WEBVTT - Monologue: We Need To Talk About CoreWeave

0:00:02.160 --> 0:00:07.280
<v Speaker 1>Zone Media burd Up. I'm at Zetron and this is

0:00:07.320 --> 0:00:18.360
<v Speaker 1>your weekly better offline monologue. So last week I put

0:00:18.360 --> 0:00:21.479
<v Speaker 1>out a newsletter about core Weave, an a cloud provider

0:00:21.520 --> 0:00:24.840
<v Speaker 1>that sells GPU compute to AI companies looking to run

0:00:24.960 --> 0:00:27.680
<v Speaker 1>or train their models, and they recently filed the paperwork

0:00:27.720 --> 0:00:31.680
<v Speaker 1>to go public. Now, the newsletter explained the shaky fundamentals

0:00:31.680 --> 0:00:34.519
<v Speaker 1>of the business, which is putting it mildly and ask

0:00:34.600 --> 0:00:37.960
<v Speaker 1>the question how the fuck does core Weave survive. This

0:00:38.080 --> 0:00:41.040
<v Speaker 1>is a company that's drowning in debt and relying on one

0:00:40.960 --> 0:00:43.599
<v Speaker 1>company for more than half of its revenue. And said company,

0:00:43.640 --> 0:00:46.600
<v Speaker 1>Microsoft is actively pulling away from building new data center

0:00:46.640 --> 0:00:49.680
<v Speaker 1>capacity and has reportedly pulled back from some contracts with

0:00:49.800 --> 0:00:53.440
<v Speaker 1>core Weave, though it's not clear what those contracts are. Now,

0:00:53.479 --> 0:00:56.080
<v Speaker 1>this is an important story and one that raises questions

0:00:56.120 --> 0:00:58.560
<v Speaker 1>not just about the viability of core Weave, but the

0:00:58.600 --> 0:01:02.360
<v Speaker 1>generative AI industry. At that said, the subject matter is

0:01:02.400 --> 0:01:04.880
<v Speaker 1>a little dense to turn into a hulking two part episode,

0:01:04.920 --> 0:01:07.480
<v Speaker 1>so I decided to condense it into a shorter monologue

0:01:07.520 --> 0:01:10.720
<v Speaker 1>for you. Now, Core We've had intended to go public

0:01:10.800 --> 0:01:12.920
<v Speaker 1>a week ago. They might go public and a week

0:01:12.959 --> 0:01:15.960
<v Speaker 1>from now, and they're apparently going with like a twenty

0:01:16.319 --> 0:01:19.840
<v Speaker 1>five to thirty billion dollar valuation. While it's hardly a

0:01:19.880 --> 0:01:23.800
<v Speaker 1>recognizable name like say open AI or Microsoft or in Vidia,

0:01:23.840 --> 0:01:26.479
<v Speaker 1>this company is definitely worth taking a look at, if

0:01:26.520 --> 0:01:28.680
<v Speaker 1>not for the fact that it's arguably the first major

0:01:28.760 --> 0:01:31.720
<v Speaker 1>IPO that we've seen from the current generative AI hYP bubble,

0:01:32.040 --> 0:01:36.440
<v Speaker 1>and it's undoubtedly the biggest. Moreover, corwave is a company

0:01:36.440 --> 0:01:39.360
<v Speaker 1>that deals in the infrastructure aspect of AI, where one

0:01:39.400 --> 0:01:41.960
<v Speaker 1>would naturally assume is where all the money really is

0:01:42.240 --> 0:01:45.200
<v Speaker 1>putting up service for hyperscals to run their hallucination pro

0:01:45.280 --> 0:01:48.800
<v Speaker 1>and nonprofitable models. If in Vidia, the company that makes

0:01:48.800 --> 0:01:51.360
<v Speaker 1>the GPUs for generative AI, is selling the pick axes

0:01:51.360 --> 0:01:53.920
<v Speaker 1>for the current gold rush, cor we've well, they own

0:01:53.960 --> 0:01:56.560
<v Speaker 1>the land, or maybe they're the shovels. I'm not going

0:01:56.600 --> 0:02:01.080
<v Speaker 1>to get too far into that metaphors side up. Until

0:02:01.120 --> 0:02:03.720
<v Speaker 1>a few years ago, Corweave was a different company, a

0:02:03.720 --> 0:02:08.080
<v Speaker 1>cryptocurrency mining data center company to be specific. Now twenty

0:02:08.120 --> 0:02:11.240
<v Speaker 1>tens were turbulent for crypto and the company then pivoted

0:02:11.280 --> 0:02:14.320
<v Speaker 1>to providing high performance computing for third parties that didn't

0:02:14.320 --> 0:02:16.639
<v Speaker 1>want to run their own infrastructure for things like three

0:02:16.720 --> 0:02:19.760
<v Speaker 1>D modeling. Around the time of the launch of chair GPT,

0:02:20.240 --> 0:02:22.920
<v Speaker 1>cor We've pivoted again, this time towards providing compute to

0:02:22.960 --> 0:02:26.800
<v Speaker 1>hyperscalers and then generative AI interest. It gambled, and it

0:02:27.000 --> 0:02:30.680
<v Speaker 1>gambled successfully. You'd imagine that such a company would be

0:02:30.720 --> 0:02:34.639
<v Speaker 1>a thriving, healthy business though, right they I mean they

0:02:34.760 --> 0:02:37.680
<v Speaker 1>sell the surfer architecture. The demand is there, the growth

0:02:37.760 --> 0:02:41.680
<v Speaker 1>is there. I mean everyone wants generative AI. Right, it

0:02:41.680 --> 0:02:46.840
<v Speaker 1>wouldn't be another thing where the reality Jesus Christ. Even

0:02:46.880 --> 0:02:51.040
<v Speaker 1>a cursory glance at Corwey's financial disclosure documents reveals a

0:02:51.080 --> 0:02:55.680
<v Speaker 1>business that's precarious at best, and in my most uncharitable opinion,

0:02:55.840 --> 0:02:59.880
<v Speaker 1>it's utterly rancid. If this company was in any other industry,

0:03:00.240 --> 0:03:03.480
<v Speaker 1>be seen as a big pile of shit, except it's

0:03:03.560 --> 0:03:05.919
<v Speaker 1>one of the standard bearers of the generative AI boom

0:03:05.960 --> 0:03:09.959
<v Speaker 1>and so exists within its own reality distortion field. Now

0:03:10.000 --> 0:03:12.880
<v Speaker 1>Callw've makes its money by renting out GPUs to companies

0:03:12.880 --> 0:03:15.200
<v Speaker 1>looking to run or train their models, which means that

0:03:15.240 --> 0:03:18.120
<v Speaker 1>it's incredibly indicative of whether there's a real business in

0:03:18.200 --> 0:03:20.840
<v Speaker 1>generative AI and you'll be shocked to hear that there

0:03:20.919 --> 0:03:23.919
<v Speaker 1>isn't call. We've made just under two billion dollars in

0:03:23.960 --> 0:03:26.560
<v Speaker 1>twenty twenty four, yet somehow managed to lose eight hundred

0:03:26.560 --> 0:03:30.519
<v Speaker 1>and sixty three million dollars. Worse still, seventy seven percent

0:03:30.520 --> 0:03:34.359
<v Speaker 1>of its revenue comes from two companies, Microsoft sixty two percent, and,

0:03:34.480 --> 0:03:38.240
<v Speaker 1>although not said explicitly, most likely in Video fifteen percent.

0:03:38.280 --> 0:03:41.520
<v Speaker 1>Now the information is reported Project Gosprey, which is basically

0:03:41.640 --> 0:03:44.920
<v Speaker 1>Nvidia's stake in the fact that they're finaling money and

0:03:44.960 --> 0:03:47.840
<v Speaker 1>it's actually not immediately obvious what it is that in

0:03:47.920 --> 0:03:51.040
<v Speaker 1>Video is paying Core week four, though Invidio is both

0:03:51.080 --> 0:03:53.520
<v Speaker 1>invested in the company and provides it with priority access

0:03:53.520 --> 0:03:56.200
<v Speaker 1>to its chips. With Core, we've getting some products before

0:03:56.240 --> 0:04:00.040
<v Speaker 1>even Amazon or Microsoft. It's bonkers. And again, yeah, I

0:04:00.120 --> 0:04:04.400
<v Speaker 1>think with all this preferential accents Core, we've just printing

0:04:04.440 --> 0:04:07.760
<v Speaker 1>money right versus burning it, burning it by the hundreds

0:04:07.800 --> 0:04:10.839
<v Speaker 1>of millions full. I wish I could get this much

0:04:10.920 --> 0:04:13.600
<v Speaker 1>money and just I don't know, I would be doing

0:04:13.600 --> 0:04:16.320
<v Speaker 1>a far weirder podcast. It would be exactly the same thing,

0:04:16.320 --> 0:04:18.880
<v Speaker 1>better offline would exist. I would just have hundreds of

0:04:18.880 --> 0:04:21.480
<v Speaker 1>millions of dollars we'd have just like Samuel L. Jackson

0:04:21.839 --> 0:04:25.480
<v Speaker 1>coming in for some episodes. Anyway, I don't have that money.

0:04:25.520 --> 0:04:27.279
<v Speaker 1>If you'd like to give it to me, please please

0:04:27.320 --> 0:04:30.200
<v Speaker 1>do now. A few weeks ago, The Financial Times reported

0:04:30.240 --> 0:04:32.799
<v Speaker 1>that Microsoft had pulled out some of their core Weave contracts,

0:04:32.839 --> 0:04:36.479
<v Speaker 1>though Core We've denied the allegations, and mysteriously, a week later,

0:04:36.600 --> 0:04:39.360
<v Speaker 1>open Ai announced that they had an eleven point nine

0:04:39.440 --> 0:04:44.160
<v Speaker 1>billion dollar contract with Corewave to provide compute services. Now,

0:04:44.200 --> 0:04:47.320
<v Speaker 1>I can't say for certain. Can you think of another

0:04:47.360 --> 0:04:50.320
<v Speaker 1>reason Microsoft needed a whole bunch of GPU compute other

0:04:50.360 --> 0:04:55.039
<v Speaker 1>than to host open AI's models. Hmm, Now, I'm just

0:04:55.760 --> 0:04:58.360
<v Speaker 1>guessing here, but one has to wonder if it wasn't

0:04:58.400 --> 0:05:01.560
<v Speaker 1>so much open Ai se a new deal and more

0:05:01.600 --> 0:05:04.920
<v Speaker 1>at taking over the future compute than Microsoft was formally handling.

0:05:05.640 --> 0:05:09.080
<v Speaker 1>And yes, you're remembering correctly. Microsoft has recently canceled more

0:05:09.080 --> 0:05:12.400
<v Speaker 1>than the gigawater future compute capacity, though from what I've heard,

0:05:12.440 --> 0:05:15.640
<v Speaker 1>its core Weave contracts future or otherwise weren't included in

0:05:15.680 --> 0:05:18.800
<v Speaker 1>that number. Anyway, For core Weave to make that eleven

0:05:18.839 --> 0:05:21.440
<v Speaker 1>point nine billion dollars in revenue from open Ai, they're

0:05:21.440 --> 0:05:23.760
<v Speaker 1>going to need compute, which will cost them tens of

0:05:23.760 --> 0:05:26.680
<v Speaker 1>billions of dollars to build, which will be difficult as

0:05:26.720 --> 0:05:28.800
<v Speaker 1>the company has burdened with about eight billion dollars of

0:05:28.839 --> 0:05:31.800
<v Speaker 1>debt with horrifying interest rates, which made lead it to

0:05:31.839 --> 0:05:34.360
<v Speaker 1>pay upwards of one point five billion dollars a year

0:05:34.360 --> 0:05:36.880
<v Speaker 1>in loan payments, with one of them requiring Core Weave

0:05:36.920 --> 0:05:39.560
<v Speaker 1>to repay the loan with any other debt they raise

0:05:39.640 --> 0:05:42.599
<v Speaker 1>in the future. Core Weave, by my estimates, needs at

0:05:42.680 --> 0:05:45.440
<v Speaker 1>least thirty billion dollars to expand. It has about one

0:05:45.480 --> 0:05:47.400
<v Speaker 1>point three billion dollars in the bank and just under

0:05:47.440 --> 0:05:49.520
<v Speaker 1>four billion dollars left on its loans that they can draw.

0:05:49.760 --> 0:05:52.000
<v Speaker 1>One of them is a term loan where you can

0:05:52.080 --> 0:05:55.680
<v Speaker 1>draw more up until about June twenty twenty five. Although

0:05:55.920 --> 0:05:58.880
<v Speaker 1>Core We've will raise capital in its IPO, the amount

0:05:58.880 --> 0:06:01.160
<v Speaker 1>will be anywhere near any to meet its needs. And

0:06:01.200 --> 0:06:03.719
<v Speaker 1>when I say its needs, I mean like the ability

0:06:03.760 --> 0:06:07.120
<v Speaker 1>to service the revenue that they need to spend more

0:06:07.120 --> 0:06:09.640
<v Speaker 1>money really does not make a lick of sense when

0:06:09.680 --> 0:06:12.120
<v Speaker 1>you think about it, and you may wonder if things

0:06:12.120 --> 0:06:14.559
<v Speaker 1>can get worse, and the answer is that they can.

0:06:14.920 --> 0:06:18.359
<v Speaker 1>Coreweave has planned one point three gigawats of expansion and

0:06:18.480 --> 0:06:21.320
<v Speaker 1>its partner to build it is a public company called

0:06:21.360 --> 0:06:26.840
<v Speaker 1>Core Scientific. Different company, similar name, genuinely different. Here are

0:06:26.839 --> 0:06:30.200
<v Speaker 1>some facts about Core Scientific. They went public in twenty

0:06:30.240 --> 0:06:33.119
<v Speaker 1>twenty two in a disastrous spack merger and then filed

0:06:33.120 --> 0:06:36.600
<v Speaker 1>for bankruptcy in the same year. Their entire business has

0:06:36.640 --> 0:06:40.239
<v Speaker 1>been focused on mining bitcoin, which requires specialized acic computer

0:06:40.360 --> 0:06:44.000
<v Speaker 1>chips that are entirely different both in their construction and

0:06:44.040 --> 0:06:47.400
<v Speaker 1>their maintenance to GPUs, and they can only really be

0:06:47.480 --> 0:06:51.039
<v Speaker 1>used for mining crypto. They're not easily repurposed for other tasks,

0:06:51.040 --> 0:06:53.239
<v Speaker 1>and indeed their data centers are not just a plug

0:06:53.240 --> 0:06:56.440
<v Speaker 1>and play thing. You basically have to demolish them. And

0:06:56.680 --> 0:06:59.360
<v Speaker 1>another thing about Core Scientific they only made twenty four

0:06:59.400 --> 0:07:02.320
<v Speaker 1>million dollars twenty twenty four from selling AI compute related

0:07:02.360 --> 0:07:05.200
<v Speaker 1>services to their customers. That's around five percent of its

0:07:05.200 --> 0:07:07.560
<v Speaker 1>total revenue, with the rest coming from mining and selling crypto.

0:07:08.160 --> 0:07:12.240
<v Speaker 1>Also really dumb ass mistaken. Just make that I said customers,

0:07:12.640 --> 0:07:17.040
<v Speaker 1>I meant customer. Course Scientific has one customer and that

0:07:17.080 --> 0:07:21.080
<v Speaker 1>customer is it's core Weave. As of right now, Corewave

0:07:21.080 --> 0:07:25.080
<v Speaker 1>has approximately three hundred and sixty megawatts of compute power. Somehow,

0:07:25.120 --> 0:07:28.119
<v Speaker 1>it intends to build another one point three gigawatts worth

0:07:28.120 --> 0:07:30.040
<v Speaker 1>of compute using a partner that has never built an

0:07:30.080 --> 0:07:32.120
<v Speaker 1>AI data center and does not be able to have

0:07:32.200 --> 0:07:35.200
<v Speaker 1>any meaningful compute right now, this is all so good.

0:07:35.280 --> 0:07:39.520
<v Speaker 1>I love this. Let's fucking go. This is the big IPO,

0:07:39.880 --> 0:07:43.400
<v Speaker 1>this is the big aiaper. I am going insane every

0:07:43.440 --> 0:07:48.000
<v Speaker 1>time I read about these goddamn companies. But let's summarize.

0:07:48.880 --> 0:07:51.400
<v Speaker 1>Core Weave is burdened by interest payments that may belloon

0:07:51.520 --> 0:07:53.600
<v Speaker 1>to as much as two billion dollars a year, and

0:07:53.640 --> 0:07:56.240
<v Speaker 1>lost eight hundred and sixty three million dollars on two

0:07:56.240 --> 0:07:59.400
<v Speaker 1>billion dollars of revenue in twenty twenty four. Sixty two

0:07:59.440 --> 0:08:02.000
<v Speaker 1>percent of that revenue is from Microsoft, which has materially

0:08:02.000 --> 0:08:04.280
<v Speaker 1>pulled back on data center buildouts and may have dropped

0:08:04.280 --> 0:08:06.560
<v Speaker 1>some core Weave contracts, though core We've denies this is

0:08:06.560 --> 0:08:09.680
<v Speaker 1>the case. Of course, Corby's expansion, which is critical to

0:08:09.720 --> 0:08:12.360
<v Speaker 1>servicing future revenue and growth, requires it to invest tens

0:08:12.360 --> 0:08:15.400
<v Speaker 1>of billion dollars the core Weave does not have. Corwy's

0:08:15.480 --> 0:08:18.080
<v Speaker 1>data center expansion is dependent on what is primarily a

0:08:18.080 --> 0:08:21.160
<v Speaker 1>bitcoin mining company called Core Scientific. It appears to have

0:08:21.280 --> 0:08:26.200
<v Speaker 1>no AI capacity building, and they're meant to build over

0:08:26.240 --> 0:08:28.679
<v Speaker 1>a gigawak capacity a time where they do not appear

0:08:28.720 --> 0:08:32.800
<v Speaker 1>to have done so. And as a reminder, converting cryptocurrency

0:08:32.840 --> 0:08:36.120
<v Speaker 1>mining data centers to HPC data centers is effectively starting

0:08:36.120 --> 0:08:40.480
<v Speaker 1>from scratch. It's not good. It's not good at all.

0:08:40.679 --> 0:08:43.000
<v Speaker 1>Corby should have been a positive signal for generative AI,

0:08:43.120 --> 0:08:46.200
<v Speaker 1>or at least the way for generative AI to kind

0:08:46.200 --> 0:08:49.160
<v Speaker 1>of go further than where it is today, or at

0:08:49.200 --> 0:08:51.319
<v Speaker 1>least a way for AI boosters to shut me up.

0:08:51.559 --> 0:08:55.040
<v Speaker 1>If GENERATIVEAI had this incredible demand, both from companies looking

0:08:55.040 --> 0:08:57.640
<v Speaker 1>to integrate it and us looking to use it, Corweve

0:08:57.679 --> 0:09:00.199
<v Speaker 1>would be making fat stacks of cash. I have a

0:09:00.240 --> 0:09:03.360
<v Speaker 1>far more diverse customer base and if I'm honest, not

0:09:03.480 --> 0:09:05.320
<v Speaker 1>have to take up more than five times its revenue

0:09:05.320 --> 0:09:07.920
<v Speaker 1>and burdens and loans with loan shark level interest rates

0:09:08.000 --> 0:09:12.079
<v Speaker 1>just to survive. But it is what it is, I guess.

0:09:12.880 --> 0:09:16.760
<v Speaker 1>And another note Jim Kramer of CNBC. He said that

0:09:16.840 --> 0:09:18.760
<v Speaker 1>Corweve was going to be one of the biggest IPOs

0:09:18.800 --> 0:09:20.880
<v Speaker 1>of the year. That should tell you about everything