WEBVTT - Monologue: OpenAI and NVIDIA's Deal Is BS

0:00:02.800 --> 0:00:07.400
<v Speaker 1>Z Media. Hell, welcome to this week's Better Offline Monologue.

0:00:07.440 --> 0:00:10.160
<v Speaker 1>I'm your host ed Zeitrono. As I wrote while drafting this,

0:00:10.240 --> 0:00:12.320
<v Speaker 1>I'm the house house, a line I'm leaving in as

0:00:12.360 --> 0:00:26.920
<v Speaker 1>I'm being accountable to you, the customer. Anyway, you've heard

0:00:26.920 --> 0:00:29.120
<v Speaker 1>about the in Video Open Ai deal. By now that

0:00:29.280 --> 0:00:31.920
<v Speaker 1>InVideo will and I put this in air quotes invest

0:00:31.960 --> 0:00:34.280
<v Speaker 1>one hundred billion dollars in open Ai. And I want

0:00:34.280 --> 0:00:36.960
<v Speaker 1>to start out by being explicit that Invidia is not

0:00:37.120 --> 0:00:40.440
<v Speaker 1>wired open Ai as single goddamn dollar. All right, So

0:00:40.600 --> 0:00:42.960
<v Speaker 1>the deal is that in VideA is investing ten billion

0:00:43.000 --> 0:00:45.960
<v Speaker 1>dollars up front, according to CNBC, despite the fact that

0:00:46.200 --> 0:00:49.280
<v Speaker 1>the date before it said that they would be investing

0:00:49.320 --> 0:00:52.520
<v Speaker 1>ten billion dollars with the first gig. But anyway, they're

0:00:52.560 --> 0:00:55.160
<v Speaker 1>investing ten billion dollars in a month, allegedly, and I

0:00:55.240 --> 0:00:58.120
<v Speaker 1>quote here once the deal is finalized, which means in

0:00:58.200 --> 0:01:01.480
<v Speaker 1>simpler terms, that nothing has actually been and Video will

0:01:01.480 --> 0:01:04.640
<v Speaker 1>then invest ten billion dollars in tranches for each Gigawa

0:01:04.720 --> 0:01:08.039
<v Speaker 1>have capacity to that open Ai built. That's right, That's right. Everyone.

0:01:08.319 --> 0:01:10.679
<v Speaker 1>Most of this funding is gated behind open Ai, a

0:01:10.720 --> 0:01:13.400
<v Speaker 1>company that burns billions of dollars building huge amounts of

0:01:13.480 --> 0:01:17.480
<v Speaker 1>data centers, something they have never done, and that also

0:01:17.600 --> 0:01:20.839
<v Speaker 1>costs billions of dollars. And just as an aside, open

0:01:20.840 --> 0:01:23.679
<v Speaker 1>ai really has built nothing themselves. Everyone else is building

0:01:23.680 --> 0:01:27.000
<v Speaker 1>it for them. These fucking silicon value libertarians. They love

0:01:27.080 --> 0:01:31.039
<v Speaker 1>themselves and welfare, don't they? Anyway? How does open ai

0:01:31.160 --> 0:01:34.640
<v Speaker 1>afford to build these data centers? I don't know. Nobody

0:01:34.640 --> 0:01:37.720
<v Speaker 1>else seems to know. Where are they being built? Who knows?

0:01:37.760 --> 0:01:40.680
<v Speaker 1>It's unclear. Has the land been purchased? Also unclear. There's

0:01:40.720 --> 0:01:43.960
<v Speaker 1>some sort of crap announcement about oh stargate, We've got

0:01:43.959 --> 0:01:47.480
<v Speaker 1>a list of different fucking new places, and I had

0:01:47.480 --> 0:01:50.160
<v Speaker 1>someone on the reddit suggests I was wrong. I just

0:01:50.200 --> 0:01:52.320
<v Speaker 1>want to be clear. Of the list of the five

0:01:52.440 --> 0:01:56.000
<v Speaker 1>new stargate locations, one of them is in Shackleford, Texas,

0:01:56.080 --> 0:01:59.080
<v Speaker 1>which was already chosen to be one of Oracle's data centers.

0:01:59.200 --> 0:02:02.200
<v Speaker 1>This was already an announced. They're re announcing stuff as

0:02:02.200 --> 0:02:05.160
<v Speaker 1>a pr professional, When you're re announcing shit, it's because

0:02:05.160 --> 0:02:08.519
<v Speaker 1>you run out of stuff to announce this real. But look,

0:02:09.080 --> 0:02:13.079
<v Speaker 1>let's let's be explicit about one thing. Data centers take

0:02:13.120 --> 0:02:15.840
<v Speaker 1>a long long time to build, right, now, Crusoe is

0:02:15.840 --> 0:02:18.840
<v Speaker 1>building a data center in Abilene, Texas, made up of

0:02:18.880 --> 0:02:21.200
<v Speaker 1>eight buildings, which will amount to about one point two

0:02:21.240 --> 0:02:23.640
<v Speaker 1>gigawats of compute capacity. This is the one for Oracle

0:02:23.639 --> 0:02:26.200
<v Speaker 1>and open AI, and at the current rate of development,

0:02:26.200 --> 0:02:30.000
<v Speaker 1>it's taken about two and a half years per giga one. Look,

0:02:30.160 --> 0:02:32.760
<v Speaker 1>I realize it's easy to get drowned in big numbers

0:02:32.760 --> 0:02:35.280
<v Speaker 1>and ridiculous promises and media headlines where the writers don't

0:02:35.280 --> 0:02:37.440
<v Speaker 1>seem to bother to learn even the simplest fucking thinks.

0:02:37.600 --> 0:02:40.440
<v Speaker 1>But data centers are expensive and take forever, and this

0:02:40.639 --> 0:02:46.680
<v Speaker 1>entire plan hinges upon open AI building data centers. In fact, look,

0:02:46.880 --> 0:02:50.000
<v Speaker 1>I've seen lots of people suggest that the ten billion

0:02:50.040 --> 0:02:52.520
<v Speaker 1>dollars the open AI is getting from Nvideo will let

0:02:52.600 --> 0:02:56.320
<v Speaker 1>them build data centers. And again, can somebody please actually

0:02:56.360 --> 0:03:00.600
<v Speaker 1>learn the basics here? The Abilene data center venture requires Crusoe,

0:03:00.639 --> 0:03:04.040
<v Speaker 1>Blue Oul and Primary Digital Infrastructure to raise fifteen billion

0:03:04.080 --> 0:03:07.000
<v Speaker 1>dollars in debt just for the data center itself, the construction,

0:03:07.080 --> 0:03:09.400
<v Speaker 1>the people, the water, the power, and all that good stuff,

0:03:09.520 --> 0:03:12.040
<v Speaker 1>with Oracle footing the bill for the hundreds of thousands

0:03:12.040 --> 0:03:14.560
<v Speaker 1>of GPUs to go inside. At a price of I

0:03:14.560 --> 0:03:17.080
<v Speaker 1>think about twenty four billion dollars. It works out to

0:03:17.680 --> 0:03:20.320
<v Speaker 1>by my calculations, open ai will need one hundred and

0:03:20.360 --> 0:03:22.919
<v Speaker 1>twenty five billion dollars just to build the data centers

0:03:22.960 --> 0:03:25.600
<v Speaker 1>to get the funding from in Video, which will require

0:03:25.639 --> 0:03:28.320
<v Speaker 1>another two hundred billion dollars in GPUs. By the way,

0:03:28.639 --> 0:03:31.200
<v Speaker 1>the information also reports that in Video is going to

0:03:31.240 --> 0:03:35.320
<v Speaker 1>be potentially leasing GPUs to open ai, setting up a

0:03:35.400 --> 0:03:39.040
<v Speaker 1>special purpose entity to buy the GPUs from themselves that

0:03:39.040 --> 0:03:43.240
<v Speaker 1>they would then lease to open AI. I've really hesitated

0:03:43.320 --> 0:03:47.200
<v Speaker 1>to bring up any Enron comparisons. I've kind of flirted

0:03:47.200 --> 0:03:51.080
<v Speaker 1>with one with core We even then that's that's straight

0:03:51.240 --> 0:03:53.640
<v Speaker 1>Enron shit. Like, man, they haven't conferred they're going to

0:03:53.680 --> 0:03:55.880
<v Speaker 1>do it. It's an information report or put a link

0:03:55.960 --> 0:03:58.080
<v Speaker 1>in there. You'll complain that you have to pay for it.

0:03:58.200 --> 0:04:03.560
<v Speaker 1>I recommend the information. I just every day, every day,

0:04:03.680 --> 0:04:05.800
<v Speaker 1>these new pieces of news where everyone kind of like

0:04:05.920 --> 0:04:08.960
<v Speaker 1>trumpets them like Jesus just woke up from behind that

0:04:09.000 --> 0:04:13.040
<v Speaker 1>big fucking rock, when in fact, I think something far

0:04:13.120 --> 0:04:16.159
<v Speaker 1>worse is happening. But I really want to break something

0:04:16.200 --> 0:04:18.320
<v Speaker 1>down for you, and you'll likely hear this again in

0:04:18.360 --> 0:04:20.760
<v Speaker 1>a future episode. I want to get into the actual

0:04:20.800 --> 0:04:25.000
<v Speaker 1>economics here, Cruso's one point two gigawatts of compute for

0:04:25.040 --> 0:04:27.800
<v Speaker 1>open Ai is a fifteen billion dollar joint venture, which

0:04:27.880 --> 0:04:30.560
<v Speaker 1>means a gigawa of compute runs about twelve point five

0:04:30.640 --> 0:04:34.080
<v Speaker 1>billion dollars. Abilene's eight buildings are meant to hold about

0:04:34.120 --> 0:04:37.080
<v Speaker 1>fifty thousand in video GB two hundred GPUs in their

0:04:37.080 --> 0:04:40.480
<v Speaker 1>associated networking infrastructure. So let's say a giga what is

0:04:40.520 --> 0:04:45.120
<v Speaker 1>about three hundred and three thousand, three hundred and thirty

0:04:45.120 --> 0:04:48.000
<v Speaker 1>three Blackwell GPUs. Yeah, this math is a little bit

0:04:48.080 --> 0:04:50.560
<v Speaker 1>funky due to in Video promising to install theirny Reubin

0:04:50.680 --> 0:04:54.920
<v Speaker 1>GPUs in these theoretical data centers. But I really do

0:04:54.960 --> 0:04:57.560
<v Speaker 1>think that they'll require about just a little under two

0:04:57.640 --> 0:05:00.279
<v Speaker 1>hundred billion dollars worth of GPUs. And this is on

0:05:00.360 --> 0:05:02.800
<v Speaker 1>top of the four hundred billion dollars the open Ai

0:05:02.920 --> 0:05:05.560
<v Speaker 1>has promised to spend with Oracle over the next five years,

0:05:05.920 --> 0:05:09.040
<v Speaker 1>when open ai had already leaked that they'd be burning

0:05:09.080 --> 0:05:11.560
<v Speaker 1>one hundred and fifteen billion dollars over the next four years.

0:05:11.800 --> 0:05:15.400
<v Speaker 1>This math does not make any sense. It's complete nonsense.

0:05:15.839 --> 0:05:18.000
<v Speaker 1>For open ai to survive, they're going to need around

0:05:18.080 --> 0:05:21.920
<v Speaker 1>six hundred billion dollars in the next five years. It's ridiculous, astonishing,

0:05:21.960 --> 0:05:24.400
<v Speaker 1>and insulting to the general public that the media keeps

0:05:24.440 --> 0:05:27.120
<v Speaker 1>acting as if these deals are signed or promoted in

0:05:27.120 --> 0:05:29.600
<v Speaker 1>good faith. And I'm shocked that more people aren't doing

0:05:29.640 --> 0:05:32.960
<v Speaker 1>the very simple maths that And I've seen a disgraceful

0:05:32.960 --> 0:05:36.280
<v Speaker 1>amount of people talking about Open Aiy's revenue projections, which

0:05:36.320 --> 0:05:39.000
<v Speaker 1>involve them making one hundred billion dollars in twenty twenty

0:05:39.040 --> 0:05:41.080
<v Speaker 1>eight and one hundred and forty five billion dollars in

0:05:41.120 --> 0:05:44.279
<v Speaker 1>twenty twenty nine, and that's revenue, not profit. They're losing

0:05:44.440 --> 0:05:46.880
<v Speaker 1>tens of billions of dollars both those years, and they're

0:05:46.920 --> 0:05:49.880
<v Speaker 1>acting as if these are serious, meaningful statements rather than

0:05:50.000 --> 0:05:54.120
<v Speaker 1>egregious lies from a gaggle of Charlatan's. It's a fucking joke.

0:05:54.440 --> 0:05:57.200
<v Speaker 1>It's sickening. Oh, by the way, Open Eyes projected to

0:05:57.200 --> 0:06:00.520
<v Speaker 1>make thirteen billion this year. From Mike calculs, they made

0:06:00.520 --> 0:06:04.560
<v Speaker 1>about six point twenty six billion. Is it's wank And

0:06:04.600 --> 0:06:07.719
<v Speaker 1>that's the technical term they use it. The financial analysts

0:06:07.760 --> 0:06:09.840
<v Speaker 1>put it. When you pay JP Morgan to see their

0:06:09.839 --> 0:06:12.240
<v Speaker 1>private shit, it's or yeah, this is all wank. This

0:06:12.400 --> 0:06:15.960
<v Speaker 1>is wink. Okay, I'll keep going now for the next

0:06:15.960 --> 0:06:18.720
<v Speaker 1>three part episode I'm gonna put out next week. I've

0:06:18.760 --> 0:06:20.800
<v Speaker 1>done a lot of digging into how much actual money

0:06:20.880 --> 0:06:24.640
<v Speaker 1>exists to invest in general, and we're genuinely running up

0:06:24.640 --> 0:06:27.800
<v Speaker 1>against the limits of private and public capital. By my calculations,

0:06:27.839 --> 0:06:30.600
<v Speaker 1>there's only about four hundred and seventy seven billion dollars

0:06:30.640 --> 0:06:33.479
<v Speaker 1>of available capital across the top ten private equity funds,

0:06:33.800 --> 0:06:36.440
<v Speaker 1>and I'm deadly serious. I will be publishing the links

0:06:36.720 --> 0:06:38.680
<v Speaker 1>and maybe one hundred and sixty four billion dollars of

0:06:38.800 --> 0:06:41.440
<v Speaker 1>US venture capital left, and maybe let's call that another

0:06:41.480 --> 0:06:44.840
<v Speaker 1>one hundred and fifty BILLI in other funds sloshing around.

0:06:46.279 --> 0:06:49.039
<v Speaker 1>Are we really going to carve out half to two

0:06:49.080 --> 0:06:52.560
<v Speaker 1>thirds of available global capital for one fucking company that

0:06:52.600 --> 0:06:56.159
<v Speaker 1>burns a billion fucking dollars every two fucking seconds. I

0:06:56.240 --> 0:07:00.600
<v Speaker 1>know it's slower than that. Shut up. I'm just I'm

0:07:00.640 --> 0:07:02.880
<v Speaker 1>a little bit ornery. I'm not going to lie. You

0:07:02.920 --> 0:07:04.680
<v Speaker 1>can hear it in my voice. And I am even

0:07:04.760 --> 0:07:08.080
<v Speaker 1>more ornery than ever, more ornery than ornery, as Rob

0:07:08.200 --> 0:07:11.760
<v Speaker 1>Zombie once sung. I'm just pissed off because when you

0:07:11.800 --> 0:07:14.440
<v Speaker 1>look at these numbers, they do not make sense. There's

0:07:14.560 --> 0:07:18.640
<v Speaker 1>no parallel in history for any company or any industry

0:07:18.680 --> 0:07:21.520
<v Speaker 1>that's ever spent this much money. There's never been one

0:07:21.640 --> 0:07:26.040
<v Speaker 1>company that has been the focus like this. And honestly,

0:07:26.080 --> 0:07:29.040
<v Speaker 1>I've never seen a company less deserving of any investment

0:07:29.080 --> 0:07:31.880
<v Speaker 1>than open AI. They've burned billions of dollars. Their product

0:07:31.920 --> 0:07:36.600
<v Speaker 1>is questionably useful, it's incredibly commoditized. They have yet to

0:07:37.080 --> 0:07:39.720
<v Speaker 1>create anything really exciting or new since And if I'm

0:07:39.760 --> 0:07:43.520
<v Speaker 1>being kind reasoning, but let's be honest. GPT four, Oh

0:07:43.760 --> 0:07:48.080
<v Speaker 1>they are washed. And yet everybody's sitting around them clapping

0:07:48.160 --> 0:07:51.360
<v Speaker 1>like they're a gifted child and insult to children and

0:07:51.520 --> 0:07:54.120
<v Speaker 1>startup founders. And that really is something that pisses me

0:07:54.160 --> 0:07:56.720
<v Speaker 1>off as well, because AI took thirty three percent of

0:07:56.840 --> 0:07:58.960
<v Speaker 1>US venture capital last year and I think they're probably

0:07:58.960 --> 0:08:02.680
<v Speaker 1>taking more this year. And what's insane is nothing has

0:08:02.720 --> 0:08:07.000
<v Speaker 1>been proven. None of these companies are profit generating, none

0:08:07.080 --> 0:08:10.360
<v Speaker 1>of these products are popular. Every single day, it feels

0:08:10.400 --> 0:08:12.640
<v Speaker 1>like a study comes out saying, yeah, no one really

0:08:12.760 --> 0:08:14.720
<v Speaker 1>likes this stuff, no one wants to pay for it.

0:08:15.600 --> 0:08:18.600
<v Speaker 1>And by my calculations, I think the whole industry's revenue

0:08:18.640 --> 0:08:21.840
<v Speaker 1>have include core we even all those shits, maybe fifty

0:08:21.840 --> 0:08:24.840
<v Speaker 1>five billion dollars in twenty twenty five. That's on four

0:08:24.920 --> 0:08:30.080
<v Speaker 1>hundred billion plus capex. Is anyone having fun? Does this

0:08:30.240 --> 0:08:32.960
<v Speaker 1>feel like the future or does this feel like the

0:08:32.960 --> 0:08:35.360
<v Speaker 1>future of how we're going to blow hundreds of billions

0:08:35.400 --> 0:08:39.640
<v Speaker 1>of dollars on nothing? And the longer it takes for

0:08:39.679 --> 0:08:42.200
<v Speaker 1>people to walk away from open Ai, and the longer

0:08:42.240 --> 0:08:45.840
<v Speaker 1>it takes for this company to die, the more damage

0:08:46.120 --> 0:08:48.880
<v Speaker 1>that's going to be caused. Because right now in video

0:08:48.960 --> 0:08:52.000
<v Speaker 1>is putting everything on open Ai. If we had functioning

0:08:52.040 --> 0:08:55.440
<v Speaker 1>regulatory boards, we'd actually have something done about this. But instead,

0:08:55.480 --> 0:08:57.320
<v Speaker 1>what's going to happen is they're going to inflate this

0:08:57.360 --> 0:09:00.000
<v Speaker 1>bubble just a little further, and then when it bursts,

0:09:00.080 --> 0:09:02.760
<v Speaker 1>it's going to fuck retail investors and a ton of

0:09:02.800 --> 0:09:05.800
<v Speaker 1>venture capitalists too, ton of private equity people as well,

0:09:05.840 --> 0:09:09.160
<v Speaker 1>which will be kind of funny, but the financial apocalypse

0:09:09.200 --> 0:09:13.320
<v Speaker 1>will hit everyone and it will be grizzly. But I'll

0:09:13.360 --> 0:09:16.040
<v Speaker 1>be here to talk about it. I'm having the time

0:09:16.080 --> 0:09:18.199
<v Speaker 1>of my life doing this show. Next week's three part

0:09:18.240 --> 0:09:21.040
<v Speaker 1>Who is Completely Insane? I wrote it in kind of

0:09:21.040 --> 0:09:25.800
<v Speaker 1>a two day, fifteen thousand word kind of like ejection.

0:09:25.880 --> 0:09:26.960
<v Speaker 1>I was going to say, I was going to say

0:09:27.000 --> 0:09:29.559
<v Speaker 1>another word, which wouldn't have sounded quite as good, But nevertheless,

0:09:29.600 --> 0:09:33.439
<v Speaker 1>that's the fun brandy part of these monologues. But I'm

0:09:33.440 --> 0:09:35.080
<v Speaker 1>really excited for you to hear it because I pulled

0:09:35.080 --> 0:09:37.600
<v Speaker 1>together a lot of numbers that show how fascical this

0:09:37.720 --> 0:09:41.040
<v Speaker 1>industry has become. And I think as the bubble approaches

0:09:41.080 --> 0:09:42.839
<v Speaker 1>its eventual burst, a lot of you are going to

0:09:42.880 --> 0:09:45.400
<v Speaker 1>feel really good about how right you are. Maybe some

0:09:45.440 --> 0:09:47.800
<v Speaker 1>of you still disagree with me, and I will work

0:09:47.880 --> 0:09:50.600
<v Speaker 1>to compel you, and even if I don't, I hope

0:09:50.640 --> 0:09:53.520
<v Speaker 1>you have a good time listening. Got great episodes coming

0:09:53.559 --> 0:09:55.480
<v Speaker 1>up in the future as well. Going to be going

0:09:55.559 --> 0:09:58.040
<v Speaker 1>to see Stephen Burke of Gamers next is, do a

0:09:58.040 --> 0:10:01.000
<v Speaker 1>long interview with him, some fun you a better offline,

0:10:01.480 --> 0:10:04.520
<v Speaker 1>and then I'm already working on some big guests. It's

0:10:04.520 --> 0:10:06.560
<v Speaker 1>going to be a really fun few months. I don't

0:10:06.600 --> 0:10:09.720
<v Speaker 1>know if anyone else, but I'm definitely enjoying doing this podcast.

0:10:09.920 --> 0:10:11.120
<v Speaker 1>Thank you, zever for listening.