Structured Private Credit Series: Purpose-Built for AI: Kristen McVeety (GC of CoreWeave) on How CoreWeave Scaled From Startup to Public Company

THOMAS PROMMER (00:07)
Hello, and welcome everyone to my podcast. I'm here with Kristen McVeety, general counsel of CoreWeave. Kristen, thanks for taking the time.

KRISTEN MCVEETY (00:14)
Thanks, Thomas. Thank you for having me.

THOMAS (00:16)
To kick it off, Kristen, I went into my favorite AI agent as appropriate for this episode.

KRISTEN (00:20)
Good to hear.

THOMAS (00:21)
And ask, “Give me a half a sentence description of CoreWeave.” And here's what came up. CoreWeave is a purpose-built cloud infrastructure provider for high-performing workloads. Anything you want to pull apart here?

KRISTEN (00:33)
Well, I love the fact that it nailed it with the purpose built because that's exactly what we are. We were very much and are a new kind of cloud built purposely for AI workloads and all AI workloads, as you see AI moving through the iterations that it's moving through from training to inference to agentic. But our cloud was really built for all AI workloads.

THOMAS (01:04)
And interesting, I looked at CoreWeave's story and your story, and I saw a parallel I quickly want to discuss. So CoreWeave starts out with the team that founded CoreWeave originally as a commodity trading and, I believe, fund management firm, right?

KRISTEN (01:22)
Yes, correct.

THOMAS (01:23)
Then they went with CoreWeave, the 1.0 version, into crypto mining with GPU, and then it became now the CoreWeave as we just discussed, the high-performance cloud computing software company. And you started out in Big Law, like many of us lawyers, then you started your own niche law firm, and then you joined CoreWeave as well.

KRISTEN (01:43)
Yeah, I really love that you started at the very beginning, because it's really quite a unique arc and parallel arcs, as you say. So it was in Big Law that I started working with the company and formed it, Natsource, where two of our co-founders were at the time. And I was a junior associate. And at that time they were doing, within the energy markets, the fund management and the trading of derivative instruments. And so you saw over time and really explains why I'm even here today sitting as GC of CoreWeave, it was working with incredibly dynamic risk managers, the most dynamic risk managers I had ever seen, and building a very dynamic working relationship with them. And that sort of gets fortified over time with an enormous amount of trust. And so it was building that relationship so that when they moved into, wasn't called CoreWeave back then, they moved into the crypto mining space, it was also seeing where they were going to take it and where they were going to go. And I got to see that vision and have the opportunity to work with just, as I said, risk managers, but incredibly visionary risk managers.

THOMAS (03:03)
Did you draft the original formation documents?

KRISTEN (03:05)
I did. Probably under the supervision of a more senior attorney, but it was an incredible opportunity to have a very close client relationship. And then when I founded my own firm, this bespoke firm, it allowed me to start working with clients in a different way, but had always started with being that M&A capital markets attorney and counsel and working with companies to really build their vision.

THOMAS (03:36)
So life does go full circle.

KRISTEN (03:37)
Yeah, full circle and been an incredible journey.

THOMAS (03:41)
And so if you talk about the more recent milestones of CoreWeave so that people can get familiar with, I'm thinking pre-IPO, Microsoft offtake contract 2023, I believe, right?

KRISTEN (03:56)
Yes.

THOMAS (03:57)
Yeah. So negotiating the first Microsoft contract, what was that like?

KRISTEN (04:01)
That was amazing to even be able to be anywhere near a transaction like that. So that really started with, we were a small company in an incredibly capital-intensive market. So we started with doing these first unprecedented customer contracts remarkably on our paper, and they had to be bankable.

And so we had to enter into those conversations with those incredible counterparties and move through and then bring those customer contracts over to the financing side. So we looked at each of those customer contracts that were based on long-term customer contracts with the most creditworthy counterparties really on the planet. You have the Microsofts of the world, the Metas of the world, it was with them. And you move that over to the financing side, and once again, you got to start to explain the vision of the company, but you were really asking for the financing of really the receivables off of those contracts in addition to the GPUs that were securing the contracts.

THOMAS (05:14)
And to unpack that maybe for the audience, so when we talk about offtake contracts, let's say Microsoft is going to CoreWeave and saying, "We want to buy cloud computing capacity from you guys so we can run our models or whatever we're trying to do." You now have a contract from an investment-grade counterparty, and you basically use those receivables coming in from this investment-grade counterparty to finance your infrastructure, correct?

KRISTEN (05:39):
That's right. I mean, it was an incredibly disciplined and important loop that we had to get into, and we knew that it was within the debt space that we had to do it because it was so capital intensive. And again, for a small company, a startup company, trying to reach hyperscale with a completely new kind of cloud, you had to go into the debt space. And the founders are incredibly comfortable within the debt space and understood it, and therefore could forge this new kind of debt facility that was a GPU-backed debt facility. And so it did sort of become this, well, what are they underwriting? What exactly? So it was understanding and working with them on how we could get them to underwrite it.

THOMAS (06:34)
Yeah. And people who are listening, they might not know that you and I worked across from each other. My team and I, we represented the lender consortium. I think it was an $8 billion —

KRISTEN (06:44)
Lead lender.

THOMAS (06:45)
Lead lender. Yeah. The $8 billion facility, which was no capital markets, no traditional bank, but rather those private credit providers. And I remember what I found fascinating was you have this brand-new technology, and as you said, you need to make it bankable, meaning you have to translate it to investors with the different investment committees. And very interesting, you said you used your form, not Microsoft’s form, because your form became the precedent in the market really.

KRISTEN (07:11)
It did. And again, that was just an insane opportunity. And in one respect, we've always said that the demand for AI, the demand for our cloud to run those workloads, to sit on top of the GPUs, that infrastructure was insatiable. So in some ways that did help push through the agreements being on our forms, but we could go into those negotiations. And I would say as counsel, it was incredibly important to go into those negotiations as really trying to be very in tune with what the business is trying to do with them. So you enter the negotiation knowing that it's got to be bankable on the other side when you take it to finance it. And then you see that work come to fruition when you do close the DDTLs. And that first one where we were on the opposite side of the table from each other, I mean, you could tell me what was the thing that got everybody comfortable?

THOMAS (08:20)
I think it was number one, really, who were the off-takers, right? Investment-grade companies. So I think there was a lot of confidence that they wouldn't walk away from the deal or couldn't walk away from the deal.

KRISTEN (08:30)
Yeah, they were taking long-term take-or-pay contracts. So that was incredibly important.

THOMAS (08:35)
And you were buying GPUs from Nvidia, right? And I think that was the time when Nvidia's stock price went through the roof, their chips were in high demand. So the lenders were looking at it and said, "Wait, we have those receivables from those investment-grade companies, but we also have these unique assets."

KRISTEN (08:48)
Yeah. And that started to evolve. The DDTLs started to evolve, and you get to see what exactly are they underwriting as we take our, we sort of formed a bridge between the technological innovation and the commercial execution and you start to get that track record, and how did the DDTLs evolve after that and what are they underwriting? Certainly in the first instance, it's the off-takers and the credit rating of the off-takers. And then you also see the GPUs because frankly, we didn't choose the GPUs, our customers chose the GPUs. That's what they demanded. And we were able to lay the most performant infrastructure on top of the GPUs because again, we were purpose-built. We weren't retrofitting a legacy cloud for those GPUs. So we were able to make them more performant than any other infrastructure company. And so then you see it evolve and then you come to Q1 of this past year, and we closed the first ever investment-grade rated DDTL,. One, you suddenly now are seeing that they were willing to investment grade it, and that was based on two things. It was still based on that you had long-term contracts with extremely creditworthy counterparties, those off-takers, but that was now in combination with the GPUs as becoming a separate asset class. And then importantly, it allowed us to just slash, literally slash our borrowing costs, our cost of capital. Because in the first DDTLs in those early years, it was really up around nearly 15%. And then you start to see that whittle down to nearing what Oracle has as its cost of capital. 

THOMAS (10:44)
Which means you slash it by approximately two-thirds.

KRISTEN (10:46)
Yeah, which was incredible.

THOMAS (10:47)
In a matter of two years or so.

KRISTEN (10:49)
Right. No more than two years. So you see that evolution, and then I would say then you continue to see what the most remarkable and most recent evolution of the DDTL, which is DDTL 5.5 that we just closed in the last quarter, Q2. And now you're seeing suddenly there are actually shorter-term contracts in there. So you see what's being underwritten is changing or evolving. And so you see now the focus is less on the off-takers. It's more again on always that GPUs as an asset class in and of itself. But what you're really starting to see that was a result of our track record that we were building is the underwriting of the CoreWeave cloud. It's the underwriting of that remarkable purpose-built technology solution for all users of AI, for all consumers of AI. And so that was really a remarkable evolution.

THOMAS (11:51)
And I think a few things we should zoom in here because they're very interesting is number one, you say DDTL for the audience, that means delayed draw term loan. Because the way these facilities work, whenever you get a new customer contract, the lenders allow you to now draw on the facility and they actually fund so you can build that infrastructure. And what's interesting in the first few iterations of the DDTL, there was a parent guarantee from CoreWeave, right?

KRISTEN (12:17)
That’s right.

THOMAS (12:18)
The way we set up these facilities is we create a new entity below the CoreWeave structure.

KRISTEN (12:23)
That's right.

THOMAS (12:24)
And then we put in the contracts, the GPUs, the infrastructure and fund it. But we still had a parent guarantee. Now in the latest investment-grade capital markets transaction you have, you no longer have that. 

KRISTEN (12:35)
Yes, it’s a non-recourse, effectively non-recourse lending facility, which is again sort of an affirmation of what they're underwriting. And of course you have, and it's not based on CoreWeave's corporate credit rating, it's just based on the CoreWeave cloud.

THOMAS (12:57)
And technology that you are managing.

KRISTEN (12:59)
And the technology that we're managing.

THOMAS (13:00)
Because I think what's important to mention is these separate facilities, these separate vehicles, they're still managed by CoreWeave, right? I think that's the important part.

KRISTEN (13:07)
Very much so. I mean, we could lay the CoreWeave cloud over anybody's GPUs. And again, they're making those GPUs as performant as possible for any kind of AI workload. That's the other thing. It’s just customers aren't coming and saying, “Well, I just need it for training or I just need it for inference.” We’re saying we can basically fit and take on any AI workload. We've always been just meeting the customers wherever they're at, where they need to go, so that running their AI workloads is going to accelerate their business and the profitability of their business.

THOMAS (13:45)
And slashing your capital, your cost of capital by two-thirds is – 

KRISTEN (13:48)
Was enormous. 

THOMAS (13:49)
Because I talked to a friend who's a credit investor and I ask him, "What do you think about CoreWeave and their competitors?" And he said, "I like the business model. The only thing I'm always conscious about is it's a CapEx treadmill.” So whoever has the lowest cost of capital has really a competitive advantage in this field that requires so much capital expenditure.

KRISTEN (14:08)
It requires so much capital. And we've really reached the point where nobody can say that they can do it alone. But it took this ... I say at the company, you have things that are working together. And I would say you have this, again, this very elegant, beautiful, purposeful technological solution with this financial engine that is now sort of becoming this loop and one serves the other. And so if you're able to offer the market the most remarkable technology for AI, then you're able to — again, we see the numbers and they're unprecedented — but you have to be able to have that be attractive enough and have lenders have an appetite for what they're going to underwrite. And you do see it with even our other financing with, I'd say it was really important, shortly after the IPO, we came out with a high-yield bond offering. That was incredibly important because it, again, was letting us go into the public debt space, again, where we're very comfortable. And again, just work at decreasing that cost of capital. And all of our public debt has been tremendously oversubscribed. So you just do see there's an understanding, and it took a while, but there's an understanding of what we do and what we offer, and there's an understanding that we're the best at it in terms of these infrastructure companies. And now you're seeing also just the evolution of our tech stack and the products, the amount of products and services that we're able to build on top of that infrastructure and now offer to a very diverse and broad range of customers that are either building intelligence or using intelligence.

THOMAS (16:17)
And now that the deals are sitting on the same side of the table, I can ask the question, what caused the most friction early on in the process when you were drafting those and negotiating those offtake agreements, the first facilities, and especially for people watching who may be in a similar spot or want to get to a similar spot, what were the biggest learnings?

KRISTEN (16:35)
You're working with Nvidia based on the merits and we're just getting it into the market. So I think it was just, probably the most tension was, again, we were in early days of our track record, so really that's what forced the idea of we know what kind of customer contract we have to bring to this initially, and it had to be long-term. It had to be with those counterparties, those investment-grade counterparties, and it had to basically then become the receivables from those contracts. Those contracts were dropped into the SPVs —

THOMAS (17:13)
Special purpose vehicle.

KRISTEN (17:14)
In that structure. It was really also those receivables became ...

THOMAS (17:20)
And how did you balance, especially from in the early days, how did you balance this rapid technological development while at the same time considering the legal framework you had to fit in?

KRISTEN (17:31)
It's a good question. Again, the demand was insatiable. So in some ways that helped us get to contract. Time kills all deals. So the demand was absolutely insatiable. The counterparties were coming into the conversation. It was very much a very transparent negotiation with those customers. It was also great to, again, explain what we could offer, give them the chance to then tell us their pain points and get to paper that really reflects the enormous investment that both sides are making. But I'd say both parties came to those discussions knowing that the technology was going to so rapidly evolve that in this case, again, months matter for at least the AI-native companies to get their offerings into market. So when months matter, you have to turn to a company like CoreWeave that is actually executing, that's actually getting them powered up. And then we have this purpose-built solution that then gives enormous visibility to the workloads that are being run. So they could see they need a reliable platform, they need a secure platform, they need something that is going to draw the most out of the chips that it's sitting on. And this is an entirely from a blank page, we love a blank page. This is an entirely purpose-built proprietary infrastructure layer.

THOMAS (19:21)
And you gave investors the option basically to be also the first-of-its-kind structure in the deal that they could participate in.

KRISTEN (19:26)
There's nothing like offering a really novel vision for people to invest in. But at the same time, you have to be very disciplined in knowing, okay, how does the contract need to look so that, and I can assure you that, and well, you were part of it, that the review of those contracts, it was nice to actually go and negotiate the contracts and then actually bring the contracts, have the lender's eye on them, and then be able to actually walk through every clause in those contracts. Does that get us more comfortable? Does that get us less comfortable?

THOMAS (20:04)
Okay. And so a step forward, you went public. Congratulations. I know it's been a few months. 

KRISTEN (20:10)
Thank you. 

THOMAS (20:11)
One thing I've always been wondering, if someone gets public, there must be this moment where you guys say, "Are we ready or not?" What was that like? What's the decision framework around that?

KRISTEN (20:21)
I would say it actually starts with, you have to understand we knew why we were going public. We knew why we're doing it. We weren't thinking of it as simply a liquidity event, not that there's anything wrong with that, that's why a lot of companies go public. We knew we had to reduce the cost of capital. We were this technology and both physical aspects of our company that were so capital intensive, if we didn't lower the cost of capital, we weren't going to be able to reach hyperscale. And so that very much explains the going public. So when you know why you're doing it, that absolutely helps with when you are looking for the signals from the business to say it's a go, no-go situation. And they're incredibly opportunistic and disciplined about finding that window. I would say we came out into enormous chop, but we had to come out when we did, because again, you are seeing these step functions play out. We couldn't get to the next step function as a private company. We just simply had to get into the public markets. When you know the why, you are just making sure that you’re not the reason that we’re not going to go and you are just waiting for the signal.

THOMAS (21:52)
And for general counsels who are listening in, who are IPO curious but maybe not ready, what's the realistic timeline?

KRISTEN (21:59)
Well, I mean, not that we followed it, but I would say the realistic timeline is probably something like 18 months to two years of really full prep. Again, we didn't follow that. It is absolutely – I would tell any GC at a company like this considering, especially a technology company where the technology is moving so fast, you do an enormous amount of triage. Again, you stay very in tuned with the business and as to what the business is trying to do, where it needs to go. And I would say always be understanding that and making sure you're aligned. You're not just sort of adjacent to the business, you're not this sort of necessary add-on, you're really in the room. And it is interesting because I look back and I think it was from those early days and working with some of our co-founders that I sort of just knew how I was always allowed to be in the room, and I was never asked to take the counsel hat off. It was always just you are in the room for a reason, so bring something of value to the room. And then you are just working incredibly close as a unified team to, again, realize their vision. There's a lot of triage that goes on.

THOMAS (23:28)
I recall you as a translator whenever we had a question, like from the technology to the legal side.

KRISTEN (23:32)
Yes, yes. That does help. 

THOMAS (23:35)
It definitely did.

KRISTEN (23:37)
When I started, I don't think I knew how to spell GPU, so that definitely does help.

THOMAS (23:41)
I'm sure you’re being humble.

KRISTEN (23:42)
Well, you learn fast, but you learn along the way. And again, you surround yourselves with just great teams. You need that. You have to be extremely intuitive about sourcing so that you're going so fast. We were going so fast that through a combination of outside counsel and inside counsel, you are surrounding yourself with people that are mission oriented and understand where your leadership needs to get to.

THOMAS (24:15)
In order to be mission oriented, you need a long-term view and always think of the companies like CoreWeave, they invest on the long-term timeline, but now you have the quarterly cadence of reporting. How is that?

KRISTEN (24:29)
It's really not something that I feel negatively about at all for a few reasons. We talk to the markets constantly. We're also talking to the markets because we were arguably the first AI company to go public. You had the hyperscalers already out there with the legacy clouds, but they had gone public years ago and they had not gone public on AI. So we had to come in and be incredibly, again, transparent and thoughtful about educating, just like we did with the lenders on the DDTLs. That sort of teed us up extremely well. We knew that we were explaining something entirely new. What is an AI cloud? What is an AI hyperscaler? Where is it going? What can you do? And so now having to report quarterly, again, we're just always in the market. We're always talking to the market. I think it's an important, especially for finance, for legal, it's an incredibly important discipline to build. I understand why certain companies who have been public for a long time, yes, we think long out, but again, the technology is developing so fast. Thinking long out, it's okay for us to have to file and check in with the market every quarter. I just don't see it as a negative thing for companies like ours.

THOMAS (26:06)
And in that regard, how has your role changed from being a strictly GC person to now also I think having a public affairs angle that comes with your role?

KRISTEN (26:16)
Yeah. I mean, I had to somewhat step back from being so close to the deals, and that's hard for a self-admitted deal junkie, but to really seeing… I remember when we went public, my first thought was, "Well, now we're just going to go global." And that immediately, because I knew we had what the world wanted, that technological solution, and we were doing it. It's always been going out and doing it for the customers and seeing what they need and building the cloud for them and bringing the cloud to them as opposed to putting them on our cloud, very different mindsets. So when it came out and I knew, "Okay, we're going to go global," but then you balance that with all politics is local, as they say, which is really true. So I knew that there was going to be this expanding aspect of my role as GC, which is that public and government affairs aspect of it, and really being able to hone in on that as a strategy.

We had started that, a very meaningful and genuine engagement with government while we were a private company and got to have those sort of magical moments of walking in a room and them not knowing what on earth we did and who we were, and then them suddenly realizing, "Oh my goodness, there's another voice in the room," and governments were trying to get as educated on AI and where it was going as quickly as possible. So now they had another voice that quite honestly came out of just having an environment where a company could compete, a small company like ours could compete, could become consequential, could reach hyperscale, and that takes a certain kind of regulatory environment. So as we became public, I knew that I was going to serve more and more in that role of just making sure that we were looking at this regulatory landscape, we're looking at public affairs and government affairs in a way that just the scope of the risk that I was assessing just became much broader and working with many more stakeholders on that. And then again, trying to just bring a very meaningful, genuine engagement with the governments in a way that served CoreWeave quite frankly, because you can have where suddenly big is great and big has advantages, but best is kind of cool too. And so —

THOMAS (29:14)
Especially, you said all politics is local, which is a great segue actually into my next question is we see a lot of, at the local level, moratoriums on data centers, given the concern with rising energy prices, water usage, and whenever we do some sort of GPU deal, we always have to consider a patchy framework environment almost because there's different moratoriums in different towns. What's your view on this, especially as you deal with the local communities, I assume?

KRISTEN (29:41)

We do. We absolutely do. Our focus is hyper local. It has to be, but it goes up and down the government ladder from federal, national, international, all the way down to local. And in each jurisdiction, it has that same ladder in it. So I would say it's interesting particularly with the moratoriums, what's interesting is we're seeing no pullback or reduction in AI demand, demand for AI and for the infrastructure. So I think what the moratoriums will do is they're not going to impact that demand. The moratoriums will more just have an effect on where the infrastructure is built. And so we obviously, we approach it very carefully. We approach it early. We come in very transparently and we engage in a way, that the way I see it is we always come in and these concerns are valid, and so you want to have good, informative, accurate conversations with the constituents, whether it's a local government, a state government, and even at the federal level. You want to basically come in and I always want to make sure we leave every engagement where I have a better understanding of the concerns. And then we also leave it where whoever we've engaged with has a better understanding of how we build. And as we get into more and more self-builds, we're about to have a project that we're extremely proud of in New Jersey start to operate towards the end of this year. And so it's not only where you build, it's how you build. And I would say not all companies are building the same way. I can tell you that we're building in a way that sees the community as a long-term partner of ours and it's imperative that we build responsibly. It's also imperative to your point about costs. We build and fund all of the grid upgrades that are required and then we pay for full freight for the power.

There's also an enormous amount. The AI data center is very, very different from the legacy data center, so we build responsibly. And so as we do it, we're paying the full price for the grid upgrade and for the power that we're taking and we're employing closed loop cooling systems so that there is less water usage and we're reducing that and we're reducing noise. The AI data centers are very different, at least the way CoreWeave builds them, very different from the legacy data centers. So they're built in a way to reduce noise. It makes as much noise as your kitchen. And then again, it's also giving the community the opportunity, quite frankly, because there are jobs and there are ancillary jobs that come out of us being a good neighbor and being a part of and bringing that community into the AI economy, quite frankly. 

So again, the moratoriums, and there's a lot of different flavors to the moratoriums. I would say what's recently been put in place is really not. It's a moratorium, but it's also giving a framework, which I think is incredibly important in acknowledging that AI is not going to stop being built. We're going to have to build it. The U.S. is poised to be the most competitive nation in terms of the AI economy, and so it's going to get built. AI doesn't have to leave anybody behind, but it has to be built responsibly.  And for Pennsylvania to put out a framework for how it should be built, which is honestly how we're building already, I think is a really responsible way of governments supporting what is happening. So yes, I don't think all moratoriums are created equal. Just like all clouds aren't created equal, I don't think the moratoriums are all created equal.

THOMAS (34:13)
Perfect analogy. And so we've covered the past, the present regulatory framework. If we look in the future, what's going to be the growth story for CoreWeave going forward, especially if you think about GPU as a commodity versus a proprietary technology that you can offer?

KRISTEN (34:32)
Yeah, our growth story, that's a great question because it is something that you literally feel.

THOMAS (34:39)
And it's hard to beat the past.

KRISTEN (34:40)
You feel it physically. It's so powerful, the growth story. Where I see it going is, first of all, it always comes from the same starting point and always has, which is serving the customer, going and meeting the customer where they're at. If you look at our growth story, what you should look at first, I would say is the diversification of our customers. So you've really gone from, and again, you see it in the contracts even in the DDTL. So you've gone from the AI native companies to now really moving into, and you see that flow into inference, which is now starting to really show the monetization of AI is really in inference. But now you're looking at the enterprise customer and you just look at the CoreWeave tech stack and how that's evolved. I mean, last quarter alone we introduced seven new products to be offered off of our tech stack to our customers. And there is an explosion of demand from enterprise customers in the CoreWeave solution, in the CoreWeave cloud. And so the growth is really happening in terms of just looking at where AI is. AI is going to change everything. It's going to be everywhere. And so when you have a cloud that is so driven to serve all AI workloads, and now you have enterprises saying whether it's you see an explosion within manufacturing and industrial automotive, we've enjoyed an incredible customer relationship with Jane Street. You now see it with Hudson River Trading. You see it in fintech, you see it in life sciences. So now you see basically us bringing AI and bringing that platform, that infrastructure and those products that we get to run, as a continuous AI loop. So we're now offering products and then that gives us incredible visibility and insight into what the customers need to do with the platform. And then you just create this AI loop, and it gets better and better and it compounds and it becomes more performative.

THOMAS (37:10)
And presumably the idea is that you customize it to the customer, you get feedback and then you further customize it and thereby you already have a proprietary angle as compared to just any sort of commodity solution out there. 

KRISTEN (37:22)
And that's the way the infrastructure was set up. That's the proprietary software that again, gives us that incredible insight and visibility into how to actually make the platform better. Then quite frankly, just this is the AI loop in motion is just their use of it is then informing the platform and the platform gets better. And the fact that we're just offering more and more products off of the platform is a result of that AI loop sort of proving out. And so I think that's really the growth story. We just are making the silicone that's most in demand. There's still no question. We run on what our customers want us to run on and that has not changed, but we're also offering just tremendous more. Now you're just seeing it run on older generations of chips, which is really, really interesting, which is something we've been telling the market for years now, that there is this longer life to these chips.

THOMAS (38:29)
And that was a big question by the investors always, what's really the longevity, the life expectancy of a GPU?

KRISTEN (38:36)
And of course that was just a big question mark, right? And now you're seeing it in motion because we have, particularly with our managed inference, which has just sort of exploded recently, you are seeing that those older generation chips are still incredibly valuable and you're seeing those being put to use in a way that actually works for enterprises. And now we're just so excited to partner with those enterprises. I think Caterpillar is a very powerful example of a customer of a logo that we brought on and we're working with, but that's really to work with them on the physical AI and them developing these sort of autonomous construction vehicles. So that's where suddenly you're just seeing the entire vision of the founders just play out in the market, which is absolutely incredible.

THOMAS (39:40)
The growth story isn't over yet.

KRISTEN (39:42)
No, the growth story is not over and I would say still early chapters. And as I said, it's just really, really cool to see, again, when I think of my role in it, you're just sort of out there telling the story, but knowing that you have to bring a credible, reliable voice to every engagement that you go into. And so it's just really super important that we stay very focused and very transparent and very responsible in how we're going about doing it.

THOMAS (40:24)
And to close out this episode, any common misconception you want to dispel about CoreWeave?

KRISTEN (40:29)
I would be lying if I said it didn't absolutely irk me every time I see a headline that just says that CoreWeave rents out GPU capacity from its data centers. I think that's a massive misconception. I think it was through just this magical combination of, again, technological innovation and technological focus on the customer with this disciplined but also innovative financing engine that you forged this totally new cloud and an AI cloud. And that's what CoreWeave is. 

THOMAS (41:17)
Great. Kristen, thanks so much for your time.

KRISTEN (41:19)
Thank you for inviting me.