Coreweave: AI Bubble Poster Child Or The Next Tech Giant? — With Michael Intrator and Brian Venturo

Channel: Alex Kantrowitz

Published at: 2026-01-07

YouTube video id: m1uh7Ka6868

Source: https://www.youtube.com/watch?v=m1uh7Ka6868

Is AI a bubble or the biggest boom of
our lifetimes? The fate [music] of one
company, Coreweave, may tell us
everything we need to know. We'll be
back with the company's founders right
after this. Welcome to Big Technology
Podcast, a show for coolheaded and
nuanced conversation of the tech world
and [music] beyond. We have a great show
for you today because in studio with us
are the founders of Cororeweave.
Cororeweave CEO Michael Intrader is here
with us. Michael, welcome.
>> Thank you very much. Great to be here.
and Corweave's uh chief strategy officer
Brian Venturo [music] is also here.
Brian, great to see you.
>> You are uh you both are running um one
of the most fascinating companies in the
AI boom. Everyone has used you
effectively as a Roshark test to read in
their beliefs or insecurities about uh
what's going to happen in this AI
moment. Some people think that you're
the poster child for the AI bubble.
Others think that you're perfectly
positioned uh to take advantage of the
boom in building that is occurring as
demand uh goes through the roof. A
couple stats about you. As of today, uh
the company is worth $42 billion uh
after an IPO earlier this year. You've
built uh eight new data centers across
the US in the third quarter alone. And
uh the latest reported numbers have you
uh in possession of something like
250,000 of Nvidia's uh GPUs, which are
the chips that companies use to run uh
AI models and grow them or train them uh
as they like to say. Let's just start
off with this because it's been heck of
a ride for you over the past couple
years. What has it been like being on
the front lines of this AI buildout?
talk a little bit help people feel it
the speed at which it's boomed and what
it's taken to do something like build
eight data centers in a quarter.
>> It's exhausting.
>> All right. So, let's start with that.
It's been exhausting.
>> Yeah. I it it's um
you know, you hit it dead on, right?
Like it it has been um incredibly
exciting. Um it has been an unbelievable
year. I mean, we just we just IPOed uh
um really eight months ago, and it feels
like it's been two lifetimes. Um the
company is uh moving at incredible
speed. Um we are um um building uh a
massive uh percentage of um uh the
global AI infrastructure that's required
to allow artificial intelligence to be
what it is. And when I say massive,
it's, you know, like a meaningful
percentage. Um
>> what's your estimate about the
percentage?
>> That's tough. Um you know look um
>> a lot
>> a lot is you know we we don't
>> we think of we think of ourselves as
providing enough of the compute that um
that we have the ability to be relevant
in the debate of how AI is going to be
built and how it's going to run into the
future. And so um we don't know what the
numbers are. you know, it's there's
there's lots of different providers of
technology. They're being used and
there's no real good way to kind of put
your fingers on the data, but you know,
meaningful, right? And that that's an
exciting place to be. Um, and it's
honestly, I mean, we talk about this in
the company all the time. It it it's a
privilege to come into work and focus
your energy, your your creativity um uh
uh every day on
building a component of uh this this uh
um uh of artificial intelligence which
is the issue of our time in many ways.
And we get to really sit there every day
and pit ourselves against those issues.
Um which is great. I mean, I have a ball
with it.
>> I'm taking a shot of this. Hold on.
Before we move on. Um, I I think that
that's really around, let's call it, the
practical side of it, right? And when
you're a company growing as fast as we
have, where we had maybe 100 employees
three years ago, now we have 2500
employees or so. Um, there's an
emotional side of this too, right? And
you know, sometimes since the IPO, we've
been under this spotlight in the world
of like what are they doing? How are
they doing it? Are they executing or do
are they doing this? And you know
internally we always set the highest bar
for how how fast can we do something how
high of a quality can we do it at. Um
and you know as this industry has
expanded so rapidly like there are
things that happen right and you know
you have weather that impacts
construction on a project. You have a
truck that hits a bridge. Like you have
all of these random exogenous or
idiosyncratic things that happen in a
supply chain and then it comes back to
us and it's like the world is like wow
you failed right and inside the company
from a culture perspective it's been so
important for us to manage like listen
we're doing something at a scale no one
no one's ever done before at a speed no
one's ever seen before. Of course,
things are going to go wrong, but take
perspective. Like, see how much we've
done, right? And for our employees, it's
if you're moving at a million miles an
hour and you hit a speed bump, it's
okay, right? It doesn't change the
trajectory of what you're doing. It just
like it just provides the battle scar so
it doesn't happen next time.
>> Yeah. Now, I can imagine it's a rough
and tumble uh world trying to build this
with very demanding customers, very
important technology that you're
deploying and the speed is crazy. I mean
it is interesting looking at your
founding story you really started uh
working on providing infrastructure for
crypto uh was it like Ethereum mining or
something like that and then um pivoted
in a very smart way to uh this AI moment
uh establishing a relationship with
Nvidia that's we'll talk about that
that's proven uh to be to be very useful
and helpful for you and probably for
Nvidia as well um and and now you're
again hyperdrive uh building uh data
centers and the data centers are if I
have it right um largely licensed uh or
or the capacity is rented out mostly the
tech giants I mean the core customer is
Microsoft something like twothirds of
the demand according to your your public
filings uh is Microsoft but there are
others as well
>> so so uh we actually uh spoke to um to
company uh a customer concentration in
our last earnings so we can kind of
there's no uh customer that represents
more than 30% of our backlog. And so
we've done an incredible job. It's been
a focus of the company uh um everything
from sales all the way through the the
build cycle to really begin to to uh
broaden uh the reach with which our
solution touches artificial
intelligence. So, Microsoft is an
important customer um uh and a large
creditworthy um and formidable uh part
of the AI uh ecosystem at large, but
they are um you know, we we've done a
really good job bringing on other
wonderful clients, wonderful customers
um that are going to continue to kind of
uh use our our solution as they as they
build their products and deliver them to
market.
>> Okay. And I definitely want to get into
customer concentration in a little bit.
So, um, but that's a good, uh, preface
to what we'll touch on and already some
new new data to me. So, good to hear
that. Um, but I wanted to again like
just get into what it what it takes to
build these things uh, these data
centers. Um, you're you're assembling
them with incredible speed. U, so I just
want to hear a little bit about like on
the ground uh, what does it take to put
together uh, these data centers? So the
um historically
um
you know let's say two years ago we were
able to go out and buy capacity or lease
capacity that was much further through
the development cycle right they were
basically the shell already existed. It
was a fit out construction process which
means going in and installing like the
last last pieces of the cooling
infrastructure cabinets conveyance for
all the cabling all the hundreds of
miles of cabling we have in these
things. Um, but it's shifted over the
past year is that now we're doing much
more uh bespoke in-house design, right?
To make sure that we're meeting the
needs of what our customers deployment
is going to be, right? So, it's
everything now from okay, how is the
cooling and electrical distribution
designed? Um, how are we ensuring
electrical redundancy and reliability?
Uh, you know, how are we cooling the air
cooled side of these things? Because you
have liquid cooling, but there's still a
component of it that has to be cooled
with air.
>> Can we pause on that? Sure.
>> These chips run extremely hot, right?
extremely hot
>> cooling. People talk about cooling for
those people who are coming to this for
the first time. Being able to run an AI
den data center, you got to be able to
cool the chips if you want to be able to
be successful.
>> So long term.
>> This is one of the things that um I
think the market misunderstands, right?
Is that everybody believes that this
that there's some differentiation in the
plumbing of the liquid cool data center,
right? That's not where the
differentiation lies. It's all the same
uh pipe and valves and fittings like
everyone's using the same things there.
The differentiation comes after you turn
it on and how you control those systems.
>> Okay.
>> Right. And that's what we've done
incredibly well as a company that we've
very consciously not spoken about
externally for the past couple years
because it is our secret sauce is how we
provision, validate, and manage those
data centers all the way from the power
cooling infrastructure up through the
GPUs, the servers. And it's why the like
the most valuable companies in the
world, the biggest AI labs actually use
us to run their most critical training
jobs,
>> right? I mean, it's a herculean task,
right?
>> It's important to understand that when
when you're when you're thinking about
the ecosystem, right? Um and you're
thinking about the different Neoclouds
that that that populate the
>> What's a neoc? So,
>> the worst term ever. [laughter] I hate
it.
uh think of it as like um you know in
the uh the common vernacular you know
everybody knows who AWS is you know
Amazon they know who Microsoft is they
know who Google is those are the
hyperscalers right um you can throw
Oracle in there if you'd like uh but
then there's a uh a class of uh
providers that can deliver this
infrastructure um and you know we are
the leader among that and what is
important to understand that if you took
all of the other uh Neoclouds and added
their GPU fleets up, we would still be a
multiple of all of them combined in
terms of the number of GPUs that are up
and running and delivered to clients.
And so large multiple
>> um what what when when when when Brian
is talking about um you know um you know
things that the market is struggling to
understand it it it is it's important to
understand that what differentiates us
what allows us to be as successful as
that we have is that the software suite
that we have built allows us to take the
commodity GPU and deliver a
decommoditized premium service that
allows people to extract as much value
from this infrastructure as possibly can
be extracted and that's really what
Coreweee is doing and it's why when when
Brian says hey you know the leading
companies in the world and the leading
uh uh labs in the world are relying upon
us to deliver our service that is why
it's because the the product that
ultimately um they receive is the
product that will allow them the
greatest probability of being successful
at using the GPUs to deliver the the
products that their company is building,
>> right? So, just to put it in plain plain
English, always helpful for me. When a
company uh like a Microsoft will work
with you on on building infrastructure
for artificial intelligence, you've
built uh some proprietary uh pieces of
the puzzle, like your cooling system,
like the software that runs the data
center, and that allows them to get more
out of the chips than they would have
typically.
>> Yeah. And the the the
nuance here is that when you build one
of these data centers and it has 3,000
miles of fiber optic uh cabling and it
has a million uh optics that connect
into the switches, like these things all
fail, right? And when they fail, um the
way that training jobs are run today is
if one component fails or one component
limits the performance, the balance of
the training run is going to be governed
by the worst performing component. Oh,
>> right. And our entire job is to build
the automation, the predictive
analytics, the you know the machine
learning models around saying okay we're
seeing a problem here. How do we
gracefully handle these things so it has
the least impact on our customers jobs
right and that's the core we've secret
sauce
>> okay
>> is that we have the world's largest data
set of how these things run how they
fail and we've built all the recovery
mechanisms and the software intelligence
to help our customers run these things.
is the demand that you're getting from
your customers uh you mentioned you know
training uh very well u is it mostly
training the AI models uh because well
that's that's what a lot of the
infrastructure has been used for build
scaling these models throwing more uh
compute at them throwing more data uh
making the models bigger and and then
the idea is that the models get better
so are you seeing most of your demand in
the training side of things or has it
gone to inference where like companies
are actually using the models uh and
deploying them into production.
>> It's a great question and I think it it
talks to the split or this kind of
delineation of where the market's been
for the last three years and where it's
going. Um you know our customer base for
the last three years has primarily been
the largest AI labs and enterprises that
are building the capabilities of AI,
right? And it's now shifted from the
people building those capabilities to
the people that want to use those
capabilities to change business
outcomes. And this is where all the
enterprise adoption is coming from. Um,
you know, it's, uh, one of my favorite
services out there is Lovable, right?
You go to Lovable, you can build any app
you want. There's a chatbot that helps
you go through it. Um, you know, we're
finally starting to see people chain
together these capabilities to build
real products that solve problems. And
our business for the last three years
has really been around the the creation
of those capabilities and has very
quickly shifted to include not just the
creation of them but the deployment of
them and use in business practices. All
right. So um one of the things that I
didn't expect was that uh what looked
like training two years ago is how
inference was going to look today.
Right? Is that you're still dependent
upon uh highly connected storage. um you
know your backend networks become
critical to this because the models are
so large. So there's really no
difference between training
infrastructure we deploy to build those
capabilities and what our customers are
ultimately using to serve them.
>> So has has inference overtaken training
for you?
>> Uh we serve a tremendous amount of
inference. Um but you know
>> I actually don't know the answer to
that. I
>> 6 months ago I would have said it was
2/3 training and one third inference.
It's probably close to 50/50 now.
>> Okay. Um, but there's also some of our
big customers that they go from they'll
use a campus for training, they'll
launch a new product, they'll have to
spill over for inference. Um, you know,
a lot of this is very dynamic and it's
been built to be so.
>> Yeah. I I uh this may provide a segue to
some of the other subjects that that
you'll ultimately get to in the in the
in this podcast, but you know, for me,
watching inference,
understanding that inference is the
monetization of the investment in
artificial intelligence is one of the
most exciting trends that exists within
AI. And we have a front row seat across
the entire cross-section of uh almost
every uh um large important lab that's
building this stuff and watching them
increasingly you know move from let's
say uh you know one-third inference
climbing towards uh you know 50% and at
times it's even over 50% of uh the fleet
being used for inference you know is is
just an amazing uh um uh indic
indication of the scale of the demand to
use artificial intelligence to serve
customer inquiry
>> and that means everything.
>> All right, one more question about this.
>> Y
>> why does why does Coree need to exist?
Why I mean we're talking about these big
companies like Microsoft like why
wouldn't they just build their own data
centers? Why are they licensing it from
a third party?
>> So it's a great question. Um
there was a void in this market, right?
And there's a couple pieces here. Um,
the biggest clouds in the world today
are built off the cash engines of
peripheral businesses, right? Google's
built on search. Amazon's built on
retail. Microsoft was Microsoft was
built on enterprise software. Uh, we
came pretty much out of nowhere, right?
And our the the moment in time for us to
be able to get ourselves into this
position was driven by crypto, right?
You mentioned earlier that we came out
of, you know, Ethereum mining. um we
were able to leverage the revenue from
Ethereum mining to go out and build and
deploy additional scale so that when
crypto went away we had the
infrastructure in place and we hopefully
had enough clients that we became like
we were at escape velocity right so um
you know we recognized that compute was
going to be valuable we didn't
necessarily know at the time what it was
going to be valuable for like I don't
think Mike and I ever had this idea of
like there's going to be this hundreds
of billions of dollars a year in capex
for AI high. But um you know we had the
thesis that compute is uh going to be
incredibly valuable and we wanted to own
a lot of it and we looked at that
compute resource as an option like and
we said okay what are the best things
that we can do with this and that's how
we've always approached different
business problems right is like what is
our asset how do we monetize it the most
effectively what's the most valuable way
to use this
>> but then so so so I'm gonna jump in here
on this but I want to go back to
something that we kind of talked through
as we started this right is that Like
we've built a software stack from the
ground up to optimize for the use cases
associated with parallelized computing.
We do it better than anyone else. The
reason we exist is because we deliver a
fantastic product that is highly in
demand
>> and incredibly differentiated
>> and incredibly differentiated. And so um
you know we we serve the largest players
but we also serve you know a ton of
other uh AI uh companies that are
building applications where they have
the choice to go and use us or to go and
use uh one of the hyperscalers and many
many many of them uh choose to use our
solution because it allows them to more
effectively deliver compute and one of
the things that's really just lost on
this is that there's not an
understanding of how fundamental the
change from cloud 1.0 into cloud 2.0 as
you moved from you know uh sequential
computing into parallelized computing.
And when you when you made that leap
right from you know um you know hosting
websites and and and data links into
driving parallelized computing for
artificial intelligence. It stands to
reason that a fundamental change in how
compute is used will also require a
fundamental change in how you build the
cloud to serve it. And we took advantage
of that transition to build
best-in-class solutions, right?
>> And that's why we exist.
>> So, uh I've heard an argument made that
basically the big tech companies um you
know to build these uh these data
centers, they have to forecast demand
out years in advance. It's a massive
capital commitment. They're not sure
whether it will pay off and coreweave is
useful to them because you're taking the
risk and then they will be able to use
your capacity and sort of rent it out as
opposed to having to make these big
investments on their own and you know
it's it's their ass is if things go
wrong.
>> Yeah. Look uh you know that that is a uh
that is a narrative. Um I don't think
that actually tracks with the reality of
the situation. I think the reality of
the situation is is the large
hyperscalers are building as fast as
they can. Uh Google went out and just,
you know, released a press release where
they're building $50 billion worth of
infrastructure while they're still
buying from everyone else they can. Uh
Microsoft is building internally and
they're buying from uh from from lots of
other players. What what
I I feel like that that argument is
model fitting, right? it is somebody's
got a preconceived notion of what this
is going to look like and now they're
reconstructing the factual the facts on
the on on the ground to fit that model
so that they can say look I'm right but
the reality is is that um I look at it
very differently right I look at the way
that we built our competitive uh
advantage over you know the hyperscalers
the way that we built our competitive
advantage over other uh neoclouds and
the way that we did that is we
understood that this type of computing
was going to be important and we built
the infrastructure and the software to
be able to serve it when the demand
emerged and we did it in a very
riskmanaged way. When I look at the
future, when I think about uh um the the
the investments that go into building a
an AI factory and I think about how much
money is being put into the data center
versus how much money is being put into
the compute that goes inside of the data
center, I think about the data centers
as being basically an option on being
able to provide and be relevant for the
delivery of compute into the future.
Right? We take our risk dollars as a
company and we invest in the long poles
and the long poles are really twofold.
One is building the best software in the
world and the second one is having
access to the data center capacity to be
able to deliver compute when a wave of
demand hits this market that requires
you to deliver it. You can't just wake
up and say, "Hey, I want to deliver a
gigawatt worth of uh infrastructure."
What you have to do is you have to start
years in advance building that gigawatt
of infrastructure so that you're in a
position that when your customers say,
"Hey, I just produced a new way of using
AI that's going to require a gigawatt
worth of infrastructure." You're able to
serve it. We're going to have a
tremendous portfolio of infrastructure
that is going to be able to be deployed
into the future and we're really excited
about that. We think it's a wonderful
way to go about building our business,
>> right? And and that's the question about
the bet, right? is that um you're
betting that AI is going to continue to
be adopted at a wild rate.
>> That's not entirely accurate. Okay.
>> What what we are doing is we are making
the majority of our uh investments by
taking
long-term contracts from creditw worthy
entities using those contracts as a way
of raising money to build the
infrastructure where the demand and the
credit and the capital has already been
uh um secured. Right? So let's say 85%
of our exposure is to deliver compute to
investment grade or AI labs or other
large consumers of compute. Right? The
other 15% is our exposure to long-term
contracts to be able to do that exact
thing in the future.
>> And that's the way I look at it. And I
think it's a much better way to think
about how we're taking on risk, how
we're dealing with leverage, and how
we're positioning ourselves. If the
market continues to grow, we're in a
great position. If the market stabilizes
in and around this, we're fine. If the
market contracts, there's some new
technology, then we will be left with
some portion of that 15%
that we may be uh in a position where it
has to wait for a few years before the
market grows back into it. And we are
fine with that. We think of it from and
and you know people have talked about
how the founders of this company kind of
look at the world with a different lens
because we don't come from Silicon
Valley. You know, we come from the
commodity space. We come from Wall
Street. We think about option value,
right? When when we think about compute,
we think about what is the option value
associated with it. When we think about
the data centers, we think about what is
the option value to be able to build to
be relevant in the future. And that's
the way we kind of go about allocating
our risks and securing the contracts
that we have in place right now.
>> Yeah. And and [clears throat] you know
to speak to one thing here you talked
about if the market contracts um I think
that we would love for that because it
presents tremendous opportunity for us.
>> How right I mean you you're in a
position where there's going to be
distressed assets. There's going to be
consolidation uh possibilities like
that's when opportunity really comes in
and you know there's a lot of times
where we sit there and say okay we're
looking for M&A we're looking to invest
in things but the valuations don't make
sense. And for Mike and I, you know,
we've made our careers on waiting for
those opportunities and saying, "Okay,
these are the things that I want to buy
when things don't necessarily go right
for them, right? And uh you know, that's
really what excites us. You know, one of
our uh one of our other founders last
week, he got on the phone with me. He's
like, "I love this, Brian." I'm like,
"What, Brian?" He's like, "This is the
one where you start like this. You're so
focused on like where are the
opportunities? How do I go take things
over?" Um, and you know, it's I say it
to some people every once in a while is
that I feel like when uh there's
headwinds in the market, it's actually
easier to do this job.
>> Right.
>> Right. Than when the tailwinds are kind
of blowing at 1,000 miles an hour.
>> But can I ask how have you set up the
company to make sure that you're not the
distressed asset when the contract if
the
>> look at our look at our construction of
uh customer of our customer contract
portfolio, right? is everybody last year
talked about how customer concentration
and exposure to Microsoft was a bad
thing, but they have a better balance
sheet than the US government, right?
Like I'm not worried about them
performing in their long-term
obligations to us. Like that's basically
the best possible position we can be in.
And we've been super thoughtful about
the way that we choose which customers
to work with and how we manage the
credit exposure so that we're like we're
certain that the investments we make
will be paid back. And if you look at
the people that are providing us the the
debt to do those projects like
Blackstone, right, they're the some of
the most sophisticated people in the
world. And for their underwriting
committee uh committees to come in and
say, "Yes, I want to do this and I want
to scale it up as aggressively as
possible." Like, you're telling me
you're going to pit some financial
analyst against John Gray? I'm going to
go with John Gray.
>> Yeah. Well, well, well, you know, I
mean, may maybe a second on just like
kind of one of the fundamental building
blocks of how we um have expanded the
way we have and how we use debt because
I think that's one of the misunderstood
uh uh components of how you build or how
we have built this company. Um and so it
is really important to understand that
we the the way that we build the
components is we go into the market.
Let's use Microsoft because we've used
them, but there's lots of other clients
you could use and they're totally
interchangeable. Um, from from the
perspective of the the structure uh is
still the same. We go to them and we
say, "Hey, um, you know, we've got we've
got access to to to this data center.
Um, they say we need compute." We say,
"Okay, we're going to sign a contract."
They sign a contract for five years. We
structure that contract um in a way that
we can go back out to the Blackstones of
the world and we can borrow money from
them to go ahead and build the
infrastructure to deliver to Microsoft
within the 5 years of the contracted
period with Microsoft. We pay for the
infrastructure, we pay for the opex, we
pay for the uh um uh interest and we
earn an enormous margin on the
infrastructure.
So yes, there is debt. We're not arguing
that. We believe fundamentally when you
build any type of uh infrastructure at
this scale, debt is the correct way to
go about doing it. The examples run
through history. Whether you're talking
about building a power plant, building a
uh um uh distribution grid for
electricity, whether you're talking
about the telephone, whether you're
talking about the steam engine and
railroads, like you go throughout
history, this is the tool that you use,
right? We didn't invent anything new
here. We just took a uh a tried andrue
method and applied it to the specifics
of depreciation associated with this
asset of the obsolescence curve
associated with this asset and made the
contours so that it worked in an
airtight manner so that guys like John
Gray or you know Blackstone or any or
Black Rockck or any of the big lenders
could look at it and say I understand
how they're going to underwrite this. I
understand the risk in this. I
understand that these guys are going to
deliver compute to that balance sheet.
they're going to get paid back and when
they get paid back, we're going to get
paid back. So, let's lend them the
money.
>> And that's lost on the market. They
think we're running around with this
like um you know, incredible capacity to
take on risk. But that's a really
lowrisk approach. Matter of fact, it's
way more lowrisisk than saying, "Hey,
we're going to do it on equity because
we're saving our equity for the long
poles that you've got to invest in."
That's where you want to put your
bullets. You want to use the debt
markets to deal with a depreciating
asset. It's the way it's done. It's the
way it's been done throughout history.
>> Yeah. By the way, it's great that we're
able to have this conversation. This is
what we want to do on the show is take
this complex stuff,
>> talk about what the reactions have been
in public, speak with the principles,
and actually get the story. So, thank
you for talking it through with me. And
on that note, let's continue. Um the the
argument I think that would be made uh
is not that Microsoft isn't good for the
money. The argument would be made that
generative AI is still a developing
category. It hasn't really shown the
ability to turn consistent profit. And
so comp the companies that are investing
in a big way in it may one day wake up
and say um you know we we can't we don't
really want to uh do that buildout. uh
Open AAI for instance, let's just use
them as an example, they have something
like 1.4 trillion uh committed to spend
on infrastructure. Uh I think Open AI
might be the only ones that believe that
they'll actually spend that 1.4 trillion
and maybe they're investors. So what do
you think about that risk that AI is
because AI is new and not as predictable
as you would have uh in a different
category you know finance by debt uh
that therefore it is riskier even if the
credit rating of a company like
Microsoft is golden. So um when when
you're
a couple things on OpenAI because
they're they you know they are um the
tip of the spear in in in many ways for
artificial intelligence. They have uh a
franchise that has um 800 million uh
monthly users of [snorts] their product
which is fully onetenth one out of every
10 human beings on the planet logs on to
open AI
>> fastest growing tech product history.
>> I use it all the time for every I am
addicted to it and and I don't even find
it in like a bad addiction way. It's an
amazing product. I won't argue with
that. So, so, so you you've got this
this this product that's out there, and
then you have this $1.4 trillion, which
I believe has been uh confirmed by
everybody, but OpenAI, who would
actually probably have issues with that
number in terms of how much they're
spending, when they're going to spend
it, um what what are options, what a
firm, all those kind of things. And so,
I just think it's a you know, narative
shaping there. There's a there's an
incredible uh amount of people out there
that are uh talking through how this is
going to be done, when it's going to be
done. Um um and I don't think that they
necessarily have all the correct
information. Um that's number one.
Number two is is that you know, you
listen to both Brian and I talk about
how we think about credit. We're pretty
sophisticated how we think about credit.
We've built our entire careers long
before we started this company uh
thinking about risk management in
credit. Open AAI will be a percentage of
our credit exposure
just like Microsoft will be a percentage
of our credit exposure. and the way that
you manage credit against a unbelievable
potential company, but a company that
may not have the credit rating that is
strong enough to support um um their
aspirations or they may have to tone it
down or they may is you just make them a
limited percentage of your overarching
business and you accept the risk on that
while you mitigate the risks using
credit from other companies. companies
um like Meta that we signed a 14 billion
contract with like Microsoft I mean just
incredible companies and so you just
think of them as how much investment
grade exposure am I going to take how
much non-investment grade exposure am I
going to take and what's the correct
ratio and how am I going to mitigate
that over time and that's the way we
look at it
>> and what happens if one of these
companies over time wants to walk away
let's say meta says yeah actually
artificial intelligence we can develop
it much more efficiently or Microsoft
says yeah AGI is actually a decade away
not three years away
>> yeah so so AGI being a decade away six
decade it doesn't matter like like the
way you know you you were you were
asking about you know how you run a
company in in this dynamic environment
how you run a company that's going
through this type of scaling and and I
talk about this internally to the
company all the time we need to be
directionally correct the world is
incredibly
uh uh fluid The world is incredibly
dynamic. We are um at the absolute
bleeding edge of a new technology that's
redefining the world. You're not going
to get everything right, but
directionally you have to go ahead and
build a company that's moving in the
correct ways to be able to take
advantage of this super cycle that's
going on. What do I think if Meta says,
"Hey, we're going to, you know, we're
not going to uh continue to invest."
That is their prerogative as a company,
but that doesn't in any way mitigate
their contractual obligation to us
through the term of the agreement that
we went to Blackstone with and said
we're going to borrow money because we
have a firm contract with Meta. That's
not open to uh uh um renegotiation. They
can't walk away like like the concept is
is is and you know there was a wave of
this that took place you know about a
year ago. Microsoft is walking away.
Like, what are you talking about? This
is a AAA company. They don't walk away
from anything. If they make a a
contractual obligation, that's a
contractual obligation. The even the
idea that they would walk away from it
is deeply misleading to the market.
>> Okay. Uh there's been some analysts that
have talked about one more thing on
debt, then we'll move on. Some analysts
that have talked about uh core weave
borrowing more money be uh because uh
they spend more money than they can get
structurally. So they borrow to pay
interest on the last loan.
>> Why don't you talk about how these DD
like these actual debt instruments are
structured from like the box perspective
and how the controls around these things
are like that'll put this to bed.
>> Yeah. So
>> like let's just be done with this.
>> There there's a there's a lot of a lot
of analysts that have a lot of opinions
[snorts]
um based on
a a deeply um incomplete understanding
of how these are built. So may maybe two
seconds on it and then Brian you can
kind of keep me on the rails here. Um
>> I'm pushing you off the as much as I can
record.
>> Once again going back to to the
contract. We did a contract with Meta,
right? When we did a contract with Meta,
we go ahead and
we sign the deal with Meta.
We go we borrow the money from a
syndicate of lenders
and then we go and we buy the
infrastructure to build that facility.
We run the facility. When we run the
facility, as we're delivering GPU
capacity to Meta, Meta sends money, but
it doesn't come to us. It goes into
what's called a box. Money flows into
the box and then it goes through a
waterfall. The first thing it does is it
pays off the opex associated with the
power and the data center. The second
after it's done paying that, the second
thing it does is it pays the interest to
the lenders. The third thing it does is
after it's paid all of the expenses is
it releases back up to our company and
>> also principal
>> and principal and interest so that it it
it completely amvertises within the
five-year term of the uh contract with
Meta.
Like there's no like it's controlled by
somebody else.
>> Yeah. And and the important piece of
this is like
>> it's not that it's hey we just barely
pay off the interest. The coverage ratio
in that box is
excellent and it can be underwritten at
a very narrow spread based on the risk
analysis of the most sophisticated
lenders in the world. Right? They're not
lending us this at 22%. They're renting
they're lending this at, you know, 250%
over uh excuse me, 250 basis points over
uh sofur, right? Which means basically
they're looking at it as like this is a
lowrisk transaction to get their money
back. It's not some crazy, you know,
you know, yolo structure. It's an
unbelievably risk mitigated structure
that's built to simply go ahead and
allow us to build the infrastructure,
deliver it, and then take the revenue.
Now, when you're scaling a company at
the rate we're scaling,
it tends to make sense that you're going
to be investing all over the place. And
we are. We're investing in data centers.
We're investing in software. We're
investing in people. We're investing in,
you know, uh uh um the the companies
that we're buying to to to help us reach
up the software stack and provide more
value. We're doing all of those things,
which is exactly what we should be doing
right now as this space opens up.
Whenever we see an opportunity, we look
at it against all the other
opportunities that are out there and say
that one makes sense for us. It drives
the company forward. the the idea that
you're at risk from
the debt.
I mean, anytime you have debt, there's
risk. Not going to I'm not going to
argue that point because you have to
generate the revenue. But what are you
talking about? You're talking about
operational risk on the GPUs that are in
the box,
>> right? You know, one of the things for
us and why our spread on that interest
rate has compressed over the last two
years is we've demonstrated incredible
capacity and capability of delivering
that infrastructure, right? The first
time we did uh uh one of these debt
syndicates, I got paraded around the
whole world and had to sit with every
single underwriter being like asking me
questions about like uh like what are
the doors to get into the data center?
Like what is the floor made out of? like
okay guys like that there had so much
there was so much risk around our
ability to operationalize it that has
been put to bed now where everyone knows
that we can do this and we can do it at
scale
>> right that our cost of capital is
significantly compressed
>> I mean it went from you know what was it
>> surp
plus 1350 down to sofur plus 400 right
once again like for those who don't
understand what that means is the higher
the the higher the interest rate the
higher the risk And what you're seeing
is the lending market understand that we
have the capacity to deliver this
infrastructure and that they are willing
to lend us money at increasingly lower
rates because they look at it as a lower
risk transaction.
>> Okay, I have uh so many more questions
and we have only uh 15 or 20 minutes
left. So uh let's take a quick break and
come back and talk about a few things uh
that I find really fascinating that is
the depreciation on uh these AI chips.
Uh maybe a little bit about the
financing structures and then power. I
think we need to talk about power. So
let's do that when we're back right
after this. And we're back here on big
technology podcast with uh the founding
team of or twothirds of the founding
team of coreweave. Uh Michael and Trader
is here. He's the core CEO and Brian
Venturo here is here. He's the core
we've coo chief strategy officer. Uh we
talked previously or in the first half
about um how these chips run hot. Um so
let's just talk a little bit about the
life cycle of these chips. I I'm trying
to figure this out. There's two
differing opinions. One is that a a GPU
like the Nvidia H100 or the GB200
will burn as hot as it possibly can for
like two or three years and then
effectively be useless like meltdown.
It's like the life cycle of a car
compressed into a couple years. Um the
other side of it is that uh no this the
GPUs can last. Uh but um they get less
valuable over time because more powerful
GPUs come out that are multiples uh in
terms of their ability to to um do AI
calculations compared to previous
generations. So can we just start with
like the basic physics of this? How long
do these things last?
>> So Oh, I'm taking this one. You're out.
>> You take the physics. I'll
>> the other side. Uh so last year is when
we saw um let's call it the the
hyperscalers that were around in the
2010s. So uh Amazon, Microsoft, and
Google finally uh retire their Nvidia
K80 fleets. And the K80 was a GPU that
was introduced in 2014. So it's it was
active in their clouds almost fully
utilized for 10 years,
right? And the number of um you know of
changes in architecture and efficiency
advancement and performance advancement
over those 10 years was massive. You
know just last week um we entered a
multi-year contract to renew Nvidia
A100s which are the GPUs that were
introduced in 2021.
Right? So we're already going beyond the
5-year contract life for GPUs that came
out, you know, four years. um that the
idea that these things burn out in 2 or
3 years like it's kind of bunk, right?
And from a physical perspective, right,
within 3 years, these things are all
still under warranty. So if they break,
they get replaced, right? But from a
like this is not they run hot. These
things are designed to run hot. Um GPUs
that we had deployed in 2019 are still
running, still have customers on them.
Um, you know, it is a
like some of it is customers that are
deploying Grace Blackwell with us today.
They're going to use Grace Blackwell for
their most frontier or bleeding edge use
cases. They're going to train their
biggest models. They're going to do the
things that they need the like the
newest
>> Nvidia's latest chip.
>> Yeah, it's Nvidia's latest chip. They're
going to do the things that they need
the most uh firepower to do and they're
going to run their inference on hoppers
or they're going to run their inference
on ampear, the A100s, right? or they're
going to run different steps of their
pipeline on A100s or they're going to
run parts of their pipeline on CPU
compute, right? There's always going to
be a use for these different levels of
compute infrastructure. It's just where
is the economic value there, right? It's
not a useful life question. It's where's
the economic value in those in that
time.
>> And this is where this is where the the
questions start to build up because um
so that the chips run. We agree on that
one. Now I've I've been um taught so
thank you. Um [laughter]
the chips
>> the chips run off the table and and so
so now the question is when it comes to
power, right?
>> Hold on, hold on. Let
>> me let me finish this question and you
can answer the last one, but I just want
to finish this one. Right. So the
question No, no, no, no, no. I want I
really do want to hear, but let me just
put this out there and then you can
answer whichever way you want. Okay. the
the old generations of of Nvidia GPUs,
they're much less powerful than the than
the newest generations. There's the
great the Grace Blackwell that's out
now. Uh there there's Ver Rubin that's
coming out. And that the argument is
that these newer chips, even if the
H100, the Hopper can continue running.
Uh the new chips are so much more
powerful that the value, right, because
those H100s are being sold at 20 $30,000
a pop. the value of those chips are
going to be much less because of the
power of of the newer generations. And
then if you think about it again, if if
these companies move from training to
inference, right? If for instance, let's
say hypothetically there's there's a
diminishing return to training a bigger
model, then those bigger those more
powerful chips can be used to run
inference. And then a company like
Cororee which has hundreds of thousands
of the older generation of chips is
faced with a depreciation problem
compared to the most powerful ones.
>> You got it.
>> So, so let's uh
>> let's go through this a couple different
ways. Okay.
>> All right.
>> Um
I feel like the depreciation narrative
is uh being uh spun up by um folks.
>> Yeah. like like people that don't
understand the space never been in a
data center.
>> So So like my my theory here is is it's
it's it's being spun up by a bunch of
folks who couldn't spell GPU two years
ago and now they are out there as
experts on how it actually works. So
let's actually go through the different
pieces of it.
The most important tool that I have for
understanding what the depreciation
curve or the obsolescence curve of
compute is is not what I think right.
It's not what you know uh some historic
short thinks. It's what are the buyers,
the most sophisticated companies in the
world willing to pay for today? And when
they come to me and they put in a
contract for a five-year deal or a
six-year deal, in what world do I not
think that they who are the consumers of
this understand that there are new, more
powerful chips coming out? Of course
they do. They understand it, but they
also understand what their various use
cases are. And they are saying to
themselves, "I'm going to buy this
because I'm going to need it today. I'm
going to need it in 3 years, and I'm
going to need it in 5 years. And what
the use is within my system will change,
but it didn't become useless. It hasn't
become obsolete, right? and they know
the new stuff's coming, yet they're
still buying it because they know better
than someone who doesn't know anything
about how compute is used. My
opinions around depreciation are
informed by the only entities that get
to vote in my world, which are the folks
that are paying for the compute over
time. Those are the guys that get to
vote. Everybody else is just looking in
and guessing, right? That's number one.
Number two is Brian kind of made a point
that we just had somebody come back and
recontract for term for a term deal the
H100s.
>> A100s.
>> No, at H100s at 95%
of the value
of what they were originally sold for.
Once again, not showing this
catastrophic depreciation curve that you
know has been voiced out there. I just
once again like For me, it's about the
data because I need to make the decision
to buy this infrastructure or not to buy
this infrastructure. And so, I've got to
kind of look through the noise and
decide,
you know, are the big hyperscalers,
are the big labs, are the big buyers of
this infrastructure who are looking at
this saying, "This stuff will be useful
for us for the next 5 years. Let's go
out and buy it." or should I go and turn
to somebody who's never really
understood how the cloud works, what a
GPU is, what are the different uses as
it moves through from the most cutting
edge models to other uses within the
training as they go all the way down
through inference to simpler, smaller
models. And I think that's the way you
got to look at this thing is like what
are you talking about, man? if if if
Microsoft and Meta and the other big
buyers are coming in and buying for five
and six years and I I don't really think
that anybody else really should or gets
to have what I would consider to be a an
informed opinion on depreciation. And
since I'm selling on term contracts
specifically to insulate my company from
the depreciation curve, right, I know
how much I'm going to make because I've
sold it to Meta for 5 years, every hour
of every day. And they're going to pay
for it every hour of every day. The what
what the curve looks like inside of that
five years, that's already been priced
into the deal I did with them.
>> Sorry. Go ahead.
>> Sorry. Well, I was trying to interrupt
you there because I think that the in
addition to the H100s which came out in
2023,
>> right? We signed a term contract for the
A100s at like within like the 95% of it
original price range for on like on term
last week or two weeks ago. Like that's
crazy.
>> Those GPUs are already 5 years old
>> and they're they're like that useful
life is there.
>> Yeah.
>> And everyone is saying, "Oh, it's not
useful." Like they have no idea. They
don't actually have the data. We're
sitting on all this data. We talked to
every single one of these customers and
you know one of the interesting things
that's happened over the past years
everyone was saying well where are all
the enterprises last year and the
enterprises weren't there because every
AI lab in the planet was like was in a
food fight for capacity and the
enterprises couldn't fight their way in
right and now as we're finally getting
enough supply to make it available to
many people like the ground swell of
enterprises that are coming in to use
this stuff is overwhelming right to the
point that we're st we're still choosing
what customers we want to work Right.
>> Right. We are like this is a supply
constrained environment, right? And the
supply constraint keeps getting tighter
and tighter and tighter, right, for
these customers.
>> Okay. I have two more questions.
Hopefully, we have time to get through
both of them.
>> Let's do it.
>> Um, we got to talk briefly about this
circular financing question. Uh, just to
set it up. Um, Nvidia owns 5% or so of
Core Wee. According to reports, it has
agreed to spend 1.3 billion uh over four
years to rent its own chips from
coreweave according to reports. Uh and
you also buy uh the the GPUs from
Nvidia. So can can you talk a little bit
about like is this is this too tight of
a relationship? Is this like sort of
demand, you know, sort of propping up
supply which is propping up demand?
>> Nvidia has made two investments in
Cororeef. uh they made an investment of
$und00 million and then they made an
investment of uh and that was early that
was in the the B round I believe um
>> at a $2 billion valuation.
>> Yeah. And then they made an investment
of $250 million at IPO. Um Cororeweave
has raised uh $25 billion
to build and scale its business. I'm
pretty sure that they don't think of
their investment of $300 million as the
secret sauce to standing up the largest
company in the world. It's just a
ridiculous narrative. Um so look, um
the reality is is you've got a
systemically
imbalanced market, right? There are not
enough GPUs out there to go ahead and uh
support the demand for compute for
artificial intelligence. And when you
have such a uh disequilibrium
uh in a market, it is not unusual to see
companies working together to try to uh
um
>> align interest
>> align interests and drive compute
buildout or any other industry as fast
as possible.
Um Nvidia has been a wonderful partner
of ours. Um and uh they have um
uh entered into uh a relationship with
us which is great. They've entered into
uh and invested in other companies which
is great. They're trying to uh invest in
the ecosystem uh and cultivate the
buildout of what they considered to be,
you know, a a generational change in the
way the world is going to work. And I
agree with them. Um, but do I think it
is circular financing to invest a
hundred billion dollars hoping that
we're going to then go ahead and spend
billions and billions of dollars? It
doesn't make any sense,
>> right? Um, you know
what their strategy is. I I don't I
don't think it's really prudent for me
to kind of guess at what Nvidia is
doing. I I think of it differently. I
think of it as um there is a
relationship that exists between us and
Nvidia. We provide um the most
performant configuration possible of
their infrastructure and deliver it to
the consumers of computational power and
they appreciate that. They build
incredible infrastructure that allows us
to build our business and we appreciate
that. Um and you know the the the it's
sort of like you're being distracted by
a fly on the butt of the elephant
>> and you know that's what this is. You're
talking about a very dimminimous
uh sum of money that was invested. I
mean, it's a lot of money, but not in
the scope of what we're talking about
here. It's a dimminimous sum of money
from the perspective of, you know, the
the company is worth $40 billion. It was
just a good investment. They looked at
what we did and they said, "These guys
rock. We're going to invest in them."
>> Y
>> right. So, you know, once again,
depreciation is one of the narratives
that you hear continuously. Circular
financing is one of the narratives you
hear and bubble is one of the narratives
you hear all the the the the other way
of looking at it is just the largest
companies in the world can't get enough
computing they're desperate to get their
hands on it so that they can serve their
clients because it is profitable for
their business and that seems to have a
lot more uh there there to me
>> right I I know we're running out of time
can I just ask the power question and
then we can head out uh Sat Nadella was
on a podcast recently and said he has
more chips than he can plug plug in
because uh the power is basically uh the
constraining factor for him. Y
>> u there's been so much I mean we talked
about you guys building eight data
centers in the most recent quarter. Uh
so much build out people are talking
about how it's going to maybe even raise
consumer prices for energy. Um is power
the limiting factor for the continued
ability to build out uh AI
infrastructure? So the um the constraint
moves right and right now um I don't
think that power itself meaning grid
connections and the generation capacity
is the limiting factor. Right now it's
the construction and trades. So it's
human labor and supply chain that are
the limiting factor is that you went
from a market that was building maybe
one gigawatt of data center capacity a
year to a market that's building 10
gawatts of data center capacity a year
and all the trade unions like they don't
scale the same way, right? And you're in
a position where you may have had a
labor force of a thousand people
building a data center and 200 of them
were experienced tradesmen that had
apprentices and now you have a thousand
people and 20 of them are experienced
tradesmen and everyone else is kind of
apprentices and you know that stretches
the supply and construction uh you know
supply chain very very thin. So I think
you're running into just this like
temporary transient problem of projects
are taking longer than people thought
they would, right? Uh that's the big
blocker is that you walk in and you say,
"Okay, I have a data center being turned
on next week." And during your
energization process, something goes
wrong and okay, now you're set back by
40 days, right? There's hiccups that are
happening along the way because things
have gotten so stretched because demand
has been so insane and has been
increasing at a step function every six
months for the last 3 years. But the
power is there. The power is there
today.
>> Oh yeah, that's the key word, right?
It's like it's it's this is a you got to
think about this stuff over time, right?
Um what will happen is data centers will
be built power within the grid will be
consumed and as that power gets consumed
there will need to be new power brought
online in order to provide for the
future growth as well as all the other
uses for power that are are required and
growing you know independent of
artificial intelligence. Um and that
will be a challenge for the US grid over
time. it will be the challenge for you
for grids all around the world. Um but
at the moment it's you know what what
what is what is the problem that you're
facing today? What is the problem you're
going to be facing in three years? In
three years power is going to be an
issue
>> today. It's it's it's it's power chips.
>> But power is going to be an issue until
somebody like some college kid at
Stanford's going to come up with a
better way to run this in the software
side and they're going to generate crazy
efficiencies like we saw that with
DeepSec back in January. The world
freaked out that this all of a sudden
got more efficient. we need that to
happen like 10 more times,
>> right? Is like those efficiencies are
good. It brings in new use cases. It
lowers your cost like your cost per
token or cost per task. And for this to
be like to really permeate society and
to develop the most like the most good
for humanity, we need the cost to drop a
lot, right? So like someone's going to
solve those problems and I hope they
solve them soon.
>> All right. Well, Michael Brian, uh,
[music] so great to speak with you.
Thank you for taking all the questions
and talking through the tricky stuff and
some of the fun stuff. And uh we hope
that you'll come back soon.
>> Thanks for having us.
>> Thank you.
>> All right, everybody. Thank you so much
for listening and watching if you're
here with us on YouTube or Spotify and
we'll see [music] you next time on Big
Technology Podcast.