0:00

Hi, this is Shirley Strawberry

0:02

from the Steve Harvey Morning

0:04

Show. Toyota has been building

0:06

a legacy of excellence for

0:09

years, from developing hybrid technology

0:11

to upping the standards of

0:13

safety and efficiency. Toyota is

0:16

always innovating, always making progress.

0:18

And with Toyota's superior lineup

0:20

of SUVs in stock at

0:23

your local Toyota dealer, You

0:25

can experience the legacy for

0:27

yourself. So check out an

0:29

adventure ready RAB 4, designed

0:32

to be the perfect mix

0:34

of style, practicality, and go

0:36

anywhere attitude. Or a test

0:39

drive a capable and affordable

0:41

Corolla Cross. With the style,

0:43

space, and available tech to

0:46

keep you cool and connected.

0:48

And both RAB 4 and

0:50

Corolla Cross are available with

0:53

all-wheel drive. Giving you the

0:55

freedom to roam. Quality. Reliability.

0:57

Reliability. That's the legacy of

0:59

Toyota. Visit biotoyota.com, the official

1:02

website for deals to find

1:04

out more. Toyota, let's go

1:07

places. Cuzo Media.

1:09

Bird Up! I'm

1:11

Ed Zetron and

1:13

this is your

1:15

weekly bat or

1:17

offline monologue. So

1:20

last week I put out a newsletter

1:22

about CoreWeave, an AI cloud provider that

1:24

sells GPU compute to AI companies, looking

1:26

to run or train their models, and

1:29

they recently filed the paperwork to go

1:31

public. Now the newsletter explained, the shaky

1:33

fundamentals of the business, which is putting

1:36

it mildly, and asked the question, how

1:38

the fuck does Corweave? This is a

1:40

company that's drowning in debt, and reliant

1:42

on one company for more than half

1:45

of its revenue, and said company, Microsoft,

1:47

is actively pulling away from building new

1:49

data center capacity, and has reportedly pulled

1:52

back from some contracts with Corweave, though

1:54

it's not clear what those contracts are.

1:56

Now this is an important story and

1:59

one that raises questions not just about

2:01

the viability viability. of Corweave, but the

2:03

generative AI industry at large. That said,

2:05

the subject matter is a little dense

2:08

to turn into a hulking two-part episode,

2:10

so I decided to condense it into

2:12

a shorter monologue for you. Now cool

2:15

we've had intended to go public a

2:17

week ago, they might go public in

2:19

a week from now, and they're apparently

2:21

going with like a 25 to 30

2:24

billion dollar valuation. While it's hardly a

2:26

recognizable name like say Open AI or

2:28

Microsoft or Invidia, this company is definitely

2:31

worth taking a look at. If not

2:33

for the fact that it's arguably the

2:35

first major IPO that we've seen from

2:37

the current generative AI hype bubble, and

2:40

it's undoubtedly the biggest. Moreover, Corweave is

2:42

a company that deals in the infrastructure

2:44

aspect of AI, where one would naturally

2:47

assume is where all the money really

2:49

is. Putting up service for hyperscalists to

2:51

run their hallucination-prone non-profitable models. If invidia,

2:53

the company that makes the GPUs regenerative

2:56

AI is selling the pickaxes for the

2:58

current gold rush, Corweave, well they own

3:00

the land, or maybe they're the shovels.

3:03

I don't know, I'm not going to

3:05

get too far into that metaphor. As

3:07

an aside, up until a few years

3:10

ago, Corweave was a different company, a

3:12

cryptocurrency mining data center company, to be

3:14

specific. Now the 2010s were turbulent for

3:16

crypto, and the company then pivoted to

3:19

providing high performance computing for third parties

3:21

that didn't want to run their own

3:23

infrastructure for things like 3D modelling. Around

3:26

the time of the launch of chair-GPT,

3:28

Corweave pivoted again, this time towards providing

3:30

compute to hyperscalers and then generative AI

3:32

interest. It gambled and it gambled successfully.

3:35

You'd imagine that such a company would

3:37

be a thriving healthy business though, right?

3:39

They, I mean, they sell the surfer

3:42

architecture. The demand is there. The growth

3:44

is there. I mean, everyone wants generative

3:46

AI. Right? It wouldn't be another thing

3:48

whether reality... Jesus Christ. Even a cursory

3:51

glance at Corwe's financial disclosure documents reveals

3:53

a business that's precarious at best. And

3:55

in my most uncharitable opinion, it's utterly

3:58

rancid. If this company was in any

4:00

other industry, it would be seen as

4:02

a big pile of shit. Except it's

4:05

one of the standard bearers of the

4:07

generative AI boom, and so exists within

4:09

its own reality distortion field. Now Corweave

4:11

makes its money by renting out GPUs

4:14

to companies looking to run or train

4:16

their models. Which means that it's incredibly

4:18

indicative of whether there's a real business

4:21

in generative AI. And you'll be shocked

4:23

to hear that there isn't. CoreWe've made

4:25

just under $2 billion in 2024 yet

4:27

somehow managed to lose $863 million. Worse

4:30

still, 77% of its revenue comes from

4:32

two companies. Microsoft, 62%, and although not

4:34

said explicitly, most likely in video, 15%

4:37

now the information is reported Project Osprey,

4:39

which is basically on video's stake and

4:41

the fact that they're filing money to

4:43

them. It's actually... Not immediately obvious what

4:46

it is that invidious paying Corwee for

4:48

though. Invidia is both invested in the

4:50

company and provides it with priority access

4:53

to its chips with Corwee getting some

4:55

products before even Amazon or Microsoft. It's

4:57

bonkers and again you think with all

4:59

this preferential accent, Corwee would be just

5:02

printing money right versus burning it by

5:04

the hundreds of millions full. I wish

5:06

I could get this much money and

5:09

just, I don't know, I would be

5:11

doing a far weirder podcast. It would

5:13

be exactly the same thing. Better offline

5:16

would exist. I would just have hundreds

5:18

of millions of dollars we'd have just

5:20

like Samuel L. Jackson coming in for

5:22

some episodes. Anyway, I don't have that

5:25

money. If you'd like to give it

5:27

to me, please, please do. Now a

5:29

few weeks ago the Financial Times reported

5:32

that Microsoft had pulled out some of

5:34

their core weave contracts, though core weave

5:36

denied the allegations. And mysteriously a week

5:38

later, Open AI announced that they had

5:41

an $11.9 billion contract with core weave

5:43

to provide compute services. Now while I

5:45

can't say for certain, can you think

5:48

of another reason Microsoft needed a whole

5:50

bunch of GPU compute other than to

5:52

host open AI's models? Hmm. Now I'm

5:54

just guessing here. One has to wonder

5:57

if it wasn't so much Open AI

5:59

sounding a new deal and more at

6:01

taking over the future compute than Microsoft

6:04

was formally handling. And yes, you're remembering

6:06

correctly, Microsoft has recently canceled more than

6:08

the gigawatt of future compute capacity. Though

6:10

from what I've heard, its Corweave contracts

6:13

future or otherwise weren't included in that

6:15

number. Anyway, for Corweave to make that

6:17

$11.9 billion in revenue from Open AI,

6:20

they're going to need compute, which will

6:22

cost them tens of billions of dollars

6:24

to build. Which will be difficult as

6:27

the company is burdened with about $8

6:29

billion of debt with horrifying interest rates,

6:31

which mainly did to pay upwards of

6:33

$1.5 billion a year in loan payments,

6:36

with one of them requiring Corweave to

6:38

repay the loan with any other debt

6:40

they raise in the future. Corweave by

6:43

my estimates needs at least $30 billion

6:45

to expand. It has about $1.3 billion

6:47

in the bank and just under $4

6:49

billion left on its loans that they

6:52

can draw. One of them is a

6:54

term loan where you can draw more

6:56

up until about June 2025 June 2025.

6:59

Although Corweave will raise capital in its

7:01

IPO, the amount will be anywhere near

7:03

enough to meet its needs. And when

7:05

I say its needs, I mean like

7:08

the ability to service the revenue that

7:10

they need to spend more money. It

7:12

really does not make a lick of

7:15

sense when you think about it. And

7:17

you may wonder if things can get

7:19

worse and the answer is that they

7:21

can. Corweave has a planned 1.3 gigawatts

7:24

of expansion and its partner to build

7:26

it is a public company called Core

7:28

Scientific. Different Similar name, genuinely different. Here

7:31

are some facts about core scientific. They

7:33

went public in 2022 in a disastrous

7:35

SPAC merger, and then filed for bankruptcy

7:38

in the same year. Their entire business

7:40

has been focused on mining Bitcoin, which

7:42

requires specialized acic computer chips that are

7:44

entirely different, both in their construction and

7:47

their maintenance to GPUs. And they can

7:49

only really be used for mining crypto.

7:51

They're not easily repurposed for other tasks,

7:54

and indeed their data centers are not

7:56

just a plug and play thing. You

7:58

basically have to demolish them. And another

8:00

thing... about Core Scientific, they only made

8:03

$24 million in 2024 from selling AI

8:05

compute related services to their customers. That's

8:07

around 5% of its total revenue, with

8:10

the rest coming from mining and selling

8:12

crypto. Also, really dumb-ass mistake is to

8:14

make that. I said customers. I meant

8:16

customer. Core Scientific has one customer. And

8:19

that customer is its core weave. As

8:21

of right now, Corweave has approximately 360

8:23

megawatts of compute power. Somehow it intends

8:26

to build another 1.3 gigawatts worth of

8:28

compute using a partner that has never

8:30

built an AI day to center and

8:32

does not appear to have any meaningful

8:35

compute right now. This is all so

8:37

good. I love this. Let's fucking go.

8:39

This is the big AI appeal. I'm

8:42

going insane every time I read about

8:44

these god damn companies. But let's summarize.

8:46

Corweave is burdened by interest payments that

8:49

may balloon to as much as $2

8:51

billion a year and lost $863 million

8:53

on $2 billion of revenue in 2024.

8:55

62% of that revenue is from Microsoft,

8:58

which has materially pulled back on data

9:00

center build-outs and may have dropped some

9:02

Corweave contracts, though Corweave denies this is

9:05

the case, of course. Corweave expansion, which

9:07

is critical to servicing future revenue and

9:09

growth requires it to invest tens of

9:11

tens of billion dollars. the Corweave does

9:14

not have. Corweave's data center expansion is

9:16

dependent on what is primarily a Bitcoin

9:18

mining company called Core Scientific that appears

9:21

to have no AI capacity building, and

9:23

they're meant to build over a gigawatt

9:25

of capacity a time where they do

9:27

not appear to have done so. And

9:30

as a reminder, converting cryptocurrency mining data

9:32

sent as to HPC data centers is

9:34

effectively starting from scratch. It's not good.

9:37

It's not good at all. Corby should

9:39

have been a positive signal for generative

9:41

AI, or at least the way for

9:43

generative AI, to kind of go further

9:46

than where it is today, or at

9:48

least a way for AI boosters to

9:50

shut me up. If generative AI had

9:53

this incredible demand, both from companies looking

9:55

to integrate it and users looking to

9:57

use it. Corweve would be making fat

10:00

stacks of cash and have a far

10:02

more diverse customer base. And if I'm

10:04

honest, not have to take up more

10:06

than five times its revenue and burdens

10:09

from loans with loan shark level interest

10:11

rates just to survive. But it is

10:13

what it is, I guess. And another

10:16

note, Jim Kramer of CNBC, he said

10:18

that Corweve was going to be one

10:20

of the biggest IPOs of the year.

10:22

That should tell you about everything. show

10:34

that Toyota Tundra and Tacoma

10:36

are designed to outlast and

10:38

outlive back by Toyota's legendary

10:40

reputation for a liability. Get

10:42

into Tundra with available I-force

10:45

Max hybrid engine delivering exceptional

10:47

torque and towing capacity or

10:49

check out a Tacoma with

10:51

available off-road features like crawl control.

10:53

It can take you beyond the

10:55

trails Toyota trucks are built to

10:58

last year after year, mile after

11:00

mile so don't wait. Get yours

11:02

today. Visit Bayota. Residents

11:05

at Brightview Senior Living communities

11:07

enjoy enhanced possibilities, independence and

11:09

choice. Brightview Dulles Corner in

11:11

Herndon and Brightview, Great Falls,

11:14

offer vibrant senior independent living,

11:16

assisted living, and memory care

11:18

services through various daily programs

11:20

and cultural events. Chef prepared

11:22

meals, safety and security, transportation,

11:24

resorts to alimentities, and high

11:27

quality care. Everything you need

11:29

is here. Discover more at

11:31

Brightview Senior Living. Com. Equal

11:33

Housing Opportunity. coast.