WEBVTT - Special Episode: Here's Why AI Costs Still Worry Investors

0:00:00.280 --> 0:00:04.040
<v Speaker 1>Hi, this is Caroline Hyde from Bloomberg Tech. Today we're

0:00:04.080 --> 0:00:07.080
<v Speaker 1>sharing something a little different in your feed, an episode

0:00:07.080 --> 0:00:10.840
<v Speaker 1>from our colleagues at Here's Why, Bloomberg's weekly show that

0:00:10.920 --> 0:00:14.800
<v Speaker 1>answers one big question in under ten minutes. Host Stephen

0:00:14.800 --> 0:00:17.960
<v Speaker 1>Carroll is joined by our Bloomberg Tech europanker Tom McKenzie

0:00:18.360 --> 0:00:20.720
<v Speaker 1>to dive into a story that's right at the heart

0:00:20.760 --> 0:00:24.200
<v Speaker 1>of the tech world, the massive investments in AI data

0:00:24.239 --> 0:00:27.200
<v Speaker 1>centers and the hidden costs that come with them. If

0:00:27.200 --> 0:00:30.280
<v Speaker 1>you'd like to hear more episodes of Here's Why, you'll

0:00:30.320 --> 0:00:32.600
<v Speaker 1>find a link to the podcast feed in the show notes.

0:00:33.120 --> 0:00:34.200
<v Speaker 1>Hope you enjoy.

0:00:36.720 --> 0:00:44.000
<v Speaker 2>Bloomberg Audio Studios, podcasts, radio news. I'm Stephen Carol and

0:00:44.120 --> 0:00:46.640
<v Speaker 2>this is Here's Why, where we take one new story

0:00:46.680 --> 0:00:48.800
<v Speaker 2>and explain it in just a few minutes with our

0:00:48.840 --> 0:00:50.360
<v Speaker 2>experts here at Bloomberg.

0:00:53.640 --> 0:00:56.880
<v Speaker 3>It's ten thirty pm in this AI party. It started

0:00:57.000 --> 0:00:59.880
<v Speaker 3>nine pm now and that party goes to four am,

0:01:00.040 --> 0:01:02.080
<v Speaker 3>and the reality is like, look, this is going to

0:01:02.080 --> 0:01:04.600
<v Speaker 3>be a two to three year left in this bull

0:01:04.680 --> 0:01:07.760
<v Speaker 3>cycle for tech. The tech sector is very strong because

0:01:08.040 --> 0:01:11.920
<v Speaker 3>artificial intelligence is really a qualitative leap in the kind

0:01:11.920 --> 0:01:15.240
<v Speaker 3>of technology that we've had over the last several decades.

0:01:15.400 --> 0:01:19.520
<v Speaker 3>You're seeing an exponential growth of adoption and use of AI.

0:01:19.880 --> 0:01:22.280
<v Speaker 4>The number of applications that are going to be using

0:01:22.280 --> 0:01:23.960
<v Speaker 4>these AI is also growing.

0:01:24.480 --> 0:01:27.600
<v Speaker 2>Everyone has an opinion on where the AI frenzy is

0:01:27.640 --> 0:01:31.839
<v Speaker 2>going next. While optimism is rampant about the technologies potential,

0:01:32.200 --> 0:01:35.720
<v Speaker 2>more questions are now being asked about AI's running costs.

0:01:36.280 --> 0:01:40.680
<v Speaker 4>We are putting mostly chips silicon into these data centers

0:01:41.000 --> 0:01:44.720
<v Speaker 4>that have a lifespan of perhaps four years, So those

0:01:44.760 --> 0:01:46.920
<v Speaker 4>chips they appreciate very quickly.

0:01:47.040 --> 0:01:49.919
<v Speaker 3>Even in video, there's a new chip every eight months

0:01:49.920 --> 0:01:52.000
<v Speaker 3>and it's ten times as powerful as the earlier ones.

0:01:52.160 --> 0:01:54.360
<v Speaker 4>The thing with the valley this here is that almost

0:01:54.360 --> 0:01:57.320
<v Speaker 4>every investor knows it's all going to turn into pumpkins

0:01:57.360 --> 0:02:00.280
<v Speaker 4>and mice at midnight. Only, as Buffett would say, no

0:02:00.320 --> 0:02:01.480
<v Speaker 4>one in the room as a clock.

0:02:02.040 --> 0:02:06.280
<v Speaker 2>Even with bumper results and bullish revenue forecasts, here's why

0:02:06.400 --> 0:02:13.359
<v Speaker 2>AI casts still worry investors. Tom McKenzie, who hosts Bloomberg

0:02:13.400 --> 0:02:16.960
<v Speaker 2>Tech You're U bum Bloberg Television, joins me now for more. Tom.

0:02:17.320 --> 0:02:20.080
<v Speaker 2>The investor Michael Burry of Big Short fame is among

0:02:20.160 --> 0:02:23.280
<v Speaker 2>those who's worried about these future casts of AI and

0:02:23.440 --> 0:02:25.600
<v Speaker 2>data centers in particular.

0:02:25.680 --> 0:02:29.680
<v Speaker 4>What's the concern, Yeah, absolutely, Michael Burry putting on famously

0:02:29.720 --> 0:02:33.320
<v Speaker 4>short positions, so shortening the stocks of Nvidia and Pallanteer

0:02:33.400 --> 0:02:37.880
<v Speaker 4>before he wrapped up his fund. His concern does focus

0:02:38.360 --> 0:02:42.240
<v Speaker 4>on the depreciation of some of these assets by assets.

0:02:42.240 --> 0:02:47.120
<v Speaker 4>I'm talking about specifically these AI chips, very expensive AI accelerators.

0:02:47.240 --> 0:02:51.079
<v Speaker 4>Ninety percent of the market share is dominated by Nvidia,

0:02:51.280 --> 0:02:54.799
<v Speaker 4>so across the sale of these chips and video has

0:02:54.840 --> 0:02:58.720
<v Speaker 4>that significant market gain versus its rivals. And the concern

0:02:58.880 --> 0:03:01.480
<v Speaker 4>is that as you get newer versions of these chips,

0:03:01.480 --> 0:03:05.440
<v Speaker 4>the older ones essentially become less valuable. And Michael Barrie

0:03:05.960 --> 0:03:09.760
<v Speaker 4>making the argument that companies the hyperscalers, so the Microsofts

0:03:09.800 --> 0:03:12.600
<v Speaker 4>and alphabets and metas of the world, are not properly

0:03:12.639 --> 0:03:18.200
<v Speaker 4>accounting for how quickly these these assets depreciate. The other

0:03:18.240 --> 0:03:20.440
<v Speaker 4>part of the concern and kind of ties into this

0:03:20.840 --> 0:03:23.440
<v Speaker 4>that you hear voice from the skeptics around the AI

0:03:23.520 --> 0:03:26.560
<v Speaker 4>bubble is that there are comparisons, they say, with what

0:03:26.680 --> 0:03:29.480
<v Speaker 4>happened in the late nineteen nineties, nineteen ninety nine, early

0:03:29.520 --> 0:03:31.560
<v Speaker 4>two thousand, the dot com bubble, when it was the

0:03:31.639 --> 0:03:36.640
<v Speaker 4>telecom equipment makers that leading up to all of the

0:03:36.760 --> 0:03:40.480
<v Speaker 4>online expectations around how our digital economy was going to change,

0:03:40.640 --> 0:03:43.880
<v Speaker 4>spent huge amounts of money on building the infrastructure to

0:03:44.080 --> 0:03:49.040
<v Speaker 4>power the dot com era and ended up losing a

0:03:49.080 --> 0:03:52.120
<v Speaker 4>lot of money because the gains didn't come as quickly,

0:03:52.160 --> 0:03:55.360
<v Speaker 4>the technology didn't evolve as rapidly as they had expected.

0:03:55.440 --> 0:03:57.200
<v Speaker 4>Of course, on the back of that, you did get

0:03:57.600 --> 0:04:01.280
<v Speaker 4>some very significant players like Amazon who came through the

0:04:01.320 --> 0:04:04.080
<v Speaker 4>dot com bubble and of course now remain one of

0:04:04.080 --> 0:04:06.480
<v Speaker 4>the most valuable companies on the planet. But there was

0:04:06.560 --> 0:04:08.480
<v Speaker 4>a lot of capital, there was a lot of investment

0:04:08.480 --> 0:04:10.720
<v Speaker 4>that was burnt in that process. And so that is

0:04:10.760 --> 0:04:13.200
<v Speaker 4>another comparison that people are making. It's the depreciation around

0:04:13.200 --> 0:04:15.480
<v Speaker 4>the assets and the chips that they're worried about, but

0:04:15.560 --> 0:04:18.320
<v Speaker 4>also comparisons with what happened during the dot com era

0:04:18.400 --> 0:04:20.960
<v Speaker 4>and the pain that was felt by those telecom equipment

0:04:20.960 --> 0:04:24.480
<v Speaker 4>makers that sunk so much money into which they accumulated

0:04:24.560 --> 0:04:25.160
<v Speaker 4>huge losses.

0:04:25.760 --> 0:04:28.799
<v Speaker 2>So how are the big AI players thinking about these

0:04:28.920 --> 0:04:30.080
<v Speaker 2>casts at the moment?

0:04:30.720 --> 0:04:35.240
<v Speaker 4>So pushback to the depreciation argument would come from in

0:04:35.400 --> 0:04:38.120
<v Speaker 4>video and we've heard that recently from the CEO Jensen Huang,

0:04:38.839 --> 0:04:41.800
<v Speaker 4>and he's made the case that in fact, even their

0:04:41.839 --> 0:04:45.400
<v Speaker 4>older AI chips, one of their older versions is called Hopper,

0:04:46.040 --> 0:04:50.000
<v Speaker 4>has a lifespan of about six years and is very versatile,

0:04:50.040 --> 0:04:52.520
<v Speaker 4>so you can use it not just for the training

0:04:53.040 --> 0:04:55.919
<v Speaker 4>of these large language models, but for the post training

0:04:55.960 --> 0:04:58.680
<v Speaker 4>and for the inference. That's when they're actually being used

0:04:58.680 --> 0:05:01.800
<v Speaker 4>by us, by consumer and by enterprise, and so you

0:05:01.839 --> 0:05:05.120
<v Speaker 4>can move them around. They have different functions and therefore

0:05:05.160 --> 0:05:07.800
<v Speaker 4>they actually have a longer lifespan than some of the

0:05:07.800 --> 0:05:11.440
<v Speaker 4>skeptics are suggesting, and our own analysis suggest that those

0:05:11.440 --> 0:05:13.520
<v Speaker 4>Hopper chips, those older varieties of chips have a life

0:05:13.520 --> 0:05:15.960
<v Speaker 4>span of about six years and are fully utilized by

0:05:16.360 --> 0:05:19.240
<v Speaker 4>most of the companies that own those. So that does

0:05:19.279 --> 0:05:25.240
<v Speaker 4>address some of that concern. The question going forward to

0:05:25.279 --> 0:05:27.800
<v Speaker 4>what extent these companies are going to be able to

0:05:27.839 --> 0:05:33.120
<v Speaker 4>find products that match the investments that they are syncing

0:05:33.200 --> 0:05:36.320
<v Speaker 4>into the AI infrastructure story. A Bane Capital came out

0:05:36.320 --> 0:05:40.040
<v Speaker 4>with a report recently suggesting that by twenty thirty, the

0:05:40.160 --> 0:05:43.560
<v Speaker 4>hyper scalers and other AI giants would have to be

0:05:43.600 --> 0:05:47.240
<v Speaker 4>turning around revenues of about two trillion dollars and that

0:05:47.400 --> 0:05:50.080
<v Speaker 4>right now there's a huge gap, hundreds of billions of

0:05:50.080 --> 0:05:53.240
<v Speaker 4>dollars in terms of the gap between the investments into

0:05:53.279 --> 0:05:57.200
<v Speaker 4>the AI infrastructure and the actual revenues that are coming

0:05:57.240 --> 0:06:01.640
<v Speaker 4>about as customers and as enterprises and companies use the

0:06:01.800 --> 0:06:03.720
<v Speaker 4>end product. So the go to market, the product fit

0:06:03.839 --> 0:06:05.599
<v Speaker 4>is going to be really, really important. And what the

0:06:05.600 --> 0:06:08.839
<v Speaker 4>big AI players say whether or that is the hyperscalers again,

0:06:08.880 --> 0:06:11.479
<v Speaker 4>the likes of Meta and in Alphabet and Amazon say,

0:06:11.800 --> 0:06:13.840
<v Speaker 4>all the likes of open A and Anthropic because we're

0:06:13.839 --> 0:06:15.960
<v Speaker 4>going to be in this world of agentic AI. We're

0:06:15.960 --> 0:06:19.320
<v Speaker 4>going to have AI agents booking our holidays, checking up

0:06:19.360 --> 0:06:23.160
<v Speaker 4>on our healthcare, finding good schools and universities for our students.

0:06:23.320 --> 0:06:25.000
<v Speaker 4>All those kind of things are going to come together.

0:06:25.400 --> 0:06:28.520
<v Speaker 4>Enterprises are going to be embedding AI much more than

0:06:28.520 --> 0:06:31.240
<v Speaker 4>they already are. We're only in the first opening stages

0:06:31.279 --> 0:06:33.000
<v Speaker 4>of that would be the argument. And then there's the

0:06:33.120 --> 0:06:36.200
<v Speaker 4>sovereign AI story where different countries, and we're seeing that

0:06:36.240 --> 0:06:38.000
<v Speaker 4>in the Middle East but also in Europe as well

0:06:38.120 --> 0:06:40.520
<v Speaker 4>and Japan are investing heavily to ensure that they have

0:06:40.600 --> 0:06:43.920
<v Speaker 4>their own AI infrastructure AI clouds. That will very early

0:06:44.000 --> 0:06:46.239
<v Speaker 4>in that story as well. Those are all the cases

0:06:46.240 --> 0:06:49.760
<v Speaker 4>that the big AI players would underscore in terms of

0:06:49.920 --> 0:06:53.200
<v Speaker 4>why this is going to be driving momentum going forward

0:06:53.279 --> 0:06:56.359
<v Speaker 4>at least through twenty twenty six. Our own team at

0:06:56.360 --> 0:06:58.800
<v Speaker 4>Bloomberg Intelligence say the end of twenty twenty six is

0:06:58.800 --> 0:07:00.440
<v Speaker 4>going to be a question mark to whether or not

0:07:00.680 --> 0:07:04.400
<v Speaker 4>investors continue to have patients. Will they continue to invest

0:07:04.440 --> 0:07:08.160
<v Speaker 4>in the hyperscalers if they're not saying real material terms,

0:07:08.160 --> 0:07:11.080
<v Speaker 4>if that product fit and that custom use isn't there

0:07:11.160 --> 0:07:13.560
<v Speaker 4>in a really really significant way. So I think the

0:07:13.640 --> 0:07:17.040
<v Speaker 4>patients of investors and to what extent they can continue

0:07:17.520 --> 0:07:20.160
<v Speaker 4>to lean into the Hyperscalers if they spend these huge

0:07:20.160 --> 0:07:22.160
<v Speaker 4>amounts is going to be a key question mark, and

0:07:22.160 --> 0:07:23.880
<v Speaker 4>our own team think that that's really going to come

0:07:23.920 --> 0:07:25.880
<v Speaker 4>to the fore at the end of twenty twenty six.

0:07:25.880 --> 0:07:28.440
<v Speaker 4>They'll need to answer that question. They've they've spent. Hyper

0:07:28.440 --> 0:07:31.520
<v Speaker 4>Scalers have spent about three hundred billion dollars on air

0:07:31.640 --> 0:07:34.360
<v Speaker 4>infrastructure this year, and the projection is that they could

0:07:34.400 --> 0:07:37.000
<v Speaker 4>be according to Vidia in the video, sees the Hyperscalar

0:07:37.040 --> 0:07:40.040
<v Speaker 4>spending upwards of about six hundred billion dollars next year.

0:07:40.480 --> 0:07:41.920
<v Speaker 2>One of the things that occurs to me in this

0:07:42.000 --> 0:07:44.480
<v Speaker 2>as well. As we're talking about some of the world's

0:07:44.640 --> 0:07:48.920
<v Speaker 2>most valuable companies. They have massive cash piles in a

0:07:48.960 --> 0:07:52.360
<v Speaker 2>lot of cases, Why is their concern at all about

0:07:52.360 --> 0:07:54.760
<v Speaker 2>how they're going to pay for this given their revenue

0:07:54.760 --> 0:07:56.240
<v Speaker 2>streams and how much money they have.

0:07:56.640 --> 0:07:59.440
<v Speaker 4>You're absolutely right. So when we talk about the hyperscalers,

0:08:00.000 --> 0:08:04.560
<v Speaker 4>these are companies with massive balance sheets and huge cash reserves.

0:08:04.560 --> 0:08:08.120
<v Speaker 4>These are incredibly profitable with businesses that come through with

0:08:08.320 --> 0:08:13.040
<v Speaker 4>very strong earnings. These are not nonprofitable major punts and

0:08:13.160 --> 0:08:15.520
<v Speaker 4>risky parts of the market. These are not companies that

0:08:15.560 --> 0:08:18.560
<v Speaker 4>no one's heard of. They're making real product, they're selling

0:08:18.560 --> 0:08:22.800
<v Speaker 4>it to customers, and they've been doing that for decades. Microsoft, Alphabet, Meta,

0:08:22.840 --> 0:08:25.440
<v Speaker 4>and Amazon. They have that balance sheet strength, they have

0:08:25.560 --> 0:08:30.280
<v Speaker 4>that cash on hand. The concern then is around other

0:08:30.320 --> 0:08:32.280
<v Speaker 4>parts of this ecosystem. So if you can think about

0:08:32.320 --> 0:08:35.360
<v Speaker 4>it in different baskets, you have those big ticket blue

0:08:35.440 --> 0:08:38.840
<v Speaker 4>chip names in one basket, and then you have maybe

0:08:38.880 --> 0:08:41.640
<v Speaker 4>neo clouds in the other basket. These are the core

0:08:41.679 --> 0:08:45.920
<v Speaker 4>weaves or the en clouds companies that lease out data

0:08:45.960 --> 0:08:48.240
<v Speaker 4>centers to some of these hyper scalers, and some of

0:08:48.240 --> 0:08:51.199
<v Speaker 4>the large language models who have business models that are

0:08:51.280 --> 0:08:55.079
<v Speaker 4>less proven than the hyperscalers. Then another bucket would be

0:08:55.920 --> 0:08:58.280
<v Speaker 4>maybe some of the key large language models themselves, the

0:08:58.320 --> 0:09:01.040
<v Speaker 4>open eyes and the anthropics that are money on an

0:09:01.040 --> 0:09:03.720
<v Speaker 4>annual basis. Even as they're seeing a lot of growth

0:09:04.040 --> 0:09:08.600
<v Speaker 4>and revenues increase year on year, they're still not profitable.

0:09:08.840 --> 0:09:10.920
<v Speaker 4>So you can break it down into different categories in

0:09:11.000 --> 0:09:14.160
<v Speaker 4>terms of the level of risk. But even amongst the

0:09:14.160 --> 0:09:16.880
<v Speaker 4>big publicly listed companies with those strong balance sheets, you

0:09:16.960 --> 0:09:19.840
<v Speaker 4>have seen examples of then tapping the public markets and

0:09:19.960 --> 0:09:22.160
<v Speaker 4>raising debt on the public markets, and so far that's

0:09:22.200 --> 0:09:25.320
<v Speaker 4>been well received by the markets. But how long is

0:09:25.360 --> 0:09:27.440
<v Speaker 4>that going to continue? And to what extent is the

0:09:27.559 --> 0:09:31.120
<v Speaker 4>leverage that now these companies are starting to tap into

0:09:31.640 --> 0:09:34.600
<v Speaker 4>going to be acceptable to investors. And again I think

0:09:34.600 --> 0:09:36.880
<v Speaker 4>you have to put a different framework over the different

0:09:36.880 --> 0:09:39.960
<v Speaker 4>companies in terms of how you answer that question. Then

0:09:39.960 --> 0:09:42.920
<v Speaker 4>there's the circularity of the financing. So open Ai, for example,

0:09:42.960 --> 0:09:46.200
<v Speaker 4>doing deals with Nvidia and Nvidia investing in open Ai,

0:09:46.360 --> 0:09:49.280
<v Speaker 4>and in response to that, open Ai committing to buying

0:09:49.320 --> 0:09:53.000
<v Speaker 4>a certain number of chips from Nvidia. Those circular financing deals,

0:09:53.040 --> 0:09:55.560
<v Speaker 4>as they've been described by some have also caused some

0:09:55.640 --> 0:09:59.840
<v Speaker 4>concern as all of these companies becoming increasingly enmeshed and

0:10:00.040 --> 0:10:03.400
<v Speaker 4>intertwined in terms of their deals and their investments on

0:10:03.440 --> 0:10:05.280
<v Speaker 4>what is a bet on the future and how the

0:10:05.320 --> 0:10:06.400
<v Speaker 4>future evolves, and.

0:10:06.480 --> 0:10:08.880
<v Speaker 2>An expensive one at that. Tom McKenzie, thank you very

0:10:08.960 --> 0:10:11.920
<v Speaker 2>much for joining us, host of Bloomberg Tech Europe on

0:10:11.960 --> 0:10:15.800
<v Speaker 2>Bloomberg Television. For more explanations like this from our team

0:10:15.840 --> 0:10:18.679
<v Speaker 2>of three thousand journalists and analysts around the world, go

0:10:18.679 --> 0:10:22.240
<v Speaker 2>to Bloomberg dot com slash explainers. I'm Stephen Carroll. This

0:10:22.320 --> 0:10:24.480
<v Speaker 2>is Here's why. I'll be back next week with more.

0:10:24.720 --> 0:10:25.480
<v Speaker 2>Thanks for listening.