WEBVTT - Monologue: AI Isn't Getting A Bailout

0:00:02.800 --> 0:00:08.320
<v Speaker 1>Zone Media. Hello and welcome to this week's Better Offline Monologue.

0:00:08.360 --> 0:00:18.880
<v Speaker 1>I'm your host ed Zitron. A lot of people have emailed,

0:00:18.920 --> 0:00:21.400
<v Speaker 1>messaged and plugged me that the natural end of the

0:00:21.440 --> 0:00:24.520
<v Speaker 1>AI bubble is some kind of government bailouts, something resembling

0:00:24.560 --> 0:00:28.760
<v Speaker 1>the TARP program that bailed out financial institutions overleveraged during

0:00:28.800 --> 0:00:31.240
<v Speaker 1>the Great Financial Crisis. And I want to be clear

0:00:31.280 --> 0:00:34.240
<v Speaker 1>about something. This is different, and a bailout is very

0:00:34.360 --> 0:00:38.000
<v Speaker 1>very very unlikely. As a reminder, TOP stood for Troubled

0:00:38.040 --> 0:00:41.720
<v Speaker 1>Asset Relief Program. It existed, unsurprisingly, to buy the distressed

0:00:41.720 --> 0:00:44.800
<v Speaker 1>assets that banks had invested in during the Great Financial Crisis,

0:00:44.960 --> 0:00:47.879
<v Speaker 1>primarily the ridiculous bets made on the housing market made

0:00:47.960 --> 0:00:50.040
<v Speaker 1>up of people who had got mortgages they could not

0:00:50.120 --> 0:00:54.280
<v Speaker 1>pay for. Now, these bailouts existed to stabilize banks that

0:00:54.320 --> 0:00:56.920
<v Speaker 1>were going out of business, along with consumers' money being

0:00:56.920 --> 0:00:59.080
<v Speaker 1>set on fire if they did so, as well as

0:00:59.120 --> 0:01:01.960
<v Speaker 1>auto manufacturer like GM and Chrysler that the contagion had

0:01:02.000 --> 0:01:03.800
<v Speaker 1>spread to. And while I'm not going to go into

0:01:03.840 --> 0:01:06.360
<v Speaker 1>massive amounts of detail, haven't got all day, I can

0:01:06.400 --> 0:01:09.920
<v Speaker 1>make it pretty simple. These bailouts existed to stabilize systemically

0:01:10.000 --> 0:01:12.800
<v Speaker 1>necessary companies that had they died, would have tanked the

0:01:12.920 --> 0:01:15.399
<v Speaker 1>US economy and markets at the same time and also

0:01:15.520 --> 0:01:19.759
<v Speaker 1>lost people's money. Unlike the Great Financial Crisis, very few

0:01:19.800 --> 0:01:23.000
<v Speaker 1>of the major players in AI are either systemically necessary

0:01:23.080 --> 0:01:24.800
<v Speaker 1>or in danger of dying as the result of the

0:01:24.840 --> 0:01:29.600
<v Speaker 1>AI bubble bursting. Furthermore, there really isn't anything too bailout.

0:01:30.040 --> 0:01:32.640
<v Speaker 1>So let's start with Nvidia, the largest company on the

0:01:32.640 --> 0:01:36.080
<v Speaker 1>stock market. While GPU sales make up woods of eighty

0:01:36.080 --> 0:01:38.520
<v Speaker 1>eight percent of its revenue and Video will not die

0:01:38.560 --> 0:01:40.640
<v Speaker 1>if the AI bubble bursts, though it will see a

0:01:40.680 --> 0:01:43.960
<v Speaker 1>massive contraction as a result of its biggest business line collapsing.

0:01:44.600 --> 0:01:46.760
<v Speaker 1>It's also important to know that in Nvidia's massive stock

0:01:46.800 --> 0:01:50.040
<v Speaker 1>price is a direct result of its credible growth. It

0:01:50.160 --> 0:01:53.240
<v Speaker 1>isn't enough for in Vidia to continue selling lots of GPUs.

0:01:53.440 --> 0:01:55.840
<v Speaker 1>It must continue to increase revenues by at least sixty

0:01:55.880 --> 0:01:59.240
<v Speaker 1>percent year of a year, every single quarter, which means

0:01:59.280 --> 0:02:01.440
<v Speaker 1>that for in Video to continue being the most beloved

0:02:01.480 --> 0:02:04.280
<v Speaker 1>boy and stock in the in history of the stock market,

0:02:04.480 --> 0:02:07.360
<v Speaker 1>it needs to make ninety one billion dollars in revenue

0:02:07.480 --> 0:02:09.720
<v Speaker 1>in a year's time, and that's just in one quarter.

0:02:09.800 --> 0:02:13.880
<v Speaker 1>By the way, this isn't something a bailout can fix.

0:02:14.000 --> 0:02:17.200
<v Speaker 1>Even in the most crony capitalist fantasy you can think of.

0:02:17.800 --> 0:02:20.560
<v Speaker 1>A bailout would have to continually sink tens of billions

0:02:20.600 --> 0:02:24.040
<v Speaker 1>of dollars directly into Invidia, buying GPUs and increasing amounts

0:02:24.240 --> 0:02:28.120
<v Speaker 1>every three months in perpetuity to continue its valuation. And

0:02:28.120 --> 0:02:30.680
<v Speaker 1>while some of you may say, booh, troub will do

0:02:30.840 --> 0:02:35.160
<v Speaker 1>themon you'd live in a boring, dumerous fancy based on nothing.

0:02:35.200 --> 0:02:38.680
<v Speaker 1>And I'm sick of hearing about it. Good lord, listen

0:02:38.760 --> 0:02:41.880
<v Speaker 1>to me. Top was a seven hundred billion dollar program

0:02:41.880 --> 0:02:44.520
<v Speaker 1>that only ended up investing around four hundred and forty

0:02:44.520 --> 0:02:48.119
<v Speaker 1>three point five billion dollars or so, or put another way,

0:02:48.280 --> 0:02:50.480
<v Speaker 1>less than than the five hundred billion dollars of orders

0:02:50.480 --> 0:02:53.880
<v Speaker 1>that Invidia CEO Jensen Je claims are on Invidia's books

0:02:53.880 --> 0:02:56.560
<v Speaker 1>through the end of twenty twenty six. Are we meant

0:02:56.560 --> 0:02:59.280
<v Speaker 1>to do this forever? Is that the plan? Are you

0:02:59.320 --> 0:03:00.760
<v Speaker 1>saying that Trump is just going to go.

0:03:00.720 --> 0:03:02.880
<v Speaker 2>There, We're going to buy gps or everyone, We're going

0:03:02.960 --> 0:03:05.440
<v Speaker 2>to bury them every year every year? Obey so many

0:03:05.440 --> 0:03:07.360
<v Speaker 2>of the most of you, No, it's not gonna happen.

0:03:07.600 --> 0:03:09.080
<v Speaker 2>Though I guess he has kind of done that with

0:03:09.120 --> 0:03:11.960
<v Speaker 2>the Department of Energy. You've probably heard of this deal.

0:03:12.000 --> 0:03:14.040
<v Speaker 2>By the way, it's one hundred thousand GPUs. That isn't

0:03:14.040 --> 0:03:14.720
<v Speaker 2>going to do shit.

0:03:15.280 --> 0:03:17.799
<v Speaker 1>Jensen Wong needs all your money and eats it now

0:03:17.840 --> 0:03:24.200
<v Speaker 1>every quarter forever. Okay, Okay, it's not going to happen now, Okay.

0:03:24.240 --> 0:03:28.120
<v Speaker 1>Another annoying talking point you've probably heard is that open

0:03:28.160 --> 0:03:30.399
<v Speaker 1>ai is somehow too big to fail, when in fact

0:03:30.440 --> 0:03:33.320
<v Speaker 1>it's too small to pull apart into enough parts to eat.

0:03:34.040 --> 0:03:37.800
<v Speaker 1>Open Ai has promised one point four trillion dollars in

0:03:37.840 --> 0:03:41.280
<v Speaker 1>compute deals, convincing Oracle and others to build massive data

0:03:41.320 --> 0:03:43.840
<v Speaker 1>centers in its honor. Yet the reality is that this

0:03:43.880 --> 0:03:46.560
<v Speaker 1>company isn't even going to hit its thirteen billion dollars

0:03:46.640 --> 0:03:49.920
<v Speaker 1>in projected revenue this year, and we'll burn billions of

0:03:49.920 --> 0:03:52.240
<v Speaker 1>dollars in inference loan and doesn't really have a path

0:03:52.240 --> 0:03:53.080
<v Speaker 1>to profitability.

0:03:53.480 --> 0:03:54.000
<v Speaker 3>Two big to.

0:03:53.960 --> 0:03:56.560
<v Speaker 1>Fail means that something may kill our economy or markets

0:03:56.560 --> 0:03:58.640
<v Speaker 1>if it dies, And as loud and annoying as Clammy

0:03:58.680 --> 0:04:02.000
<v Speaker 1>sam Utman might be, open AI's death wouldn't kill either

0:04:02.040 --> 0:04:06.320
<v Speaker 1>of those. Let's break it down. Open Ai dying would

0:04:06.320 --> 0:04:08.920
<v Speaker 1>be symbolically lethal to the large language model era and

0:04:08.920 --> 0:04:11.880
<v Speaker 1>create an immediate and permanent chill in AI investments. Don't

0:04:11.880 --> 0:04:14.280
<v Speaker 1>get me wrong. Its death would also likely be an

0:04:14.280 --> 0:04:16.960
<v Speaker 1>existential threat to Crusoe, the company building its data center

0:04:16.960 --> 0:04:19.640
<v Speaker 1>in Abilene, Texas for Oracle. I spoke to someone on

0:04:19.680 --> 0:04:21.600
<v Speaker 1>the ground there and they say, if the money runs out,

0:04:21.640 --> 0:04:23.279
<v Speaker 1>everyone just stops and goes home.

0:04:23.400 --> 0:04:24.240
<v Speaker 3>That'll be fine.

0:04:24.520 --> 0:04:27.080
<v Speaker 1>I don't know the extent of Cruser's revenues, however, but

0:04:27.120 --> 0:04:28.640
<v Speaker 1>I do know that Oracle is on the hook for

0:04:28.680 --> 0:04:32.200
<v Speaker 1>a billion dollars a year for Abilene for fifteen goddamn years.

0:04:32.720 --> 0:04:35.160
<v Speaker 1>An Oracle has also taken on massive amounts of debt

0:04:35.200 --> 0:04:38.440
<v Speaker 1>to build Abilene and other data centers in Shackleford, Texas,

0:04:38.440 --> 0:04:41.000
<v Speaker 1>and I think somewhere in Wisconsin is in New Mexico

0:04:41.160 --> 0:04:44.400
<v Speaker 1>one now and started this with an eighteen billion dollar

0:04:44.440 --> 0:04:47.040
<v Speaker 1>bond sale and in the works eighty eight billion dollar

0:04:47.080 --> 0:04:49.880
<v Speaker 1>debt package to build data centers for Open AI. Well,

0:04:49.920 --> 0:04:52.680
<v Speaker 1>Oracle does run the risk of default and credit default swaps,

0:04:52.680 --> 0:04:55.159
<v Speaker 1>which is people betting again well betting that they might

0:04:55.279 --> 0:04:59.400
<v Speaker 1>default on their obligations. That's going up. The risk is there,

0:05:00.080 --> 0:05:02.520
<v Speaker 1>but there's still a real company that makes real things.

0:05:03.040 --> 0:05:06.560
<v Speaker 1>Microsoft owns open Ai IP and research, which would mean

0:05:06.600 --> 0:05:09.279
<v Speaker 1>that Microsoft pretty much absorbed those eight one hundred million

0:05:09.320 --> 0:05:12.480
<v Speaker 1>weekly active users directly into Copilot. And that's if it

0:05:12.600 --> 0:05:15.600
<v Speaker 1>even chose to retain them and just shut down chat

0:05:15.640 --> 0:05:19.080
<v Speaker 1>GBT directed it to a very flimsy version of Copilot

0:05:19.080 --> 0:05:20.480
<v Speaker 1>and didn't let people do half the.

0:05:20.440 --> 0:05:22.080
<v Speaker 3>Shit they did with them.

0:05:22.760 --> 0:05:25.960
<v Speaker 1>Bailing out open Ai would not do anything, in part

0:05:26.000 --> 0:05:29.400
<v Speaker 1>because open ai burns billions of dollars and doesn't really

0:05:29.800 --> 0:05:33.360
<v Speaker 1>do much in the large economy itself. It's revenues, which

0:05:33.400 --> 0:05:36.440
<v Speaker 1>I estimate to be most four point five billion dollars

0:05:36.440 --> 0:05:37.520
<v Speaker 1>through the end of September.

0:05:37.880 --> 0:05:39.600
<v Speaker 3>By the way, if you're listening to this and you.

0:05:39.600 --> 0:05:42.640
<v Speaker 1>Think that my revenue numbers for open Ai, my inference

0:05:42.720 --> 0:05:47.000
<v Speaker 1>numbers were delayed somehow, it's a cruel accounting. Stop fucking

0:05:47.040 --> 0:05:49.200
<v Speaker 1>with me. I know what I'm doing. Anyway, those revenues

0:05:49.240 --> 0:05:51.839
<v Speaker 1>are inconsequential, and the eight point six to seven billion

0:05:51.880 --> 0:05:54.080
<v Speaker 1>dollars it spent on inference in that period is while

0:05:54.080 --> 0:05:57.240
<v Speaker 1>a large chunk of Vizio revenue from Microsoft only really

0:05:57.279 --> 0:06:00.080
<v Speaker 1>accounting for three point eight eight percent or so, the

0:06:00.080 --> 0:06:02.479
<v Speaker 1>two hundred and twenty three billion dollars Microsoft has made

0:06:02.480 --> 0:06:05.800
<v Speaker 1>in that period. Losing open AI would hurt but hardly

0:06:05.839 --> 0:06:08.680
<v Speaker 1>threaten the health of anyone in Redmond. The same ghost

0:06:08.720 --> 0:06:12.640
<v Speaker 1>for Anthropic or any other generative AI company. These companies

0:06:12.640 --> 0:06:16.080
<v Speaker 1>aren't really spending that much, nor are they fueling much

0:06:16.440 --> 0:06:20.080
<v Speaker 1>of any actual real economic activity. There's only around sixty

0:06:20.120 --> 0:06:22.839
<v Speaker 1>billion dollars of actual AI revenue in twenty twenty five.

0:06:23.040 --> 0:06:25.600
<v Speaker 1>In a tech industry where Amazon made one hundred and

0:06:25.640 --> 0:06:28.440
<v Speaker 1>eighty billion dollars and Google one hundred and two billion

0:06:28.480 --> 0:06:32.320
<v Speaker 1>dollars in their last quarters. The disappearance of llm's writ

0:06:32.480 --> 0:06:35.120
<v Speaker 1>large would have a miniscule effect on actual real dollars

0:06:35.240 --> 0:06:39.520
<v Speaker 1>entering the economy, other than investments in data centers, which

0:06:39.560 --> 0:06:43.320
<v Speaker 1>is where I think there's, if any, some kind of bailout.

0:06:44.680 --> 0:06:48.240
<v Speaker 1>As I've mentioned recently, there's been about fifty billion dollars

0:06:48.760 --> 0:06:51.080
<v Speaker 1>or more even every quarter for the last three quarters

0:06:51.080 --> 0:06:53.200
<v Speaker 1>going into building data centers, and I think that many

0:06:53.240 --> 0:06:57.120
<v Speaker 1>of these projects end up defaulting. There's a chance a

0:06:57.320 --> 0:07:00.360
<v Speaker 1>chance that these are somehow bailed out, if an only

0:07:00.400 --> 0:07:03.599
<v Speaker 1>if they provide an actual existential risk to a major

0:07:03.640 --> 0:07:06.680
<v Speaker 1>asset firm, A Blackstone or areas of what have you.

0:07:07.040 --> 0:07:08.400
<v Speaker 1>I don't know if Areas is in them and just

0:07:08.480 --> 0:07:11.840
<v Speaker 1>naming them at this point. Nevertheless, there is one that

0:07:11.920 --> 0:07:15.360
<v Speaker 1>I the one company I could see if any bailout happens.

0:07:15.360 --> 0:07:18.320
<v Speaker 1>And I'm really hesitant to say I think there's any chance.

0:07:19.200 --> 0:07:23.040
<v Speaker 1>Oracle Oracle I think is the one chance. Oracle is

0:07:23.040 --> 0:07:25.400
<v Speaker 1>currently massively leveraged and going into debt to the point

0:07:25.440 --> 0:07:27.400
<v Speaker 1>it's eating into its free cash flow, and it's a

0:07:27.440 --> 0:07:29.800
<v Speaker 1>close ally of the Trump administration. I could see them

0:07:29.800 --> 0:07:33.720
<v Speaker 1>getting a lifeline to escape these ridiculous loans. And while

0:07:33.760 --> 0:07:36.040
<v Speaker 1>you may say there's some sort of grand plan or

0:07:36.040 --> 0:07:38.480
<v Speaker 1>the administration will agree to pay these companies to keep

0:07:38.480 --> 0:07:40.960
<v Speaker 1>them alive, I don't see it happening. Is doing so

0:07:41.120 --> 0:07:43.840
<v Speaker 1>won't support the economy, the markets really do anything other

0:07:43.880 --> 0:07:49.080
<v Speaker 1>than prolongedly inevitable, It's not going to happen, and I

0:07:49.080 --> 0:07:51.000
<v Speaker 1>I don't know. I'm I want to phrase this in

0:07:51.040 --> 0:07:52.520
<v Speaker 1>a nice way because I don't want it seem like

0:07:52.560 --> 0:07:55.000
<v Speaker 1>an attack. Every time you say something like this, you're

0:07:55.000 --> 0:07:57.440
<v Speaker 1>doing marketing for the AI industry.

0:07:57.760 --> 0:07:59.360
<v Speaker 3>Dumerism is marketing.

0:07:59.480 --> 0:08:03.040
<v Speaker 1>At this point, I'm so scared of what they'll they'll create,

0:08:03.240 --> 0:08:05.840
<v Speaker 1>they are gonna get bail out open Aye is too

0:08:05.840 --> 0:08:06.520
<v Speaker 1>big to fail.

0:08:06.760 --> 0:08:09.680
<v Speaker 3>All these are extrapolations of your fear. I get it.

0:08:09.840 --> 0:08:13.160
<v Speaker 1>They're also marketing. You are helping them. You are helping

0:08:13.200 --> 0:08:17.200
<v Speaker 1>them market their services by making them seem systemically relevant

0:08:17.200 --> 0:08:19.920
<v Speaker 1>when they are not, by making them seem like they're

0:08:19.920 --> 0:08:24.360
<v Speaker 1>innovative when they're not. Honestly, I need you to just

0:08:24.400 --> 0:08:26.680
<v Speaker 1>live in reality and look at what's actually happening, which

0:08:26.720 --> 0:08:30.240
<v Speaker 1>is sweet fuck all. In many cases, every one of

0:08:30.280 --> 0:08:32.440
<v Speaker 1>these wanky studies you hear come out so on, it's

0:08:32.480 --> 0:08:35.079
<v Speaker 1>like ten percent of jobs can already be automated by AI.

0:08:35.360 --> 0:08:37.959
<v Speaker 1>CNBC should be fucking ashamed of themselves for running a

0:08:37.960 --> 0:08:38.600
<v Speaker 1>story about it.

0:08:38.600 --> 0:08:39.280
<v Speaker 3>It's a PhD.

0:08:39.640 --> 0:08:42.680
<v Speaker 1>Congrats to the Better Offline Reddit user resource. It's a

0:08:42.679 --> 0:08:47.800
<v Speaker 1>PhD paper. And also it's referring to random skills rather

0:08:47.840 --> 0:08:50.760
<v Speaker 1>than jobs, and skills and jobs are two different fucking things.

0:08:51.240 --> 0:08:56.280
<v Speaker 1>I'm so goddamn tired of this era, good Lord, But nevertheless,

0:08:56.559 --> 0:08:58.520
<v Speaker 1>the next two weeks are going to be really interesting.

0:08:58.559 --> 0:09:00.360
<v Speaker 1>I'm working on a three or four part or about

0:09:00.360 --> 0:09:02.319
<v Speaker 1>my worries around in video, and I'm going to tell

0:09:02.320 --> 0:09:05.400
<v Speaker 1>you all about in Video, and how not similar they

0:09:05.440 --> 0:09:08.559
<v Speaker 1>are to Enron and Video is nothing like Enron or

0:09:08.640 --> 0:09:12.240
<v Speaker 1>Loosen or Welcome or WinStar or all these other companies.

0:09:12.280 --> 0:09:14.640
<v Speaker 3>And we're going to go into as well. And tomorrow I've.

0:09:14.480 --> 0:09:18.560
<v Speaker 1>Got a fun interview with Nathan Grayson of Aftermath Your Love.

0:09:19.240 --> 0:09:23.440
<v Speaker 1>I might next week take a knee and just run

0:09:23.520 --> 0:09:26.319
<v Speaker 1>the better offline in studio thing I just did with

0:09:26.840 --> 0:09:29.319
<v Speaker 1>Lisa from CNN, or I might do the three or

0:09:29.320 --> 0:09:31.360
<v Speaker 1>four bar. It really comes down to how I'm feeling.

0:09:31.440 --> 0:09:34.400
<v Speaker 1>It's been a rough few weeks. Health wise. I'm doing fine.

0:09:34.400 --> 0:09:36.720
<v Speaker 1>I've just been sick and I've just been quite stressed.

0:09:37.120 --> 0:09:39.720
<v Speaker 1>I must again say thank you for all the lovely

0:09:39.760 --> 0:09:42.120
<v Speaker 1>emails and messages you've been sending.

0:09:42.160 --> 0:09:43.600
<v Speaker 3>You're all so lovely. I love you all.

0:09:44.320 --> 0:09:47.240
<v Speaker 1>Couldn't be lucky with the listeners I've got. Tell your

0:09:47.240 --> 0:09:49.640
<v Speaker 1>friends about the show, tell them about the newsletter, tell

0:09:49.679 --> 0:09:53.160
<v Speaker 1>them about me, tell them about my strange proclivities, but

0:09:53.240 --> 0:09:55.640
<v Speaker 1>also keep listening. I really do appreciate all of you

0:09:56.080 --> 0:09:58.679
<v Speaker 1>enjoyed tomorrow's episode and whatever I put out next week,

0:09:58.720 --> 0:10:01.000
<v Speaker 1>and there will be a monologue regard Godless, unless I

0:10:01.080 --> 0:10:02.760
<v Speaker 1>do the three or four par in which case, that's

0:10:02.760 --> 0:10:04.680
<v Speaker 1>going to be a monologue for the week. And yes,

0:10:04.800 --> 0:10:06.920
<v Speaker 1>by the way, I realize me just talking is always

0:10:06.960 --> 0:10:07.520
<v Speaker 1>a monologue.

0:10:07.559 --> 0:10:10.079
<v Speaker 3>I get it. You're so smart. You are though, really

0:10:10.200 --> 0:10:11.040
<v Speaker 3>thank you for listening.