- 14 hours ago
Category
🗞
NewsTranscript
00:00GKG's portfolios have recently turned significantly underweight tech on concerns of deteriorating
00:06fundamentals. And a part of what Brody was discussing, right, I know that we can talk
00:10about Microsoft here because there's some exposure in the funds, but the idea that you rely on the
00:15NeoCloud so it doesn't show up on the balance sheet in the CapEx, what was your sort of reaction
00:21to hearing that? Yeah, so I think what it speaks to from a Microsoft perspective is sort of that
00:27CapEx sort of notion where you don't necessarily want to spend all of your CapEx on sort of an asset
00:32that could depreciate or may not necessarily be as advantageous for you on a go-forward basis.
00:36There's a lot of things that are evolving really quickly when it comes down to these things.
00:42But if I were to take a step back, and you had mentioned sort of we've become a little bit more
00:46cautious, so to speak, on sort of these names in general and from an AI perspective,
00:52I think one of the reasons for that is there's been a whole lot of spending on the CapEx side
00:56of things, $600 billion in spending on CapEx. And really, if you take out the infrastructure
01:00spending side of this, there's only been about $30 billion in revenue that have been generated
01:04off of this. So our issue here is that there is a lack of headroom sort of returns that are coming
01:10through on a lot of these types of businesses over the course of time. Now, Microsoft, to its credit,
01:14has a software business. They have sort of steady earnings. They have been able to deliver some
01:18decent results over the course of time. But we are becoming more skeptical about sort of a lot of
01:22things on the AI side. Where are these returns coming from? You open AI in particular? Yeah,
01:27I'm sorry. I mean, Brian, we're looking at a note that your team put out September the 11th,
01:31you rang that alarm bell basically saying, we believe that the sector stands at a significant
01:36inflection point. And everyone's making a one way bet as you see it on AI mania. And they're ignoring
01:42the alarming fundamentals for you. The alarming fundamentals are that there is a lack of revenue
01:47today. Is it not? Therefore, can you not just make that bet that eventually open AI will make
01:53$300 billion worth in revenue by 2030 that vindicates the amount that they have to spend on all this
01:58compute? So I think it becomes hard because if you actually look at the data behind this,
02:03you look at open AI, they have about a 2% conversion rate in terms of people that actually want to pay for
02:08the service. That means 98% of people that use open AI aren't actually paying for it. So now they have
02:13700 million users globally. About half of those are coming from the emerging markets. And what's
02:19interesting there, if you think about the unit economics of sort of cloud, and I'm sorry, sort
02:23of the AI sort of side of things, is that this isn't like SaaS. So this isn't like a CRM business
02:29where you add additional users and it all sort of the revenue drops to the bottom line. There is a
02:33high cost to compute that comes along with this. So you need to generate some sort of revenues off of
02:38each one of these users. Now with half of that user base coming from the emerging markets,
02:43a significant chunk coming from India, for example, if I can get a 5G telephone plan within
02:48India for less than $10 a month, do I really think that folks are going to pay $20 a month for
02:53a subscription to ChatGBT? Okay, here we go. I like this because I think we're looking at the
02:59same data set. So the top story today is open AI being valued at $500 billion on the latest secondary
03:05round. But the big question here is where the future for open AI lays on subscriptions or on
03:12enterprise, the data point that I look at is that it has a user base. We know it's like $700 million
03:17monthly, right, Caro? How much of that user base is free and then converted to being a paying
03:23subscriber? Because that kind of answers all your questions.
03:27Yeah, so that's exactly the stat that I was referring to. From what we've seen, only about
03:312% of those folks actually convert over to being paying users.
03:352%?
03:35Which means 2%, yes. And so that means that 98% of folks that are using this actually don't find
03:41enough value to actually sort of spend money on this. Now, going back to that India data point that
03:46I was referencing earlier, if you think about those folks and you need to sort of charge,
03:50call it $20 a month for some sort of subscription to make this break even profitability or even come
03:55close to that. In India, you have, you know, Barthi Airtel that has a deal with Perplexity that offers
04:01this service for free across Parthi. So it becomes really hard to see where the monetization path
04:06comes on a lot of these things over the course of time. Now, the other side of this is on the
04:11enterprise side. So a lot of people say, OK, well, maybe it's more of a B2B sales. And there's a lot
04:16of things that people are investigating or looking at from the enterprise side in terms of I can use AI
04:20to get efficiencies and things like that. The MIT study, I think we all know that by now being quoted
04:26in terms of the lack of sort of effectiveness in a lot of those. We've talked to other tech
04:30consultants recently. In fact, one of the big three consultant firms said that 85 percent of the
04:35projects that they were working on. So the 400 projects they've done on a year to date basis,
04:3985 percent of those projects were absolutely useless. I said 15 percent generates some sort of benefit.
04:44You're literally sort of echoing exactly the conversation that we had with Syntheja's CEO
04:51founder yesterday saying basically only about 15 percent or even vaguely working and only five percent
04:56actually working well. But Brian, can you not think that eventually they will work? And that's
05:02actually more to implementation issues rather than actually the fact that they're not adding value.
05:08So I think you have to show me sort of the math and the monetization and the pathway for that
05:13working. There's a lot of data organization that needs to happen. There's a lot of things that need
05:18to happen to get to that point. And what we're seeing in terms of large language models is we're kind
05:22of peaking out in terms of what the capabilities are. And you saw that sort of transitioning from
05:26GPT-4 to GPT-5. It's not simply you throw more compute at the problem and you solve bigger and
05:33more complex large language models. There's more post-training types of things that are coming
05:37through and you're seeing that the models are actually peaking out in terms of their effectiveness.
05:41And at the end of the day, these are extrapolators. So you're guessing what the next letter is or the
05:46next word is based on a large training set. It can't think for you and it can't sort of make those
05:52decisions for you. And I think you're starting to see that within the enterprise side of things.
05:56Now to pivot to an even bigger question here is where has this spending actually come through
06:01and where are we actually seeing the money spent? And that is actually more on the hyperscaler side
06:06of things. So actually providing the cloud services where the data is coming through
06:10and where you're paying for it on the cloud side of things. And this is also fairly concerned.
06:15I'm sorry to jump in because we'll run out of time, but linked to that, including the top-line
06:19growth discussion, what we've asked private market and public market participants a week
06:24is their assessment of the role debt is playing in all of these infrastructure projects and how
06:29worried or not one should be about that. Yeah. So debt or no debt. One of the points that I was
06:36trying to make earlier was that if you look at the pricing dynamics within cloud, they're coming
06:40under a lot of pressure and there's a lot of increased competition that's coming through. And I think
06:44that's where we're really struggling on a lot of these things, where you have Oracle coming in and
06:49undercutting price by 40% to 70% on a lot of these enterprise deals. And you're seeing that dragging
06:54down in terms of the pricing across a lot of the cloud players, including like an AWS and things like
06:59that. So it becomes a less profitable venture. The switching costs are becoming a little bit lower
07:04and the economics aren't quite as good. It's becoming more commoditized. And that's where we
07:08really struggle because there's a whole host of investments that's happening in this area and
07:12it's becoming increasingly commoditized, similar to the fiber build out, so to speak,
07:15back in the dot-com boom and bust cycle. The commoditized element of concern. What about
07:21the circularity argument that we keep hearing? And that feeds into the debt question in many ways.
07:26Absolutely. So then the other question to ask is if this is such a fantastic investment on a go
07:31forward basis, why do you have participants in the ecosystem that are actually funding their
07:37customers? And then those cash flows are then coming back to them. So then a lot of the sort of
07:42obscure sort of arrangements and deals as well in terms of special purpose vehicles, JV structures,
07:47where are you putting some of these assets into other sort of places where you can depreciate the
07:52debt or you can depreciate the assets within those other vehicles and it's not sitting directly on
07:57your balance sheet. So this tends to happen later in a cycle where you start to see a little bit more
08:02aggressive accounting coming through and you start to see some of these things that are starting to
08:06become a little bit more obscure. That has us concerned that we can't see a true sort of trajectory
08:11where the economics are coming through. That's what has us concerned right now.
Recommended
6:33
|
Up next
1:58
1:57
5:51
4:31
2:45
5:43
5:18
5:38
5:32
5:36
3:58
4:34
6:21
7:38
4:54
8:45
5:55
4:47
1:46
2:27
3:52
6:24
Be the first to comment