WEBVTT - Here's Why AI Is Creating The Biggest Monopolies in History

0:00:02.560 --> 0:00:10.280
<v Speaker 1>Bloomberg Audio Studios, podcasts, radio news. I'm Stephen Carroll and

0:00:10.320 --> 0:00:12.760
<v Speaker 1>this is Here's Why, where we take one new story

0:00:12.800 --> 0:00:14.840
<v Speaker 1>and explain it in just a few minutes with our

0:00:14.880 --> 0:00:22.520
<v Speaker 1>experts here at Bloomberg. Chat shipt fast as app in

0:00:22.560 --> 0:00:24.880
<v Speaker 1>the world to one hundred million users and today well

0:00:24.920 --> 0:00:28.080
<v Speaker 1>over three hundred million users. Were the very beginning of

0:00:28.080 --> 0:00:31.160
<v Speaker 1>a twenty year cycle where productivity will really increase.

0:00:31.240 --> 0:00:32.280
<v Speaker 2>Using these technologies.

0:00:32.320 --> 0:00:36.280
<v Speaker 1>It'll change everything, It'll be integrated into every business. Chat

0:00:36.320 --> 0:00:39.839
<v Speaker 1>GPT has been a game changer in artificial intelligence. The

0:00:39.880 --> 0:00:43.720
<v Speaker 1>sudden rush into large language models has propelled a handful

0:00:43.760 --> 0:00:48.040
<v Speaker 1>of companies into markets superstars. So it is in video Day,

0:00:48.280 --> 0:00:51.320
<v Speaker 1>happy in video data all who celebrate t SMC revenue

0:00:51.320 --> 0:00:53.800
<v Speaker 1>climbing thirty nine percent in the first two months of

0:00:53.800 --> 0:00:57.920
<v Speaker 1>the year. Korean shipmaker sk Highnecks delivering record quarterly profit.

0:00:58.920 --> 0:01:03.000
<v Speaker 1>Previous revolutions in technology, like with personal computers or search engines,

0:01:03.040 --> 0:01:07.360
<v Speaker 1>have created industry giants. This time around, the scale is

0:01:07.400 --> 0:01:11.840
<v Speaker 1>even bigger. Here's why AI is creating the biggest monopolies

0:01:12.160 --> 0:01:18.640
<v Speaker 1>in history. Our technology reporter Mark Bergen joins me now

0:01:18.840 --> 0:01:22.440
<v Speaker 1>for more. Mark, who are first of all these megastars

0:01:22.480 --> 0:01:23.840
<v Speaker 1>of AI and what do they do.

0:01:24.319 --> 0:01:26.760
<v Speaker 2>The household name now is in Nvidia, which has gone

0:01:26.840 --> 0:01:29.240
<v Speaker 2>from this sort of in the past few years a

0:01:29.360 --> 0:01:32.880
<v Speaker 2>niche chips company mostly for the gaming industry into one

0:01:32.880 --> 0:01:35.800
<v Speaker 2>of the world's largest, multi trillion dollar companies sort of

0:01:35.840 --> 0:01:38.040
<v Speaker 2>associated with this AI boom, and they are of course

0:01:38.120 --> 0:01:42.560
<v Speaker 2>the primary company designing what are called AI accelerators, basically

0:01:42.640 --> 0:01:44.800
<v Speaker 2>the chips that go into these data centers that make

0:01:45.240 --> 0:01:48.960
<v Speaker 2>things like CHATBT work. What we found in this really

0:01:49.000 --> 0:01:52.320
<v Speaker 2>interesting is that behind those companies are actually three of

0:01:52.360 --> 0:01:56.880
<v Speaker 2>their suppliers and partners that have a somewhat justice fortified

0:01:56.920 --> 0:02:00.360
<v Speaker 2>market position. So in Vidia has about ninety two percent

0:02:00.440 --> 0:02:03.320
<v Speaker 2>of the market for a accelerator, which is just phenomenal.

0:02:03.800 --> 0:02:06.680
<v Speaker 2>There's a company called sk Heinex out of Korea that

0:02:07.000 --> 0:02:09.200
<v Speaker 2>produces memory chips and some of the highest end memory

0:02:09.280 --> 0:02:11.799
<v Speaker 2>chips that go into the sort of GPUs that in

0:02:11.919 --> 0:02:15.400
<v Speaker 2>video makes. They have and estimates maybe eighty percent of

0:02:15.440 --> 0:02:19.040
<v Speaker 2>that market. There is TSMC, which produces the chips for

0:02:19.320 --> 0:02:21.480
<v Speaker 2>Nvidia and a lot of the rest of the world.

0:02:21.600 --> 0:02:24.440
<v Speaker 2>They have for the AI chips maybe ninety five percent.

0:02:24.440 --> 0:02:27.360
<v Speaker 2>Then there's a Dutch company called ASML that makes these

0:02:27.639 --> 0:02:32.080
<v Speaker 2>very expensive lithography machines that are necessary to build these chips,

0:02:32.160 --> 0:02:35.600
<v Speaker 2>and they have for those specific types of lithography machines

0:02:35.720 --> 0:02:38.480
<v Speaker 2>that are making these chips one hundred percent market show

0:02:38.520 --> 0:02:40.160
<v Speaker 2>no other company in the world.

0:02:40.400 --> 0:02:42.680
<v Speaker 1>And is that how we can qualify them as monopolies?

0:02:42.680 --> 0:02:44.280
<v Speaker 1>I mean, that's a big word in the business roils

0:02:44.280 --> 0:02:45.359
<v Speaker 1>when we talk about enough it.

0:02:45.320 --> 0:02:46.639
<v Speaker 2>Is, Yeah, And I think it should be very clear,

0:02:46.680 --> 0:02:48.120
<v Speaker 2>and we say this in the story that there's a

0:02:48.120 --> 0:02:50.880
<v Speaker 2>difference here where between the sort of definition of monopoly

0:02:50.960 --> 0:02:55.000
<v Speaker 2>around the market share and the legal definition of a monopoly,

0:02:55.360 --> 0:02:58.840
<v Speaker 2>which may be associate with anti competitive behavior. And so

0:02:58.919 --> 0:03:03.800
<v Speaker 2>far there've been so antitrust cases opened against Nvidia, but

0:03:03.919 --> 0:03:07.760
<v Speaker 2>no findings about these poor companies that they've demonstrated anti

0:03:07.800 --> 0:03:11.400
<v Speaker 2>competitive behavior in getting this positions. So I think maybe

0:03:11.480 --> 0:03:13.959
<v Speaker 2>they're not necessarily what we call an abusive monopoly.

0:03:14.240 --> 0:03:16.079
<v Speaker 1>They're just good at what they do. I suppose it'd

0:03:16.080 --> 0:03:17.520
<v Speaker 1>be another way to look at it as well.

0:03:17.639 --> 0:03:21.080
<v Speaker 2>Yeah, I think that's one of several factors behind why

0:03:21.280 --> 0:03:22.480
<v Speaker 2>we have these positions.

0:03:22.760 --> 0:03:25.360
<v Speaker 1>They have huge market shares. You say in these areas,

0:03:25.400 --> 0:03:27.920
<v Speaker 1>are there any competitors that we should be watching that

0:03:28.000 --> 0:03:29.519
<v Speaker 1>could potentially disrupt that.

0:03:29.960 --> 0:03:32.760
<v Speaker 2>One of the things that's happening in AI and this

0:03:32.840 --> 0:03:35.720
<v Speaker 2>moment is everything is moving much faster than sort of

0:03:35.760 --> 0:03:41.040
<v Speaker 2>the historical timelines for technology sectors or something like the

0:03:41.360 --> 0:03:45.680
<v Speaker 2>personal computing and mobile phones and browsers. In Nvidia is

0:03:45.840 --> 0:03:48.280
<v Speaker 2>certainly the most prominent and had gained the most from

0:03:48.280 --> 0:03:51.160
<v Speaker 2>the AI boom. In some ways, is the most vulnerable

0:03:51.200 --> 0:03:54.600
<v Speaker 2>here in part because they have this really fascinating dynamic

0:03:54.640 --> 0:03:58.720
<v Speaker 2>where their biggest customers are now effectively coming to compete

0:03:58.720 --> 0:04:02.160
<v Speaker 2>with them. Almost half their revenue comes from the big

0:04:02.280 --> 0:04:06.320
<v Speaker 2>giants of Silicon Valley. They have Microsoft, Amazon, Google, and Meta.

0:04:06.360 --> 0:04:09.840
<v Speaker 2>All of those companies, how have been exploring to various

0:04:09.840 --> 0:04:14.160
<v Speaker 2>degrees alternatives to in video, whether it's through partnerships, whether

0:04:14.160 --> 0:04:17.200
<v Speaker 2>it's to building their own customs chips and Google's cases,

0:04:17.360 --> 0:04:20.279
<v Speaker 2>Google's had a custom chip for a long time, certainly

0:04:20.320 --> 0:04:23.200
<v Speaker 2>developing that. Amazon, as we reported, has done some very

0:04:23.200 --> 0:04:26.159
<v Speaker 2>aggressive in house design, right these companies don't want to

0:04:26.160 --> 0:04:29.520
<v Speaker 2>be reliant on a partner like that for something they

0:04:29.600 --> 0:04:32.039
<v Speaker 2>see is so mission critical opening eye, it's been some

0:04:32.120 --> 0:04:35.159
<v Speaker 2>reporting has also been designing its own custom chip. To

0:04:35.160 --> 0:04:36.919
<v Speaker 2>be clear, like we haven't seen this reflected at all,

0:04:36.960 --> 0:04:40.000
<v Speaker 2>and in video sales or its projections. There are a

0:04:40.120 --> 0:04:43.400
<v Speaker 2>number of startups that have come around that are also

0:04:43.440 --> 0:04:45.960
<v Speaker 2>trying to replicate what in video's done or try to

0:04:45.960 --> 0:04:49.599
<v Speaker 2>produce like a next generation of chip. Intel is probably

0:04:49.640 --> 0:04:51.680
<v Speaker 2>the most interesting example right now. They just got a

0:04:51.720 --> 0:04:55.920
<v Speaker 2>new CEO who has signaled that there want to get

0:04:55.920 --> 0:04:58.000
<v Speaker 2>into this, and we call the foundry game basically being

0:04:58.000 --> 0:05:01.600
<v Speaker 2>a factory like TSMC, Samsung has probably been the closest

0:05:01.600 --> 0:05:06.080
<v Speaker 2>competitor sk Heinez has a lot of more competition there

0:05:06.080 --> 0:05:09.400
<v Speaker 2>In the memory chips of Samsung, you have Micron, and

0:05:09.640 --> 0:05:13.480
<v Speaker 2>at this point ASML is probably of those four the

0:05:13.480 --> 0:05:16.960
<v Speaker 2>most entrenched. Unless some sort of black Swan event happens,

0:05:16.960 --> 0:05:18.960
<v Speaker 2>it's very unlikely that we're going to see another company

0:05:19.480 --> 0:05:23.839
<v Speaker 2>develop that machinery and those supply chain partnerships.

0:05:23.200 --> 0:05:29.400
<v Speaker 1>To compete our competition. Regulators worried about these companies and

0:05:29.520 --> 0:05:32.599
<v Speaker 1>the I suppose dominance they have in these spaces.

0:05:33.000 --> 0:05:35.360
<v Speaker 2>I think short answer is yes and no. We've seen

0:05:35.600 --> 0:05:38.400
<v Speaker 2>in Vidia's case, the Department of Justice last year had

0:05:38.440 --> 0:05:41.400
<v Speaker 2>opened a potential case and as far as our reporting shows,

0:05:41.440 --> 0:05:44.000
<v Speaker 2>the concerns here is that they are kind of bundling

0:05:44.040 --> 0:05:46.520
<v Speaker 2>their software with some of their hardware products, where it

0:05:46.600 --> 0:05:51.280
<v Speaker 2>basically makes it really difficult for say, someone to use

0:05:51.960 --> 0:05:55.080
<v Speaker 2>a different type of chipset, because in video software is

0:05:55.120 --> 0:05:57.920
<v Speaker 2>so prominent and so widely used across AI. If you're

0:05:57.920 --> 0:06:00.960
<v Speaker 2>building a chatpot, if you're building a large language. One

0:06:00.960 --> 0:06:03.039
<v Speaker 2>of the reasons that in Vidia has grown so well

0:06:03.120 --> 0:06:05.520
<v Speaker 2>is because they were out in ahead of having this

0:06:05.600 --> 0:06:10.159
<v Speaker 2>support software that worked really well with these chips The company,

0:06:10.160 --> 0:06:12.239
<v Speaker 2>to be clear, has pushed back on that and said,

0:06:12.480 --> 0:06:14.600
<v Speaker 2>you know, it doesn't bundle its software. It has a

0:06:14.600 --> 0:06:16.719
<v Speaker 2>lot of faces, a lot of competition. There was some

0:06:16.760 --> 0:06:19.800
<v Speaker 2>scrutiny around an acquisition they made this Israeli company that

0:06:19.839 --> 0:06:22.919
<v Speaker 2>makes off were called run Ai, and I've talked to

0:06:22.960 --> 0:06:25.200
<v Speaker 2>some people who have some critics who point out that

0:06:25.240 --> 0:06:27.000
<v Speaker 2>this is sort of a traditional move that they've seen

0:06:27.040 --> 0:06:29.919
<v Speaker 2>of other companies building these kind of walled gardens. In

0:06:29.920 --> 0:06:32.760
<v Speaker 2>the US. Certainly, the Trump administration seems to be the

0:06:32.920 --> 0:06:37.479
<v Speaker 2>signaled less interest instead of going after monopolies than the

0:06:37.480 --> 0:06:41.920
<v Speaker 2>Biden administration. We haven't seen in that acquisition of Runai

0:06:42.040 --> 0:06:44.960
<v Speaker 2>by Nvidia cleared in both the US and the EU,

0:06:45.040 --> 0:06:49.159
<v Speaker 2>which has traditionally been much more aggressive about monopolies. So

0:06:49.720 --> 0:06:54.120
<v Speaker 2>as far as there's certainly not, say, the antitrust concerns

0:06:54.480 --> 0:06:58.440
<v Speaker 2>about Google or microsoftist paced in the past, or even

0:06:58.560 --> 0:07:01.279
<v Speaker 2>going back further in the hardware the companies like IBM

0:07:01.360 --> 0:07:02.520
<v Speaker 2>that faced mark.

0:07:02.560 --> 0:07:04.720
<v Speaker 1>You've been reporting on the tech industry for many years.

0:07:04.800 --> 0:07:08.080
<v Speaker 1>I wonder how you compare this moment and the emergence

0:07:08.080 --> 0:07:13.240
<v Speaker 1>of these giants in art facial intelligence chips to previous

0:07:13.560 --> 0:07:16.920
<v Speaker 1>big bang moments in tech. Is this an iPhone moment?

0:07:17.080 --> 0:07:19.640
<v Speaker 1>Is this a Google moment in search engines?

0:07:19.920 --> 0:07:21.840
<v Speaker 2>I say it's more of a Wintel moment, which is

0:07:21.920 --> 0:07:24.640
<v Speaker 2>less familiar maybe, But the Wintel is that term for

0:07:25.080 --> 0:07:29.480
<v Speaker 2>the combination of Microsoft and Intel sort of duopoly around

0:07:29.560 --> 0:07:33.800
<v Speaker 2>personal computers that form in the eighties and really solified that.

0:07:34.000 --> 0:07:37.440
<v Speaker 2>You know, every single desktop computer that shipped had Microsoft

0:07:37.760 --> 0:07:41.000
<v Speaker 2>software with it and had an Intel chip inside. And that,

0:07:41.320 --> 0:07:45.200
<v Speaker 2>to be clear, is that that's still a fairly dominant

0:07:45.400 --> 0:07:48.960
<v Speaker 2>market position at this point. A lot of the market's attention,

0:07:49.600 --> 0:07:52.080
<v Speaker 2>a lot of the legal and political tension has shifted

0:07:52.200 --> 0:07:55.080
<v Speaker 2>sage to mobile and maybe now shifting to AI. But

0:07:55.120 --> 0:07:56.840
<v Speaker 2>we're still you know, you and I are looking at

0:07:57.000 --> 0:08:00.960
<v Speaker 2>PCs right now. Is they're still prominent parts of life,

0:08:01.000 --> 0:08:04.800
<v Speaker 2>and most of them still ship with Microsoft and Intel.

0:08:04.880 --> 0:08:07.680
<v Speaker 2>I think just because you know, the ship industry has

0:08:07.720 --> 0:08:12.760
<v Speaker 2>traditionally had companies that have tended to dominate particular components

0:08:12.840 --> 0:08:16.560
<v Speaker 2>or niches. That's not necessarily new. But what is new

0:08:16.680 --> 0:08:20.480
<v Speaker 2>is the attention that this is receiving, I say, globally

0:08:20.680 --> 0:08:23.560
<v Speaker 2>politically and certainly in the market. Right that the companies

0:08:23.600 --> 0:08:28.000
<v Speaker 2>look and Vidia and its partners, the amount of money

0:08:28.040 --> 0:08:30.480
<v Speaker 2>tied up with them in the market and the amount

0:08:30.480 --> 0:08:33.040
<v Speaker 2>of money tied up in this thesis that they're just

0:08:33.120 --> 0:08:37.040
<v Speaker 2>going to be increased increased in spending around AI and

0:08:37.320 --> 0:08:39.719
<v Speaker 2>AI hardware. That is unprecedented.

0:08:40.040 --> 0:08:42.880
<v Speaker 1>Okay, Mark Bergen, our technology reporter, Thank you very much.

0:08:43.240 --> 0:08:45.720
<v Speaker 1>For more explanations like this from our team of three

0:08:45.720 --> 0:08:48.520
<v Speaker 1>thousand journalists and analysts around the world, go to Bloomberg

0:08:48.559 --> 0:08:53.679
<v Speaker 1>dot com slash explainers. I'm Stephen Carroll. This is here's why.

0:08:53.760 --> 0:08:56.040
<v Speaker 1>I'll be back next week with more. Thanks for listening.