CoreWeave co-founder Brian Venturo on GPU cloud infrastructure, Aston Martin F1 deal, and why late entrants can't catch up

there. Um well, our next guest is ready to join us. This is Brian from Cororeweave. We will be talking about the Neoclouds and uh AI training and what it's like being a public company. Yeah. So, good to meet you, Brian. How you doing? What's going on? Pretty good. Welcome to the show for having me.

Congratulations on all the success. Um, can you uh just take us through like a brief introduction and kind of maybe a little bit of history on the company, but kind of then kind of where you are today because I I I imagine we can jump into a bunch of stuff. Most people are probably familiar, but Yeah, sure.

So, uh, first off, my name is Brian Benturo. I'm one of the founders. I'm the chief strategy officer of Cororeweave. I started out as CTO until they found me out. I've been demoted to chief strategy officer. So, uh, it's promoted. It's got to be more fun. More fun. No, it's it's demoted, dude. Definitely demoted.

Um well well I'm just saying like you know dayto day I I imagine you're you're you know you still have the stress of you know running the business and and all that but uh but but maybe you're not on call which would be nice. Nothing's nothing's changed. They just don't let me write code anymore. Yeah.

Um which is probably better for everybody. Uh but you know I'm the chief strategy officer. I I run um a lot of product strategy, infrastructure strategy. I work with clients on big uh big deals. Mhm. Um, you know, probably half the company indirectly reports to me. Um, so I'm on the board of directors, etc. , etc.

So, I've been I'm I've been involved in pretty much every aspect of the business. Um, but I'm happy to talk about whatever you guys would like. Today, I want to talk about naming every strategy. I want to talk about the overarching strategy. Um, walk me through how you would think about defining the current strategy.

what is the highest priority relationship with hyperscalers, new companies, creating a fluid compute layer, um all the different like government incentives like what what is top of mind for you uh on a day-to-day basis in terms of positioning core against competitors or just broadly solving a problem in the market like how do you actually break down these problems?

So um it's two things right the the first is um serving the customers and partners that got us here right that's the big AI labs it's the hyperscalers it's the people that have an insatiable need for compute infrastructure they're the ones that are blocked from delivering to their next customer from growing their user base because they don't have any GPUs um that has been um it's been insane to be a part of right is because every time it increases an order of magnitude you're like it can't go again and then it goes again 6 months later.

Yeah. Um and you know I was listening to the show before you were talking about uh you know meta's 5 gigawatt data center campuses etc. Um you know it's uh like we're headed there right and I think that right now it's the question of you know everybody's in a race and how quickly you can build that like that large.

Uh but it's also understanding what are the local constraints what are the actual grid constraints what are the supply constraints to be able to get there. You know you run into things like labor constraints.

you know, there's not enough electricians in the world to go out and build these things and the timelines these people need. How do you modularize the data center construction? Um, you know, everybody's kind of solving this as the planes in the air already. Yeah. Right.

Um, and you know, it it's it's a race and you know, I I always say internally like I can always I can solve any problem with enough time or capital, right? And a lot of the times you don't get both. Um, so it's choosing, you know, what's the right path to go, you know, how are we going to solve this for them?

Um so that's that's kind of one side right is like okay how do we build to everything that they need how do we understand what their demand's going to be and when you look at um the thing that I think the world misses is that everybody on the hypers scale side and like in our seat as well like we're looking at it going we need to build x number of gigawatts of data center capacity right like we know the demand signals are there but the capital and the speculative capital isn't there to do it interesting right is I can go out and I can build x gigawatts of capacity for the next two or three right from my balance sheet.

But any more than that and the demand doesn't show up and I put my entire company at risk. Yeah. Right. So we're we're basically having to haircut the demand signals that we get and only build to a certain extent of it and then by the time we get out there we're in the same constraint position we're in today. Got it.

Um talk to me about how how do you um real quick how do you think so so it was 2019 that you guys decided to focus more on AI and and cloud and for broadly is that correct roughly? Yeah.

So, um, and the reason the reason I the reason I asked this is because I saw a company as, you know, there's been a bunch of companies that were, uh, effectively came to the conclusion that you guys did, but maybe six, five, six years later in terms of, hey, we're going to sort of pivot away from, you know, Bitcoin mining into AI because that's where the real growth is.

And it feels like a lot of those company from from an outside looking in uh you know it feels like those companies are going to be at a massive disadvantage. But I don't know if I have if I have the right read there. Yeah.

So we um when we started out in 2017 um we made the decision that we were never going to be as good as the ASIC producer in mining Bitcoin. Yeah. Right. And the idea there was, you know, there's always going to be something they don't release. It actually wound up being true.

Uh they had released like their low power mode, I forget what it was called in 2018 and everybody found out that they were hashing 30% better uh power efficiency. The world lost their mind about it and it was kind of like the okay, we were actually right.

Like we were kind of were paranoid about it, but we never actually knew until then. Um so, you know, we said we're not going to do it as good as them. Where is there a level playing field? And we thought there was a level playing field in GPUs.

Y and we also thought, okay, if crypto goes to zero, the GPUs are repurposable. They're used for gaming. They're used for graphics rendering. Like we could either sell them on the open market or we can build another service with them.

Um what we were most surprised by is when we first launched uh our you know let's say first non-crypto product in 2018 2019. Um it was a rendering service for an open source uh uh 3D uh graphics platform called Blender. Blender. We had like a thous a thousand people sign up in the first day. Wow. Wow. Yeah.

And I was just like, "Oh boy, like there's a market here. People need this stuff. " And like we had never expected it to be that large. And people would come to us and they send us like uh support tickets saying, "Hey, just want to tell you it's so awesome.

I was able to do so much more work because I got access to GPUs I've never had before. " Yeah. Right. And that was like signal number one. And we're like, "Okay, we convinced our early investors like we're going to go all in on this thing. We're going to use crypto to pay for it as we're building out more uh compute.

" And uh we set it up so that as we didn't have uh actual cloud or compute workloads, it would turn back to mining and the mining would cover all of our fixed and variable expenses and all of the cloud stuff was just upside. Yeah. Right.

So we were able to show over time this like constant penetration or kind of constant growth of our margin um which was really easy to grow. Right. We were able to lever the cryptocurrency into a massive resource base.

So the timing played a lot a lot into our success there and that we had that permissionless crypto revenue to lean back on. Yeah. but also just having having I think incredible foresight to some degree. Uh which which uh you're being a little bit more modest about it.

But the fact that companies, you know, even I I saw a company uh that was getting pumped on X last week that is still in the process of transitioning from Bitcoin mining to uh you know some something that will look like a Neocloud.

And the idea that you're going to come into this market six years late and really be a a key player uh feels like a long shot. It's um so I think it's even worse than that. And it's even worse it's even worse than that because the talent pool is so small. Yeah.

Um and to be successful as the technology has gotten more sophisticated and harder to run um you have to be the best in the world, right? And you know, we have some of our low-level hardware engineers. They're amazing. Like, I've seen them um, you know, solve problems that nobody even knew existed.

They they identify the problem. They're able to say, "Okay, this is what this is what the problem is. " They go back to Nvidia and they jointly develop a solution for it, right? Um, that that's really important as you're building at scale, right?

You know, anybody can run a hundred or 500 or a thousand GPUs, but if you're running hundreds of thousands of these things, like it's really really hard to do, especially at high quality. Mhm. All right. So, um, people think that there's folks that are 12 to 18 to 24 months behind us. Um, good luck.

Uh, really quick question. Uh, I have to ask, uh, what's the story with the wheels behind you? Uh, all right. So, um, there are F1 wind tunnel tires. Yeah. Uh, and we just did a So, we're uh, we just did a sponsorship with Aston Martin. Um, and we also and we also executed a uh compute agreement with them.

Um, I'm uh so excited about this one. You know, when I talked before about like the way that we think about our first cohort of clients, it's the AI labs. The second is the enterprises. Yep.

And I feel that if we can walk into Aston Martin and help them modern modernize their AI and ML stack and in the most hyperco competitive sport in the world with so much data and so many eyeballs already on it, we can do it for anybody. Yeah. Right. So, so being more concrete there, you have an F1 car.

You need to simulate how air flows over it. That's a particle simulation and that's perfect for accelerated parallelized computing GPUs. They're probably doing it on premise at some point. They're going to take it to the cloud. They're going to take it to your cloud. It It's so much Yes. But it's so much more than that.

It's not just the CFD workloads. It's also things like taking radio data from other teams and trying to decipher what their strategies are in real time. That's awesome. Uh I mean it's running tire degradation simulations like how does track temperature change with cloud cover?

How does weather come in and actually impact this stuff? It's helping them make better decisions on track in real time. Wow. Um it's like it's the coolest confluence of the technologies that I can imagine and I'm like I'm totally into it. That's amazing. Yeah. translate all of Ferrari's comms from Italian into English.

Yeah. Well, I mean, good luck with Ferrari strategy. So, talk about uh talk about uh sort of data center buildout timelines. I think Colossus and and the XAI team were getting a lot of credit for bringing a lot of compute online very quickly, a lot of energy online.

Uh how much was that an outlier or are they just really good at marketing? Um I I think it's both, right?

Um you know, one of the things that I admire about Elon, um is, you know, he sent an email a couple years ago on Thanksgiving and it was like the biggest problem that we have right now now is there's nobody to turn a screwdriver on the line. I'm going to go turn a screwdriver in the line, right?

And he is so good at identifying what the bottleneck is and breaking through it. And if you're the richest guy in the world and you don't really care about the consequences, you can do that a lot more than you can as a director of a public company is what I've learned.

Um is, you know, they're able to go and I think some of what they've done is a bit of the you told me I can't build it, but I already built it. It's already there. And they're like, "No, you can't build it. " No, no, it's there. Right. Um and when you're in a race like this, that bravado matters. Yep. Right.

Um I I don't know. I don't think anybody knows the truth of what's actually there. Uh but they've definitely been incredible in how they've been able to execute. Yeah. Yeah.

I mean the the that data center seems like it's performing very well, but what I learned from Dylan Patel over at semi analysis with the cluster max ranking is that not all clouds are created equal. And a 100,000 H100s is not perfectly substitutable for another 100,000 H100s over here.

And uh coreweave did extremely well on cluster max.

And I'm interested in pe peeling back a layer deeper into understanding first how do you think you did so well like like what what concretely makes core stand out as a neocloud uh on a on the features level from a customer perspective and then internally what's the secret sauce like is it just great employees?

Are they aligned? Are they working harder? Is there key insight? are you like how do you actually deliver that product at uh at a higher level than you know competitors that Dylan Patel reviewed? Yeah. So so let's let's not get off the XAI thing yet. I'm going to trans transfer into that.

Um you know if they have 100,000 GPUs and they get 60,000 of them running and they don't care about the other 40,000 of them it's kind of enough. Got it. Right. And if they're looking at it saying I don't care how much this costs. I have to win. Mhm. and they're in the position to do that.

And if you're Elon Musk, you are right. That's a powerful position to be in. Sure. Now, the clients that we serve um are typically going to be more costconscious and are going to look to optimize, you know, their spend and make sure they get the whatever they need out of their investment.

Um and what we found is that, you know, these jobs and this infrastructure, it it breaks all the time, right?

And um when you accept that something's going to break all the time and you design for it, your solution is going to be very different than if you're just kind of um you know square peg in a round hole after the fact saying I have to go deal with these failures. Yep. Right.

So what we did initially is we built uh with a incredibly high level of observability at the bare metal layer. Right. And the bare metal visibility is really important because things do fail and you know you have uh memory failures, you have thermal failures, you have all these things that impact your jobs.

But if we're able to help our customers identify what's happening, triage it for them and explain to them why something failed, you know, they don't have to go back and say, "Oh, was it my code or was it one of the 700,000 connections in my cluster? " Right?

They can immediately point to our software stack that says, "Oh, it was this link flap over here. We took it out of service. " you restart your job, you'll be totally fine. Um, that's been very, very, very powerful.

And it's not just that visibility layer, it's also the storage and networking um, and the quality control that we put around those things. Um, you know, I have to give credit to our new C our now CTO, Peter Sanki, who used to work for me until they found me out.

Um he is uh he is incredibly focused on the low-level visibility and quality and drives a lot of that culture throughout the company, right? Of like if one of 500,000 things is going to break, I need to know immediately so I can go in and fix it to make sure my customers are back up and running like 3 minutes later.

Um so it's it's building from that low base and understanding what the problems are going to be. Uh and then culturally, you know, we have people that are here that take so much pride in being first and being best. Yeah. Right.

And you know, if we're not first to market with a new release, like they're like they're going to cry for a week kind of thing.

Um so when we did GB300 a couple weeks ago, or maybe it was just a week ago, like people went out to one of our data centers, like all of our low-level engineers went out there and they lived there for like two weeks to get it done. Mhm. Right. like they go and they they work from 8 am to 4:00 a. m.

every day and they love it and it's a cultural thing. Um, and I think we're really lucky to have that. Uh, bit of a tangent. Were you an H20 buyer at all over the last few months or or not needed? No. And and there there was H20 sales happening in the United States from what we heard.

Who are the kind of uh players that are that were were making use of that? I don't know um if that did happen. I don't know who it is. Um but for us, you know, if we had any power available, we're going to buy the best, you know, the best accelerator we can.

Um you know, there's no reason for us to buy the cut down accelerator if we have power available. Yeah, that makes sense. You you mentioned that one of the key things is like the insight into what's happening in the data center. what uh if a specific chip is failing, you want your customer to be able to know.

You want to be able to I mean, we talked to Dylan Patelli who was saying like sometimes you just need to turn it off, turn it back on, or like there's a whole bunch of different things that can happen uh in that feel um very abstract when you're looking at it from the outside, but are very concrete when you're on the inside.

And that's actually what separates great data center performance is just like the handtohand combat that happens on an inside day.

Um my question is like um we've we've talked to some people who are like we're going to put data centers in space or we're going to put data centers like on the moon or doesn't seem to align with it with your framework of being like we need to build it in in a way knowing that it's going to break often and we're going to need to fix it and hard to do that if it's it's yeah I don't think space is the right place for a data center.

Okay. Um but you know to to the point there of sometimes you have to turn it off and turn it back on to solve the problem. Um we don't find that to be true. Okay. break down is if you if something needs to be turned off, there's a reason why it needs to be turned off. Sure.

And that like that one comment may be what separates us is, you know, our engineering team if they're like, I'm not going to reboot the system. That doesn't solve problems. I want to know exactly what happened there because we have to go solve it because if it happens one time, it's going to happen another time.

Like the scale that we run at these like these skeletons come out. Yep. Right. Um and you know being making making sure that the data center is on Earth and is acceptable by is accessible by like human technicians is really important. Yep.

Um because sometimes you have to make judgment calls on like what's actually causing this and um you know how do you fix it? Yeah. Um so the the one kind of I mean the there's a few things that the space data center folks will tell you about like you know free energy or free cooling or something like that.

We we we don't need to go too deep into that. I guess the question that I'm more interested in is uh how are the recent results from different training runs? Uh we saw this with Grock 4 that the reinforcement learning post training was 50% of the overall cost.

That's kind of the rumor and and there's been kind of mutterings about hey maybe in the future uh you'll need a whole bunch of distributed compute all over the place generating reinforcement data and doing rollouts and bringing those back and you won't need as much of this hyperconentrated compute in one place.

And so maybe having 10 1 gigawatt data centers is better than one 5 gawatt data center. It's just power and compute all over the place. How do you think like the the shape of distribution of uh of compute clusters will evolve over the next few years if you have any insight there? Yeah, so great question.

Um let's let's put it into perspective because you said one 5 gawatt or 10 one gigawatts those are insane by the way like those are huge. Okay. Um like even building a gigawatt data center and its impact on the local grid like that's we'll see.

I mean, you talk to people and they're like and they're like, "There's going to be 100 gigawatts data centers in three years. " Like the the crazy AI people get crazy and so I I'm maybe I'm a little bit on, you know, like out in space here literally. But but bringing back time to Earth. Yeah.

So, um people definitely want to build at that scale, right? When you think about 100 gawatt data center or a 10 gigawatt data center, Yeah. the capital required to build these things and to install the compute inside them, like that's country level. Yeah. Right.

And you know that I don't think that um the broader market understands the investment required yet to do this. Mh. Right. Um but you know back to the question of is it going to be more distributed? Is it going to be more centralized?

I think that right now while the ability to centralize it exists people would prefer to do that because the optionality of having it all together is you know is worth more than having it distributed. Mhm. Um I think that as you have workloads that become more latency driven, right?

So, you know, we're starting to see customers that say, "Hey, I need something that's actually local to a metropolitan area because I'm running um you know, aentic AI for one of my customers or for a banking customer. Um I I need to make sure it's there.

" That's a very different story than 12 months ago where people didn't care where inference was because your first response wasn't for like two or three seconds. Yeah. You actually want it on the other side of the world ideally where it's low load. Yeah. then then you're managing to optimize for cost. Yeah. Right.

But there there's a group of people that are going to optimize for cost. There's a group of people that are going to optimize for performance.

Um you know, for us, one of the products that we're really excited to develop is um you know, almost like a global load balancer product that helps people choose what they're optimizing for, right? Um and you know, we've been thinking about that for a long time now.

Uh because like you said, you have to have that compute distributed and it may not be be because you're trying like if you could choose you'd always have it in one place, right? But when you build it distributed, then you have to have the software services that can intelligently say, okay, where am I going to run this?

What am I optimizing for? Uh where's the data that I need? Right? And that's where um you know, we've been investing in, you know, we've built a massive global uh network backbone. We have a ton of our own dark fiber um to be able to move workloads around and have What does that mean?

What does that what does that exactly mean? Your own dark. So uh we've leased dark fiber between a bunch of our data centers across the United States and in Europe that allows us to install our own optimal uh optical optical gear um so that we can you know flex up bandwidth as we need to.

Uh you know we have some customers that need things like 64 terabs per second between sites like crazy amounts of bandwidth. And the idea there is that they're moving synchronization data. They're synchronizing models between sites. Yep.

Um so you know that's where you know we have to make these big capital investments knowing what's coming right. So when our customers start asking for it was like oh yes we already have it there.

Um so you know some of that is just it's the focus that we have and the specialization to know hey like we better start investing in this because a year down the road like we're going to be screwed if we don't. Yeah. Yeah.

How how has it uh been, you know, what what are some of the differences been in building out data centers in Europe? I know you guys have Norway, Sweden, Spain, and the UK. Is that correct? Uh what what's it been like developing uh new data centers there versus uh what you've done in the United States?

Um not so different actually, right? Uh you know, everybody was initially saying, "Oh, it's going to be so hard to work in Europe. uh you know are people going to show up and work as hard?

And um the one of the things that we've done I think that we've done really well in our onboarding process for employees is we've actually brought all the European teams over to work with our tiger teams inside of our data centers.

And you just blare the Star Spangled Banner just loudly if you're just if you blast music continuously all of a sudden like they have Yeah. moment.

Um, and you know the the culture is infectious and the what's one of the best things about the company is that all of our like the company's so young that there's been no real attrition, right? Nobody's left. Yeah. So the people that are running the data center start as started as data center tech a couple years ago.

Like everybody's still fired up. They still have like this the same drive. They want to be first. They want to be best. And when you have new employees that walk into that environment and those people are genuinely happy to teach them, um, it transfers pretty well, right?

So, I think our European investments have gone a lot better than I thought they were going to. Um, which I attribute largely to our people.

A couple years ago, I feel like everyone in tech had to learn about Nvidia and then everyone in tech had to learn about TSMC and then everyone in tech had to learn about ASML and now people are kind of getting familiar with SKHEX.

Uh what do you think the next big company in the semiconductor stack or data center stack or even networking like what is the next big theme or or subcategory of important foundational technology?

uh people are starting to dig into the the rare earth materials and what's going on like what is on the horizon in terms of like underexplored or or misunderstood as important as having a role as we try and build out these 1 gawatt 10 gawatt data centers what are we going to be hearing about that's such a such a good question and the kind of question man I don't even know um I'm not sure I have an answer for you yeah you may maybe it's networking gear I mean, you mentioned you're putting optical gear at uh and and just hearing what was it 42 terabs a second.

Like that seems like something that could be become very important. Not that you have to call out a single company, just like the the the the overall industry is interesting to be like, oh, you know, everyone's going to be thinking about this industry. Yeah, you know, that's definitely one of them.

Um fiber capacity between metros is going to be an issue. Um I don't know that it's uh an issue that's insurmountable. Yeah. Um there are definitely some gotchas that I can see coming down the the line at people.

Um I I don't want to say it because then I'll give them I'll give them warning and since we've already identified what they are and I feel like I'm in a better position. Sorry. Yeah. I don't mean to put you on the spot too much. No, it's it's okay. What about electrical What about electrical transformers?

Um I've heard a little bit about like there there sometimes there's a shortage. they're very complicated. Like it's this is this like big technical thing that's big piece of hardware. Uh is that like an important uh thing for people to be aware of going forward or like an important industry to understand?

So um the a bit of a tangent off of that, right? Um everybody talks about how there's no power left in America. Yeah. Right. And if you actually look at the data, there's a tremendous amount of power available from base load and load following generation. That's great to hear. Thank you.

Um, and the the problem exists on peak days. Mhm. Right. And like what solves peak load problems is solar. Like solar and batteries will solve peak load problems. Like you're worried about, you know, 5:00 on the hottest July day when everybody gets home to turn their air conditioning on. Sure.

Like that's that's the thing. Um, and you know, everybody wants to talk about how uh they want to use demand response as a way to, you know, bring supply back to the power grid from data centers. Um, you know, this stuff is really hard to bring up and bring down. It's it's not really intermittent load, right?

Um, you know, I think that a lot of this shortage narrative from a power perspective is coming from the fact that everybody and their brother has become a data center developer in the past year. And there's a lot of people out there squatting on power, right? Or squatting on power rates.

And I don't think that the uh local regulatory commissions understood what they were signing up for when they started doing load studies for free and giving power allocations. Um you know but it goes back to my point earlier that the biggest constraint is speculative capital to build out to meet demand that we expect.

Interesting, right? So like yeah transformers are a problem but they're not outside the lead time window. Um you know UPS is everything's a problem if you need it tomorrow. Sure.

But if you're able to plan and make the investments in like in a reasonable period of time and you trust your demand forecasts and the market is there and believes in it um like this is all fine. Yeah. Right.

Do you still uh I feel like the chart that people have been tracking is like China bringing on you know you know basically like copy and pasting um nuclear reactors. Do you think that the US should be seriously, you know, there there seems to be enough people uh now from tech to Washington that care about nuclear?

So hope it used to be very contrarian. It was very very edgy to say, oh yeah, nuclear it's like it's it's safe and we should do it and people it's it's still scary. But yeah, in your view, would you like to see more focus on more attention on nuclear or more attention on on solar if you had to pick?

Uh I mean so nuclear they solve two different problems. Mhm. Right. nuclear builds your base load and your and like it's not nuclear is not load following. Um it's really hard to bring those things up and down. They have long outage cycles. Um there it's a base load solution, right?

So as your data center demand is going to increase so much and you're shifting your base load requirements up, that's the solution. Particularly if you want to retire old coal or you want to retire some of the base load natural gas, um nuclear I believe is the solution there. Mhm.

Um but as you're dealing with peak loads, right, you're looking to match that load profile and you know, solar is a great one to do it. Yeah. Um you know, wind doesn't necessarily solve that problem during peaking. Like wind doesn't peak during the afternoon, it peaks overnight. Sure.

Um so you just have to match like what's the load profile you're trying to match and which part of the curve are you trying to solve for? Is the load profile for uh AI not perfectly matched with nuclear and and is that because of uh inference demand from actual customers?

Like I've heard people use chat GPT during the work week, but they don't use it when they're sleeping. So there's actually maybe it follows more of like the solar load profile because when I think of training, I think of we're going to run this entire data center for months.

I don't know how far off I am on that, but like that feels like perfect. Oh, build a nuclear power plant right next to the big data center. run it continuously, just keep training, and then when you get done with one model, train the next one and just keep it running perfectly matched forever.

Um, but but h how are we actually like like the actual AI workloads, how how do they match to different load profiles? Yeah. So, if you're in a global constraint environment, right, you're serving as much load as you can around the clock from anywhere. Yeah. Yeah. Right.

So, I I don't think that you have any load shape right now. I think it's just flat load and it's like it's pinned to 100%. Got it. um as bullish. Yeah. Yeah. It's insane. You're like, "It's stressful. Stop talking. " It's it's incredibly stressful, but it's it's a good problem to have. Yeah. Yeah.

Um I have people yelling at me every day like, "Where's my stuff? Where's my stuff? Where's my stuff? " And I'm like, "I promise I'm doing everything I can. " Um but I I think that as that like as you have the investments where things are more latency sensitive, um you'll see that the load curve may not change.

It may still be pinned at 100, right? But the types of workloads may be more may be more shaped. Um I'm not sure how it plays out. Right. I I think that you know the the problem that you're trying to solve there is not necessarily data center load shape. It's residential and commercial load shape. Mhm. Right.

Because you're not running like your big commercial buildings in New York aren't running at 3:00 a. m. They're running at 3 p. m. That's the problem. Mhm. Makes a ton of sense. Well, this was really fun. I hope uh that we see you at the track soon.

One of our other uh one of our sponsors is also an Aston F1 uh sponsor, so hopefully we run into you uh on the paddic. I I'll be there soon actually. Fantastic. Amazing. Uh great having you on. Let's make this a regular thing when you guys have news. Uh it was uh awesome to get all your insights. Yeah, this is great.

Yep. Thanks, guys. Thank you so much. Cheers. Bye. Bye. And since we're mentioning public. com, investing for those who take it seriously, multiasset investing, industryleading yields, they're trusted by millions. Uh, our next guest will be joining us in just a few minutes. But first, we can tell you about Wander.

Find your happy place. Find your happy place. Book a wander with inspiring views, hotel grade amenities, dreamy beds, top, and 247 concier service. It's a vacation home, but better, folks. Okay, I got an update from uh a friend uh friend of a friend uh which says that uh his friend No, no, no, sorry.

Our friend says that he has a buddy that owns some RM boutiques and has not heard of the conspiracy. Really, is interesting. Maybe it was it might have been hallucinated. I don't know. I need to find a source for that. Um where where was that actually in here? I will figure