WEBVTT - Monologue: The Questions I'm Asking

0:00:02.480 --> 0:00:06.720
<v Speaker 1>All zo media. You know what time it is. It's

0:00:06.720 --> 0:00:08.800
<v Speaker 1>your better offline monologue for the week, and I'm ed

0:00:08.880 --> 0:00:19.680
<v Speaker 1>z Tron. Now. Today's monologue is an attempt to bring

0:00:19.720 --> 0:00:21.920
<v Speaker 1>you further into my work, which I've kind of already

0:00:21.960 --> 0:00:25.120
<v Speaker 1>been doing with multiple listeners materially contributing to the reporting

0:00:25.160 --> 0:00:28.160
<v Speaker 1>behind this show. So many of you are eager to help,

0:00:28.200 --> 0:00:30.319
<v Speaker 1>and if you want to help, well, reach out to

0:00:30.360 --> 0:00:34.040
<v Speaker 1>me with information and do so on signal ez itron

0:00:34.120 --> 0:00:38.879
<v Speaker 1>dot seventy six. That's ez it r o N or

0:00:38.920 --> 0:00:41.159
<v Speaker 1>e z it r o N if you're Canadian or

0:00:41.200 --> 0:00:44.720
<v Speaker 1>British dot seventy six on signal. Also, before I go

0:00:44.800 --> 0:00:46.919
<v Speaker 1>any further, I've had a lot of emails and reddits

0:00:46.960 --> 0:00:49.240
<v Speaker 1>and as such, if you're wondering about my thoughts on

0:00:49.280 --> 0:00:52.560
<v Speaker 1>the Open Ai Disney deal, they are as followers. Wow,

0:00:52.680 --> 0:00:55.640
<v Speaker 1>Disney invested a billion dollars in open Ai. That's about

0:00:55.760 --> 0:00:59.200
<v Speaker 1>month's worth of inference. This only delays the inevitable. No

0:00:59.240 --> 0:01:02.920
<v Speaker 1>one cares. They're threatening Google with legal action. Google will

0:01:02.920 --> 0:01:05.400
<v Speaker 1>settle and they'll do exactly the same crap with VO.

0:01:05.880 --> 0:01:09.080
<v Speaker 1>Nothing will happen, a bunch of money will get lost. Similarly,

0:01:09.080 --> 0:01:11.759
<v Speaker 1>if you've heard about time naming AI executives as its

0:01:11.840 --> 0:01:15.120
<v Speaker 1>Person of the Year. Remember that Mark Bennioff of Salesforce

0:01:15.280 --> 0:01:19.440
<v Speaker 1>Huji Booster owns the publication and literally ran advertorial for

0:01:19.560 --> 0:01:25.520
<v Speaker 1>Salesforce's agent Force. Wank. It's nothing, nothing's happening. It's just

0:01:25.600 --> 0:01:27.800
<v Speaker 1>more money being passed around, so we can all get

0:01:27.840 --> 0:01:31.119
<v Speaker 1>blown on nothing. But back to the rest of the monologue.

0:01:31.480 --> 0:01:33.360
<v Speaker 1>I'm going to tell you about the things I need

0:01:33.400 --> 0:01:36.800
<v Speaker 1>to do a better job. Not don't literally mean the research.

0:01:36.840 --> 0:01:39.320
<v Speaker 1>I mean, if you know this stuff, I need your help.

0:01:39.920 --> 0:01:42.000
<v Speaker 1>Let's start with a big one. If you have first

0:01:42.080 --> 0:01:46.360
<v Speaker 1>hand knowledge of GPUs being warehouse, this is Nvidia AI GPUs.

0:01:46.440 --> 0:01:50.080
<v Speaker 1>Please God, if you reach out with this about gaming GPUs,

0:01:50.120 --> 0:01:52.320
<v Speaker 1>I'll be genuinely pissed off of you. But I need

0:01:52.360 --> 0:01:54.440
<v Speaker 1>to know about this. I want pictures if you can

0:01:54.440 --> 0:01:56.560
<v Speaker 1>get them. I want numbers. No, I want to know

0:01:56.720 --> 0:01:59.400
<v Speaker 1>whose GPUs they are and ideally what models A one

0:01:59.480 --> 0:02:01.680
<v Speaker 1>hundreds eight one hundred's probably not going to be that,

0:02:01.960 --> 0:02:05.560
<v Speaker 1>it's going to be those blackweld GPUs if you see them. Similarly,

0:02:05.640 --> 0:02:09.200
<v Speaker 1>if you've heard anything about Nvidia's aigpus that might not

0:02:09.280 --> 0:02:11.600
<v Speaker 1>be well known, reach out especially if it pertains to

0:02:11.600 --> 0:02:14.280
<v Speaker 1>how much they cost to run. Similarly, if you know

0:02:14.560 --> 0:02:18.040
<v Speaker 1>anything about Nvidia's blackweld GPUs that you suspect I don't,

0:02:18.040 --> 0:02:20.600
<v Speaker 1>please do reach out. I had somebody recently tell me

0:02:20.639 --> 0:02:24.760
<v Speaker 1>about remarkable failure rates and I want to know more. Similarly, costs.

0:02:24.919 --> 0:02:28.440
<v Speaker 1>I cannot ask for costs enough. Help me. Please, if

0:02:28.440 --> 0:02:31.520
<v Speaker 1>you have any information pretending to the revenues behind open Ai,

0:02:31.639 --> 0:02:35.760
<v Speaker 1>Anthropic or any major AI company, please get in touch. Similarly,

0:02:35.840 --> 0:02:37.440
<v Speaker 1>I want to hear from you if you work at

0:02:37.480 --> 0:02:40.800
<v Speaker 1>any of the large cloud companies and know anything about

0:02:40.800 --> 0:02:43.840
<v Speaker 1>the AI revenues or indeed the associated costs of running

0:02:43.840 --> 0:02:47.480
<v Speaker 1>AI on Google Cloud, Amazon Web Services and Microsoft Asure.

0:02:47.960 --> 0:02:50.679
<v Speaker 1>And yeah, if you have anything from open ai or

0:02:50.680 --> 0:02:53.119
<v Speaker 1>Anthropic about what they might be paying show as your

0:02:53.320 --> 0:02:56.440
<v Speaker 1>Google Cloud or Amazon Web services, please please please reach

0:02:56.480 --> 0:02:59.760
<v Speaker 1>out to me. Similarly, if you have anything of the

0:02:59.760 --> 0:03:03.000
<v Speaker 1>same from Corweave or Lambda or Nebius or any of those,

0:03:03.520 --> 0:03:06.800
<v Speaker 1>please reach out. Also, if you work at an AI startup,

0:03:06.840 --> 0:03:09.079
<v Speaker 1>one that pays open ai or Anthropic to use their

0:03:09.120 --> 0:03:11.920
<v Speaker 1>services or models, please tell me about those costs. We

0:03:11.960 --> 0:03:14.440
<v Speaker 1>don't know very much about them at all, and your

0:03:14.520 --> 0:03:17.919
<v Speaker 1>thoughts and your help, well, it'd be very much respected

0:03:17.960 --> 0:03:20.880
<v Speaker 1>and loved, and you'll be the better offline legend of

0:03:20.880 --> 0:03:22.960
<v Speaker 1>the week if you can help me out here, maybe

0:03:22.960 --> 0:03:26.040
<v Speaker 1>even the legend of the year. I'm also looking at

0:03:26.280 --> 0:03:29.440
<v Speaker 1>any and all information related to data center costs. I

0:03:29.480 --> 0:03:32.160
<v Speaker 1>want to know the total cost of ownership of any AIGP,

0:03:32.320 --> 0:03:35.480
<v Speaker 1>and that includes the A one hundred, the H one hundred,

0:03:35.600 --> 0:03:37.680
<v Speaker 1>h two hundred, B one hundred, B two hundred, B

0:03:37.720 --> 0:03:41.960
<v Speaker 1>three hundred and so on. Blackwell obviously my biggest priority.

0:03:42.440 --> 0:03:45.320
<v Speaker 1>I want to know how many GPUs hyperscalers have too,

0:03:45.600 --> 0:03:48.000
<v Speaker 1>So if you work at Amazon, Microsoft or Google and

0:03:48.040 --> 0:03:50.560
<v Speaker 1>you know how many of the goddamn GPUs they've got,

0:03:50.680 --> 0:03:52.920
<v Speaker 1>please reach out. And I want to know the hourly

0:03:53.000 --> 0:03:56.400
<v Speaker 1>cost of running these GPUs, ideally on a per GPU basis.

0:03:56.560 --> 0:03:58.320
<v Speaker 1>I'm trying to work out how much it actually costs

0:03:58.320 --> 0:04:00.280
<v Speaker 1>to run these goddamn things, and it's in the same

0:04:00.360 --> 0:04:03.320
<v Speaker 1>we don't know now my Grail data. The things that

0:04:03.360 --> 0:04:06.400
<v Speaker 1>would materially change my reporting other than everything I just

0:04:06.480 --> 0:04:10.680
<v Speaker 1>mentioned would be the underlying cost of running large language models,

0:04:10.720 --> 0:04:13.680
<v Speaker 1>which means understanding both how many GPUs are used for

0:04:13.800 --> 0:04:17.440
<v Speaker 1>inference in training and the actual costs of running said GPUs.

0:04:17.960 --> 0:04:19.880
<v Speaker 1>The costs of running GPUs are at the center of

0:04:19.920 --> 0:04:21.599
<v Speaker 1>the bubble, and I believe that truth will be what

0:04:21.720 --> 0:04:25.560
<v Speaker 1>bursts in. Though I'm repeating myself a little. My other

0:04:25.600 --> 0:04:28.560
<v Speaker 1>grail data is the underlying cost of running CHET, GPT, Google,

0:04:28.640 --> 0:04:32.200
<v Speaker 1>Gemini or any other major popular LLLM service such as

0:04:32.279 --> 0:04:36.400
<v Speaker 1>Cursor or rep them the same gohoest for Microsoft, Meta, Amazon, Google, Oracle,

0:04:36.480 --> 0:04:40.359
<v Speaker 1>or really any other major cloud provider. Look together, I

0:04:40.360 --> 0:04:42.200
<v Speaker 1>believe we can make the world better by providing the

0:04:42.240 --> 0:04:45.120
<v Speaker 1>public with the truth. And this era is one steeped

0:04:45.120 --> 0:04:48.719
<v Speaker 1>in outright lines and stal selfishness that deprives good companies

0:04:48.760 --> 0:04:51.640
<v Speaker 1>of funding and shoves dysfunctional software on millions of people

0:04:51.839 --> 0:04:55.000
<v Speaker 1>that just wish it would go the fuck away. Now,

0:04:55.080 --> 0:04:57.280
<v Speaker 1>so many of you have already been so helpful, and

0:04:57.320 --> 0:04:59.479
<v Speaker 1>I hope I can I can hear from more of you.

0:05:00.080 --> 0:05:02.279
<v Speaker 1>The show continues to grow. It's been a crazy year.

0:05:02.560 --> 0:05:05.080
<v Speaker 1>I'm insanely grateful to have all of you. And next

0:05:05.080 --> 0:05:07.200
<v Speaker 1>week we have a three part in video podcast that

0:05:07.240 --> 0:05:10.360
<v Speaker 1>tells you all about the largest we'd just weirdest. I'm

0:05:10.400 --> 0:05:13.039
<v Speaker 1>not fixing it. I'm not fixing that you want the

0:05:13.120 --> 0:05:16.440
<v Speaker 1>raw stuff anyway. In video, it's the weirdest company in

0:05:16.480 --> 0:05:19.760
<v Speaker 1>stock market history. Oh God, I'm gonna hear from you

0:05:19.800 --> 0:05:22.400
<v Speaker 1>all about that one. I really do love hearing from

0:05:22.440 --> 0:05:26.320
<v Speaker 1>you though it's been really crazy. Twenty twenty five has

0:05:26.360 --> 0:05:29.240
<v Speaker 1>been crazy and bad and good in many different ways.

0:05:29.760 --> 0:05:32.000
<v Speaker 1>And I'm sure I'm gonna I'm gonna hear fun stories

0:05:32.000 --> 0:05:33.960
<v Speaker 1>from you all because a bunch of you reach out

0:05:33.960 --> 0:05:36.080
<v Speaker 1>and no matter what I say, to reach out about or

0:05:36.120 --> 0:05:38.040
<v Speaker 1>not reach out about, and I kind of love it.

0:05:38.200 --> 0:05:41.440
<v Speaker 1>I love how chatty you all are. So yeah, look

0:05:41.480 --> 0:05:43.840
<v Speaker 1>forward to hearing from you, and I hope you enjoy

0:05:43.920 --> 0:05:46.800
<v Speaker 1>the podcast about the Weirdest company on the stock market.