mean, I think the question is, like, I've been successful on the research side.
01:49You know, the challenge for my employer will eventually be like, or if employers at large is, what is that
01:55ROI look like at the bottom line?
01:57And that is a lot more to do than just curious employees playing with these tools, but actually a more
02:03systematic rethinking, re-architecturing of labor.
02:09And that is really yet to come.
02:10The reason I asked about the outcome of the Vibe coding is because I did some, I tried to do
02:14some Vibe coding.
02:14The Vibes were bad.
02:15I knew that's where this was going.
02:17And it didn't work out.
02:18The Vibes were mediocre at best.
02:19And I'm still doing this whole show documentary.
02:21It is a new skill to learn.
02:23By hand.
02:23So I think I was, I wasn't using the right software.
02:25I was just using, like, an enterprise version of ChatGVT.
02:28What I need to use is, like, Cloud Code.
02:29Okay, wait.
02:30But to that point, when we did this show together a little while ago, you were utilizing some AI tools.
02:35I was not, and we were kind of doing it side by side.
02:38And a lot of times, I was able to find the answers faster just going the old school way.
02:43And I'm wondering, if you can talk to us, I'm seeing a little bit of a disconnect now between what
02:48Tim was just talking about, the actual ROI for the product here.
02:51Because we've got some reporting, you know, Microsoft has canceled a lot of its Cloud licensing.
02:56Uber, excuse me, Uber burned through its entire 2026 AI budget in four months.
03:01You've got a Fortune 20 CEO ordering token spending to be dramatically slashed.
03:07It seems like there are places it works.
03:09But as these companies are discovering, there's a lot of places it's not working.
03:12And yet we're seeing this spike in investment on the hard end, on chips, on storage, on memory.
03:17So I want to take a step back.
03:19And, you know, as an economist, like every new technology that enters the scene, it has this period of, this
03:25is the J curve, right?
03:26It goes negative in the beginning because you need to spend more money on the technology than it's making for
03:30you.
03:31People need to figure out how to use the technology.
03:33That takes time.
03:34Self-discovery, business discovery, all of that should be expected.
03:38We're just in a market where, yes, so much is relying on AI.
03:41So we're rightfully being very, you know, inquisitive and questioning what that ROI is going to be.
03:46But we got to let this thing play out a little bit.
03:49You know, when you first use a new tool, right, you're not going to be immediately more productive and effective
03:54with it until you build your own skills around Claude,
03:57until you sit down and workshop agents that work for you and your workflow and your audience.
04:02But all of that takes either your own time or it takes your business's time to actually engineer all of
04:08that.
04:08I think what has challenged these budgets right now is this trend of token maxing, you know.
04:14Because AI was so heavily subsidized, you could just use as much as you can.
04:19And that was the badge of honor.
04:21Throw it all at the wall and see what works.
04:21Throw it all at the wall, which is great in the context of we do need to experiment with this.
04:26Everyone needs to just get on the bandwagon, you know.
04:30But it's not efficient when we're in a world of very tight compute capacity.
04:35So it raises the question that I ask everybody because nobody knows the answer to this, including former Fed Chair
04:42Jay Powell,
04:42when he was asked about this all the time, which is the productivity gains that we will see and how
04:47people and economists are quantifying those.
04:50Right now, how are you quantifying those?
04:52In terms of the aggregate data, we don't see a lot of evidence, which is also to be expected.
04:59Don't see a lot of evidence for productivity gains?
05:00Well, I guess in terms of the broad macro.
05:02Like you said, Christina, we're like.
05:04In terms of the broad macro.
05:05Okay.
05:05Because this is the thing.
05:06Again, put an economist hat, right?
05:08When you're in the early innings of a transformative technology, you're going to see pockets of significant productivity.
05:14You can see it.
05:15You hear anecdotes.
05:16You have companies.
05:17You have the new era of firms that are just founded by one person and they're only one person.
05:21You know, all of those are examples of really significant productivity gains.
05:24Is that scaled across corporate America?
05:26I mean, the tech sector accounts for 50% of the S&P 500, 2% of the labor market,
05:33and just about 5% of GDP.
05:36So, you know, there is a big disconnect there between what is actually diffused in the broad economy.
05:41All of that is to come.
05:42The early signals do tell us a pretty constructive picture of what is to come.
05:47And I will also say, beyond just the innate productivity of all of us being more productive at work, there
05:52is the capital investment wave underway.
05:55And that, we think, is the most tangible impact on the economy right now because it's going into the real
06:01economy.
06:01It's going to real hardware and infrastructure, and that is showing up at the data.
06:05Is that why you think we're still seeing that hard investment on almost the back end of this technology?
06:10Because they figure that we'll figure out, Tim and I will figure out how to use it at some point,
06:15but they want that infrastructure to be there when that moment comes.
06:18Well, I think the demand is already there for that infrastructure.
06:22I mean, every earning season, the hyperscalers are saying, you know, our revenues would be higher today if we had
06:26the capacity to service it.
06:28So, demand is accelerating massively.
06:30We're still pretty early footholds of usage of these tools.
06:34I mean, how many agents do we interact with on a day-to-day basis?
06:38How many do you think we will be interacting with in five, ten years from now?
06:41We haven't even gotten into the world of physical AI.
06:44Healthcare is often, you know, Jensen Wong, every time he gives an interview, he's like, it's going to be the
06:47one industry that is most impacted.
06:49And there's no killer healthcare AI stocks just yet, you know, at least in the broad scheme of it all.
06:54So, there's a lot more to come.
06:55Do you have any agents acting on your behalf yet?
06:59I have a few, like, research agents I have to call on specifically.
07:02Yeah.
07:02But they are, they've been trained, you know, as I would train my own research analyst to prioritize data sources.
07:10And are they, like, I've heard different feedback from how effective the work has been.
07:15Sometimes it's like, okay, I've heard people say, yeah, it's like kind of an intern, you know, you need to,
07:19you know.
07:20Check its work.
07:21Yeah, you need to check the work.
07:22You do need to check the work.
07:22But they're eager.
07:23Yeah.
07:25What would you say?
07:26Like, how would you characterize it?
07:27So, I have, there's this great chart in one of our decks showing hallucination rates by Frontier Models.
07:32And they range from, like, 25%, the best you can get, to 90% today.
07:38Okay?
07:38But the trick is, what is so great about these reasoning models is you can just ramp up the reasoning.
07:44You can ramp up the iteration.
07:45So, you have one response that it gives you, right?
07:48And then I have it go through a number of skills around fact-checking, around voice correcting, around, you know,
07:54poking holes at what I've just written.
07:56And if you have it iterate and iterate on its own before it even gets to you, what you get
08:01is going to look a lot better than the standard, you know, generative AI query that you were using and
08:07getting a year ago.
08:08So, we have made a lot of progress.
08:09Yes, the question now is around orchestration, right?
08:12Because hallucinations will never be zero, nor are they zero for humans, you know?
08:16But I think the question is, how can we manage that?
08:18I would like to say I hate all of that.
08:20But I do understand.
08:21I do understand.
08:21All right.
08:22It's a crazy world.
08:22I want to ask in the bigger scope.
08:24The U.S. is still really dominating the high momentum aspects of that AI portfolio.
08:28But it's not a big club.
08:30Not that many countries are in this portfolio.
08:32Are there other places that you can see strategic competition coming up?
08:35And what does that mean for the field overall?
08:36Yeah.
08:36Well, I will say, even though it's not that many countries, the fact that it is more than the U
08:40.S. right now is quite important.
08:42I mean, in emerging markets, we've looked at this.
08:45And if you look at the tech sector weighting in EM relative to the U.S., EM is now more
08:50weighted towards tech than the U.S.
08:52Oh, wow.
08:52Because of the dominance that you've seen in Korea and Taiwan and just a few companies there, right?
08:58But it is more than that.
08:59The EM Asia complex is so central to so many pieces of the semi-ecosystem and also all of the
09:07equipment that is going to be needed around, you know, the peripheral execution and implementation of these tools.
09:13And then beyond that, materials.
09:15You know, if you're selling copper in this market, like, you're pretty well positioned, right?
09:20And a lot of those economies are in Latin America.
09:22So we have seen a global broadening out of AI.
09:25Things are still concentrated in the sense of it all.
09:28But, I mean, the world outside of AI is a lot less exciting than the structural growth dynamics that we
09:32are seeing right now.
09:33Is there an area geographically in this world you don't want to be invested in?
09:39Well, I think there's some areas that I'm definitely a lot more cautious in.
09:42I mean, look, you know, I mentioned Latin America just right now.
09:44Like, there's some, like, you know, pockets or resources, you know, resource exports do well.
09:49But then, you know, those economies have not been very well equipped at harnessing new technologies or even advancing where
09:57the puck is going.
09:58Also, a lot of political swings throughout that region that could change policy and government support and import, export, all
10:04those things.
10:04And I think, look, the risk around, you know, the Strait of Hormuz is not just evaporated.
10:09And there are some economies that are going to be more impacted.
10:12I think Europe definitely isn't as benefiting from being at the frontier of AI and they're more sensitive to this.
10:18So some more caution there.
10:20But I think, you know, exposure in the broad sense of it all makes sense in a globally diversified portfolio.
10:25Stephanie Aliaga, good to see you.
10:27Thanks for coming in.
Comments