The Manosphere Report

0:00I'm here at NVIDIA's annual GTC conference, and I'm going to interview four amazing AI

0:06CEOs. Stick with us. Our episode is sponsored by the New York Stock Exchange.

0:18Are you looking to change the world and raise capital?

0:20Do it at the NYSE.

0:22The NYSE is a modern marketplace and a massive platform built for scale and long

0:29-term impact. So if you're building for the future, the NYSE is where it happens.

0:37One of the great companies of the AI era is, of course, CoreWeave.

0:41They're building massive infrastructure for these hyperscalers.

0:44And in some ways, Michael Intrader, welcome to the program.

0:47You're the original hyperscaler. You guys got in very early and secured your, I don't

0:54know which GPUs you wound up getting.

0:57You were very early to this trend.

1:00How did you get to it so early?

1:02And how did you build out this first, I guess at the time, neocloud?

1:07Yeah. So we didn't really start it as a neocloud.

1:10And I was running an algorithmic hedge fund focused on natural gas.

1:16And when you build an algorithmic hedge fund, once the algorithms are built, you're really

1:23just monitoring it and testing different theses and doing all that.

1:27But there's also a lot of downtime.

1:29And we got super interested in crypto.

1:33And, you know, we're pretty nerdy.

1:35We kind of dig under the hood.

1:36And we started to get interested in this security layer.

1:38We looked at Bitcoin and the mining for Bitcoin and we didn't like it.

1:42We just thought that, like, there's some brilliant engineer that built the ASIC and they're

1:47probably going to be better at running it than we are.

1:49So we really began to focus on the GPUs, mostly because the GPUs were you

1:55can mine Ethereum with them, but you could also do all these other things.

1:59And really, so right from the start, we looked at the compute as an option

2:06to be able to deploy our computing power to different use cases.

2:10And so, you know, began the company in 2017, you know, spent the first kind

2:17of three years mining crypto, went through a couple of crypto winters.

2:23Because we had come from a hedge fund, our, you know, we have real chops

2:28in risk management and how we think about capital and risk exposure and allocation and

2:33all of that. And so we were really careful around that right from the start.

2:36So we weathered crypto winter really well and began to scale the company and immediately

2:41started to look for other use cases that you could use this compute for because

2:46crypto was pretty volatile. Yeah.

2:48And crypto was a question mark at that time.

2:50Absolutely. Yeah. I mean, Bitcoin was speculative and there were many other speculative projects.

2:55The only other people using this type of hardware, quants, medical, researchers.

3:00So a good way to think about it is like the progression of products that

3:04we kind of started to work on.

3:06You know, first was crypto, but we immediately moved from crypto to CGI rendering and

3:12we built projects that would allow folks that were trying to animate and render images,

3:20you know, kind of what makes the movies cool.

3:23Right. And we started to work on that and then we moved to batch computing

3:27and started to look at medical research and different ways of using the compute to

3:30be able to drive science.

3:32And we just kind of kept moving up the stack in terms of complexity.

3:36Uh, uh, uh, uh. on how GPUs could be used and ultimately in like, call

3:42it like 2020, 2021, we started to really try to figure out how you can

3:49go ahead and use GPUs for neural networks.

3:51And that was not something that we knew how to do.

3:54And so we actually went out and bought a bunch of A100s and donated them

3:59to a group that was working on Eleuther AI.

4:03They were working on an open source project with the thought that these guys are

4:08taking the GPU compute because we're donating it.

4:13They can't really get pissed at us if we're not very good at it initially.

4:15And that worked out really well because - They can't complain about the SLA.

4:18They kept telling us like, we need more of this.

4:21You got to work on this.

4:22And that began to really give us an understanding of what was necessary to run

4:28scale parallelized computing. And we went through it.

4:34I kind of feel like buying those initial GPUs was the tuition we paid to

4:38learn how to run this business.

4:40And then one of the interesting things is all of those guys went back to

4:44their day jobs because they were all volunteers working on this.

4:47They were like -minded scientists.

4:49And when they got to their day jobs, they were all like, I want that

4:52infrastructure. It's built the right way.

4:54That's the way that researchers are going to want to use it.

4:56And that launched our business.

4:59It was an amazing story.

5:00And so you went from crypto to these researchers into academia and deep research.

5:05What's the next card to turn over in the poker game?

5:08Yeah, so what became very clear to us very, very early on was that the

5:15scaling laws were going to drive.

5:17And remember, this is really back in 2020, 2021, before ChatGPT moment occurred.

5:25And we began to understand that computing decommoditizes its scale.

5:32Anybody can run a GPU, but can you run a cluster that's large enough to

5:35train a model that can change the world?

5:37And that's a different question.

5:39And so we really began to think about, like, how do you go about scaling

5:44up your delivery of this computing to clients, larger and larger clients?

5:49And that was the next card to turn is to think about it from a,

5:53okay, you know, there's a component of this that is going to lean into our

5:58ability to access the capital to be able to deliver our solution to the broadest

6:04possible audience, to the most sophisticated consumers of this compute.

6:07And that was really the next card is thinking about it as a business rather

6:11than as a engineering project to be able to deliver the infrastructure and the software

6:19and really everything between, you know, when you're thinking about what we do, we kind

6:23of live above the NVIDIA GPUs, but below the models.

6:27And everything in there, all the software, the integration of software and operations and observability

6:34and all the things that you need to be able to build a cloud that's

6:38purpose built for this one specific use case, right?

6:42So we don't, we don't do everything.

6:43We really focus on one use case, which allows us to - You want to

6:46do web servers. You got AWS.

6:48You know what? They do a great job.

6:50It's like, it's a, it's a great solution.

6:51It was a brilliant solution to solve a problem.

6:54We just looked at it and said, there's a new problem.

6:56And let's go about, let's go about looking at this problem and try and come

6:59up with a solution to deliver compute that solves that problem.

7:03And when did the language model start dialing and calling you for, you know, capacity?

7:08Yeah. So our, our, our first, well, our, our first language model was really a

7:15Luther. Yeah. But our first, like, large commercial was Inflection.

7:22And so, you know, we work with Mustafa and Inflection, and then we really diversified

7:29from there into the hyperscalers, into, you know, open AI across the model, the foundation

7:38models across, you know, and just kept scaling and scaling with the belief that, you

7:44know, once again, the decommoditization of compute, the ability to deliver a solution, and the

7:53solution is building supercomputers that can change the world.

7:56And that's really what we began to focus on.

7:59That was the lead into training.

8:01And now the world has gone through, you know, this moment where we've moved from

8:06research into the productization of this.

8:09It's beginning to work its way in from the fringe of organizations into the core

8:15of what they do. And you can see that every day in the amount of

8:21inference compute that is being driven through, you know, our infrastructure layer, which is just

8:27massive, which is just like one of the massive layers.

8:29And the inference shows your people are consuming it, not just building models, but they're

8:33deploying them and utilizing them.

8:35I always think of inference as the monetization of the investment in artificial intelligence.

8:42So when we see our compute being used to stand up the massive scale of

8:50inference that's hitting our compute every day, and like, you know, inference is when people

8:54ask the model a question, it comes back with an answer.

8:57That's an inference. Or when you ask the model a question and then to go

9:00do something, that's inference, right?

9:02And that's actually where you have the opportunity to really drive value outside of the

9:09model itself, but into the real world.

9:12And that's really exciting for us.

9:15That's what we like to watch.

9:16That's what I like to watch in terms of gauging the health.

9:19What chips are those? So really, you know, we are the tip of the spear

9:27in bringing the new architecture out of NVIDIA into commercial production at scale.

9:35And so when, you know, we were the first ones to bring the H100s at

9:40scale, we were the first ones to bring the H200s at scale, first ones with

9:44the GB200s, and now you've got the GB300s.

9:47And one of the things that's amazing and really fascinating for us is, you know,

9:52people are using the bleeding edge GPUs to train models as the new architectures come

9:58out, and then they take those GPUs, and they move them into different experiments.

10:04And then over time, they move them into inference, and they continue to use them

10:09in inference for a very, very long time.

10:11What is the shelf life of a 100 right now?

10:13So that's been a big debate is, I think, for your company, for Microsoft, and

10:19I guess Michael Burry, you know, who you must have known when you were a

10:22quant, you know, saying, oh, my God, the whole industry is the sky's falling.

10:26And then we all know in the industry that people don't just throw this hardware

10:29away, that they find uses for it.

10:31The street finds its own use for technology.

10:33So what's the reality of the lifespan of these things?

10:36So my take on the GPU depreciation debate is that it's nonsense, right?

10:44It's a debate that is being brought to the forefront by some traders that have

10:49a short position in the stock, and they're trying to talk down.

10:52Look, here's what we know, right?

10:55um when when we buy infrastructure we're a success -based company right we're a small

11:00company on a relative basis compared to the enormous companies that we're competing with and

11:05so they come our clients come into us and they buy compute for five years

11:10for six years our average contract is five years so any commentary by anyone either

11:17inside or outside of the industry that this stuff becomes obsolete in 16 months or

11:21whatever nonsense you're spewing it's it doesn't it doesn't in any way match up with

11:26the facts on the ground the facts on the ground is they're buying it for

11:29five years right if and my approach to this has always been if people are

11:34willing to pay me for it it still has value correct pretty simple way of

11:39uh of approaching it we use a six -year depreciation um we believe that the

11:44gpus will last in excess of six years but we felt like that was a

11:47fair and reasonable approach to a technology cycle that's moving at this velocity um the

11:53a100s the amperes this year the price has appreciated through the year why is that

11:58i i think it's because one of the things that happens is as more installed

12:04capacity becomes available you have new companies that come into existence that have new use

12:09cases that have different size models that are trying to uh build new commercial ventures

12:15that maybe have been blocked out of the h100s yeah and never had an opportunity

12:19to run on that i mean to make a very simple example for the audience

12:22like when you trade in your iphone after three or four years you're like who's

12:27going to use an iphone 12 and it's like have you been to south america

12:32or africa where you go to the store and you buy an iphone 12 or

12:36you buy the pixel 7 and it costs 50 dollars that's still got great life

12:41left in it absolutely yeah you know and so look you know we we find

12:45these amazing use cases new companies that have come into existence or existing companies that

12:51have integrated new models into their workflow that are able to use the amperes and

12:58so they keep buying any gpus that we have available and once again you know

13:03the concept that a gpu is no longer relevant or commercially viable after 16 more

13:0918 months or two years yeah that's farcical it just it just doesn't make obviously

13:12farcical i think sometimes people get caught up in moore's law or in just how

13:17fast our industry is growing yeah and that there's so much at stake that big

13:22companies are demanding the most recent products that doesn't mean that the lifespan has gotten

13:29shorter it means the opportunity and the surface area of the opportunity has gotten much

13:32larger yeah uh one of the things is is like you know the the uh

13:37the the industry has gotten so much attention for the unprecedented scale of capital that

13:45is coming to bear on this yeah and because of that there tends to be

13:50a incredible focus on the companies that are building on these most advanced chipsets and

13:59the truth of the matter is is you know even within those companies they have

14:04a long tail of useful life to provide inference horsepower to work on other experiments

14:10to do less bleeding edge activity but still needs to be done and yeah i

14:15mean rendering comes to mind as well uh or yeah we're making images on nano

14:19banana like there will be a use for it there is a moment in time

14:23where maybe the compute to power ratio doesn't make sense my my expectation is is

14:29obsolescence will be defined by the moment in time where the power in the data

14:38center for me will be able to be repurposed for a higher margin than the

14:44existing infrastructure provides. And, you know, like I said, I fully expect this infrastructure to

14:50last in excess of six years.

14:52But the standard in the space has really been used with one exception, which is

14:59Amazon, which is, yeah, it's six years.

15:01That seems like the right schedule.

15:03I'm not making it up.

15:04That's what everybody's using. Yeah.

15:05And the energy cost is the opportunity because, hey, it's just we need that space.

15:12There's a better reward here.

15:14And that might get resold that hardware to somebody else who wants it, a hobbyist

15:19or something. It's yeah. Or it could be sent someplace else where they have more

15:23capacity when they can repurpose it there.

15:25But I kind of feel like, you know, we'll deal with that part of the

15:30business when we get there.

15:31What I know right now is it is extraordinarily profitable.

15:36It's very creative to my company to continue to keep the infrastructure that's been up

15:41and running, that's been on these long term contracts.

15:43And as it rolls off, as it's been in use for five years, you know,

15:47as it becomes available, I am still able to sell it at a higher price

15:50than it was at a year ago.

15:52There's competition now when you were buying these from Jensen back in the day.

15:57Yeah, you could buy them and have them shipped, I would assume, within 30 days

16:00or less. Nowadays, what's the wait like even for you, a loyal old customer?

16:06And is there a bit of a battle?

16:08Is there politics to who gets the servers?

16:11Like you see some like very big names talking about they got to get an

16:15allocation. Is it still a little bit crazy?

16:18What's it like to be in that category, having to buy something everybody wants?

16:22Look, you know, I think of it as an affirmation of the business that we're

16:27in, right? Like the fact that we are attracting competitors, the means that the business

16:32is healthy. And there's a lot of people trying to deliver this service, because the

16:36need for this infrastructure, the need to integrate the infrastructure, you know, into the software

16:42layers to deliver it to artificial intelligence, either at the model level or the inference

16:48level or the application level or whatever, you know, level of the five layer cake

16:51that Jensen's, you know, focused on.

16:55The fact that there are more people coming into this, it doesn't discourage me.

16:59Yeah. As far as getting access to the GPUs, we show up like everybody else

17:03with a, you know, we'd like to buy, here's a PO and we're ready to

17:07pay. The one. What's the wait time like?

17:11And is it just really competitive or not?

17:15Because I talked to Jensen about, he said, I said, how do you manage all

17:18these like big egos and names and companies trying to buy stuff?

17:21And he said, well, they order it and we give it to them in the

17:24order in which they order it.

17:26Does it really like that?

17:27It really is. Right. Like, you know, he doesn't want to be in the position

17:32of playing favorites or like that, that just seems like a bad place to be

17:36with your clients. Or auctioning them off.

17:38Yeah. So you imagine. Yeah.

17:40That would, that, that. That'd be crazy.

17:42Yeah. I don't, I'm not sure that would be good for the long -term business.

17:46No. Yeah. So, so our, our, our approach is.

17:49You might get some sovereigns coming in and saying, I'll pay double.

17:51Yeah. They do that with Ferraris too sometimes.

17:54I guess these are the Ferraris of computing, right?

17:56In a way they are.

17:57Yeah. But I mean, Bugatti's our, our approach is to work with clients across the

18:03entire space to find opportunities that are really interesting companies that can fit into our

18:09contraction, contracting requirements. the way we're going to be able to go out and structure

18:14the debt that we require in order to go out and build infrastructure at this

18:19scale. How does all that debt work?

18:22That is something that you guys specialize in.

18:26Corporate debt. I'm in the venture business.

18:28People are like, why should I be in venture when corporate debt pays so well?

18:32Corporate paper is so huge.

18:33I'm curious how this fits in and what interest rate people are paying on a

18:40billion dollars in infrastructure. Where do they pay on that?

18:44Yeah. So CoreWeave has really been the innovator around a lot of the financing engines

18:53that have come to bear on this.

18:54We did the first GPU -based loans.

18:58And I think it's important, or I'm going to try to explain this in a

19:01way people can understand. So what we do is we go out and we find

19:07a client. Let's use Microsoft.

19:08You brought them up before, right?

19:09And Microsoft comes to us and says, we'd like to buy some compute for you.

19:13And we say, okay, great.

19:14We're going to sign a contract.

19:15Once I have a contract in hand, then what I do is I create something.

19:19It's not a particularly creative name.

19:21It's called the box, right?

19:23And what I do with the box is I take my contract with Microsoft and

19:26I put it in the box.

19:27I go to Jensen and I buy the GPUs.

19:30I put it in the box.

19:31I take my data center contract.

19:33I put it in the box.

19:34And now the box governs cashflow.

19:37And it has a waterfall of cashflow that comes into it and goes out of

19:41it. And so the way it works is then I build the compute and then

19:45I deliver the compute to Microsoft and they pay the box.

19:48They don't pay me. Right.

19:49It goes into the box.

19:51And the first thing it does is it pays the data center.

19:54It pays the power bill.

19:55It pays the interest and the principal.

19:58And then whatever's left flows back to us.

20:01Right. And so it is an incredibly well -structured, time -tested, pressure -tested vehicle to

20:09be able to borrow money against client paper and all of the other collateral around

20:15the deal, which is why CoreWeave, which is a company that many people haven't ever

20:19heard of, was able to go out and raise $35 billion in 18 months to

20:24build infrastructure at scale. But what's important to understand is the economics in this box

20:30are such that within two and a half years of a five -year deal, we

20:36have paid for everything. The principal has been paid off.

20:40The principal has been paid off.

20:41The interest has been paid off.

20:43The return into the box is such that we are able to generate returns to

20:48our company at the box level, which gives the most sophisticated lenders in the world,

20:55whether it's banks or private equity funds or whoever, confidence that they're going to be

21:06able to achieve the one rule of lending, which is give me my money back.

21:10Yes. And so it's better when that happens.

21:12So they look at this box and they're like, wow, we're really confident we're going

21:15to get our money back.

21:16And maybe they want 10 boxes.

21:17That's correct. And if any one box goes upside down, you're going to deal with

21:23it and it's not as acute.

21:25That's correct. And they don't cross -pollinate.

21:27They don't cause a contagion across the boxes.

21:30They're all independent and discreet.

21:32One. And number two is, as you do this and as you show the lenders

21:38how this financing tool and how this financing mechanism works, what they do is they

21:44continue to lend you money at progressively lower rates.

21:48And so when you think about our Our cost of capital over the last two

21:53years, we have dropped our cost of capital by 600 basis points.

21:58Wow. It is enormous, right?

22:00And so you're seeing a company that is driving its cost of capital down towards

22:04where the hyperscalers borrow, which will enable us to be able to be competitive with

22:11them over time. And we have been extremely militant and diligent about feeding, watering, and

22:19caring for those boxes so that we continue to have access to the capital markets

22:24in a way that allows us to build and drive our business.

22:26It means you have to say no to maybe some people who want to be

22:29in the box? Yeah. So we look at some deals and we're just like, they

22:34want to buy GPUs for a year.

22:36And I look at it and say, that's not a deal that I can do

22:39because it's too short for me to amortize the expenses.

22:44And so I won't do that, right?

22:45And they can go to another provider, who maybe wants to take that risk on,

22:48who has extra capacity. Absolutely.

22:50But our business is really built around the risk management of being able to get

22:55to scale. Because in my mind, during this period of disequilibrium, during this period where

23:02there were not enough GPUs in the world to provide the compute for all of

23:08the different use cases in artificial intelligence, the part that's important for me and for

23:12my company is to get enormously large so we can drive down our cost of

23:17capital, so that we have information flow coming in from all different parts of the

23:21market, large language models, high -speed trading, search, all of these things.

23:27And they're feeding information back into us that is letting us know what the next

23:32product we need to build is or where they need help scaling or what type

23:38of compute they need. And all of that information flow is incredibly valuable to us.

23:44What can you tell us about demand?

23:45There's been reports of, hey, maybe the Oracle Starbase thing with OpenAI has been downsized

23:53or maybe not. And then other folks, Microsoft is going big and Google is going

24:00big, Meta is going big.

24:01And those people obviously have massive cash flow.

24:03Apple seems to be MIA, they don't seem to want to play.

24:07You've named a lot of really big companies with really big balance sheets that have

24:11the capacity to drive a lot of demand.

24:13Look, I have been truly steadfast in this for years now, for four years.

24:21The depth of the demand for the service we provide has been relentless and overwhelms

24:28the global capacity of the world to deliver enough compute to enable all of the

24:35demand for artificial intelligence to be stated.

24:39And that has been, we have been relentless about that.

24:42Sounds like Nick's tickets during the Patrick Ewing era.

24:44Yeah. They got up to 50 ,000 people on the wait list.

24:48So if magically the wait list went away, if the constraint went away and we

24:53just had a large amount of GPUs available, a lot of energy available, a lot

24:59of data center available, how much capacity would just all of a sudden come out

25:05of the system or would be deployed, I should say.

25:08So remember how we build our business through this box.

25:12And it's a five year box.

25:14So if we had an air pocket, if demand were suddenly to disappear because of

25:21a technology breakthrough, because of a war, anything, right?

25:25Like the why from a risk management perspective does not matter.

25:29You have to prepare your company for the what happens if it happens.

25:33And so by entering into these long term contracts and entering into contracts with counterparties

25:40that have large balance sheets, you are or we are protecting ourselves and our lenders

25:46so that we are confident and they are confident because you can see how confident

25:51they are by the rate that they're charging us continuing to decline, that they're ultimately

25:56going to get their money back.

25:57And that is the one rule of lending.

25:59And so, you know, if.

26:02But just in terms of the capacity, if you were unconstrained and NVIDIA Jensen says,

26:06hey, order as many as you want, what would happen?

26:09So it's also important to understand the constraints aren't just GPUs.

26:15Right. Electricity. It's power shells.

26:18It's memory. It's storage. It's networking.

26:21It's optics. All of the things.

26:23And there's there's various throttles that will limit the memory is a throttle right now.

26:28Oh, yeah, it is. Oh, yeah, it is.

26:30Why? How did memory become the throttle?

26:32If memory and it has historically been a cyclical business.

26:40Right. We have seen these waves of demand driving up the cost for memory and

26:45then it collapses and then it drives it up.

26:47It's a very boom and bust business is cyclical in its nature because the fabs

26:52are so capital intensive that people invest in the fabs, build a ton of capacity

26:58and then overbuild if there is any type of turndown.

27:02And we've seen that cycle again and again.

27:05What's happening right now is the confluence of two things.

27:08Right. One is, is with all the demand for artificial intelligence and the corresponding demand

27:16for compute and the ancillary services around the GPU, the demand is through the roof.

27:22That's number one. Number two is, is that there was probably an investment cycle that

27:27needed to happen back in 2023 that would have brought on the necessary fab capacity

27:33to be able to serve.

27:35Impossible to predict what just happened just with energy.

27:38It's impossible to predict what just happened.

27:39And now people are chasing energy.

27:41The data centers are going where the energy is.

27:44It's not based on real estate.

27:45That's correct. It's based on where is there some wind.

27:48And any time you have a very, not every, any time, but many times when

27:53you have a capital intensive business, like, you know, building fabs, you will get this

27:59boom and bust cycle, just like in energy.

28:01They overbuild. Yeah. And then, you know.

28:03Fiber. Yeah. I mean, there's, there's, there's a lot of examples of that.

28:08Our approach. In some ways, when you look at that, it's a beautiful aspect of

28:12capitalism that we're able to have a boom bust cycle, that we're able to weather

28:18it. Right. If you think just that capitalism from first principles, something like that happens

28:22and we have too much fiber, it creates an opportunity for Google to buy it

28:25all up or the next person.

28:27Listen, the, the, the, um, um, you know, it, it does, it does a lot

28:31of things having a boom bust cycle.

28:32It clears out the underbrush.

28:34Yeah. The strongest companies will be able to survive and take advantage of that.

28:37And it sows the seeds of future business.

28:41The other thing that it does is you put that infrastructure into the ground.

28:45You put the fiber into the ground, which became the backbone of how, you know,

28:50we watch movies every day and how we, you know, uh, communicate and how we

28:55help on a zoom and, you know, COVID and all of these things were based

28:59on that infrastructure that was available to be consumed.

29:03Yeah. People don't recognize this fact.

29:05If you, the, the premise.

29:07of YouTube from the founders who I knew, Chad Hurley and his other partner, they

29:12basically had the realization at this curve, storage is coming down so quickly, we could

29:18offer free unlimited uploads and bandwidth is coming down.

29:22So I guess we don't have to charge people for sharing a video online.

29:26Before that, if your video went viral, people are going to have their minds blown,

29:31but your server would turn off and it would say, this person needs to pay

29:35their bill because they were getting charged for carriage by the Megavit going out.

29:41Yes. I mean, look, and the business models change and evolve and like you said,

29:48Moore's Law and certainly Jensen will talk about the fact that what is going on

29:53within the accelerated compute space dwarfs Moore's Law, right?

29:59And all of that is going to lead to more opportunity to build more companies

30:05that are going to do things like YouTube did, which has really changed the world.

30:09Yeah. Yeah. I mean, the concept that, I don't know if it was like a

30:13million hours being uploaded every hour or minute, but at some point, Susan Wojcicki, rest

30:19in peace, said, told me just like how much was being uploaded every minute.

30:25And it made no logical sense until you realized, well, there's 3 billion people, 2

30:31or 3 billion people in the service and 1 % upload or 0 .1, 10

30:34bips upload. And it's like, okay, one in a thousand people upload.

30:38It's a big, it's a big denominator.

30:40I was sitting on a panel with Sarah Fryer, CFO of OpenAI.

30:46And every once in a while, she really puts out like interesting information.

30:55And so she was talking about the cost of a million tokens when ChatGP3 came

31:01out and it was $32 in change.

31:05And now a million tokens cost nine cents.

31:08Yeah. And so you just see like the incredible power of how the capital markets,

31:16how capitalism is fueling engineering and fueling competition.

31:23And it's become recursive now too.

31:24I mean, these models, if you say to the model, hey, make yourself more efficient,

31:28spend less money and lower the cost of tokens, it'd be like, okay, Captain.

31:32Yeah. I don't know if you saw Karpathy's recursive thing last weekend, but it's like

31:37now civilians who've never worked in a language model during computer science are like, I'm

31:41going to try to do something recursive this weekend.

31:43You know, it's one of the things that I, that I talked to, you know,

31:50the other founders about, you know, and it's like when you think about some of

31:57the things that AI does, right?

31:58It's lowering the barrier to operations.

32:02So if you have a good idea or a great idea, you can open up

32:06your model and you can tell your model, you can vibe code it.

32:11You can do all kinds of different things and create things that never existed before.

32:16That's amazing, right? Like that's bringing down this incredible barrier that kept human creativity contained.

32:22And now all of a sudden this whole new vector of, uh, uh, you know,

32:27medical research or different approaches to, you know, baseball cards or whatever you want.

32:32If you've got a great idea, if you've got a new creative idea, that's the

32:36valuable kernel right now that allows you to, to build new things and to create

32:41new things. And I just think that's incredibly exciting, right?

32:44Like you're bringing the minds.

32:46of eight billion people, a tool that allows them to overcome what was insurmountable for

32:52forever. For humanity. Yeah. It's a bright new future, Michael.

32:56I appreciate you sharing the information with us and the vision.

33:00I am really delighted to have Aravind Srinivas on the program.

33:05Thank you for having me here, Jason.

33:06It's so great. I want to go through three stages in which I fell in

33:11love with your product. The first phase was I could go and pick my language

33:17model if I wanted to use OpenAI, if I wanted to use Claude, whatever it

33:21was. That was like a real unlock for me.

33:24And on the sidebar, I noticed you had done essentially like what Yahoo did in

33:32the early days, finance, sports.

33:35And when I pulled my Nick game up, it gave me a live version of

33:39that. When I pulled my stocks up, it summarized the news in real time.

33:42And I was like, wow, this execution is great.

33:45And I kind of made you my front door to different models and it made

33:49it easier for me checking.

33:50Then you came out with the Comet browser and I was like, holy cow, I

33:55can give this a series of instructions.

33:56Go to my LinkedIn, find everybody from this company, put them into a Google sheet

34:01and boom, you were the first out of the gate with that.

34:03And then just the last couple of weeks, I had been claw -pilled and using

34:07OpenClaw, but you came out with computer and I started using computer and boy, it's

34:13good. It's a really strong start allowing me to do repetitive tasks, very similar in

34:20some ways to co -work from Claude or basically an engineer or developer using it.

34:26So are these the evolution of the company?

34:31And I should think about it that way.

34:32But how do you look at perplexity now?

34:35You have a very loyal fan base.

34:36You're making a lot of money.

34:38I don't know if you disclose it, but I think it's hundreds of millions to

34:41billions. You can tell us.

34:42But what is perplexity in the face of, wow, Claude's having a great run, OpenAI

34:47still doing strong, Grok doing very well, Gemini coming on strong.

34:51There's like six or seven of you and you just happen to be one of

34:55my top twos right now.

34:57Thank you. First of all, thank you.

34:59Thank you so much. Perplexity has always been built for people who are always looking

35:03for the extra edge, the curious people.

35:06So it's very natural that you are one of our flower users.

35:11One common theme for us for the last three and a half years is accuracy.

35:18Perplexity wants to be the company that's building the most accurate AI.

35:22So when you want to give somebody answers, accuracy is very essential for building trust

35:27because only then the user is going to ask the next set of questions.

35:30It turns out it was a great idea to give AI access to the internet

35:34to be accurate. So that's the perplexity ask product.

35:37It turns out it's a great idea for AI to have full access to a

35:41browser so that it can be accurate when you task it to go do something

35:45that you would do yourself on a browser.

35:47Agentec browsing, comment. Now the last phase is it turns out it's a great idea

35:53for AI to be given a full access to a computer so that it can

35:58do whatever you do on a computer on its own, essentially becoming the computer itself,

36:03an orchestra of everything AI can do today.

36:07Every single capability each individual AI model has, be it GPT or Claude or Gemini

36:14or anything else, an orchestra of all those capabilities.

36:18That's what perplexity computer is.

36:20And all these sub -agents that are running inside computer.

36:24are the musicians. The models are essentially the instruments and they're like hundreds of models

36:30out there, each having their own specialization.

36:33Some are good at coding, some are good at writing, some are good at multimodal,

36:36visual synthesis, image generation, video generation, audio.

36:40But what matters is the end output, the music you play.

36:44That's the work AI gets done for you and that's what perplexity computers.

36:48The AI itself is the computer now.

36:50It still lives inside of a browser.

36:53Have you considered giving it desktop root access?

36:56That feels like the next place this is going, but that comes with a lot

37:00of security issues, a lot of trust issues.

37:02As you mentioned, trust is paramount.

37:04Getting the right answer is what builds it, but also not getting hacked and not

37:08having it delete your files.

37:10So how do you think about root access to my Windows machine?

37:14Obviously iOS, they won't let you, but with an Android phone, it would let you.

37:18Yes. So do you have that in the works?

37:20Yes. So we announced something called personal computer, perplexity personal computer.

37:24That's essentially going to take all the trust and reliability and the server side execution

37:29of perplexity computer, but synchronize it with your local computer so that you can use

37:35it from your phone. And we're going to do this with the Mac mini where

37:39you synchronize your computer with the Mac mini.

37:42So that becomes your local server.

37:44All the agent orchestration that has to do with your local private data will run

37:48on that local orchestration loop, that runtime with the Mac mini.

37:52Not on your servers, not on Anthropics.

37:55Exactly. Yeah. It could still ping Frontier models if it needs to with your permission,

38:00but it will be orchestrating everything on your local hardware.

38:05Yeah. And if it needs to run on the server side hardware, if you don't

38:08want very complicated, long running tasks to be running on your local hardware, you can

38:14delegate it to run on your server side computer, which is again, only accessible to

38:19you and you alone. So that way we're going to bring this perfect hybrid of

38:24trustworthy hybrid between local and server side.

38:29And you'll make it easy to do it.

38:30Just be abstracted. You install one executable.

38:33Boom. It's done. It's like open claw for dummies.

38:35Nobody needs to learn how to use it.

38:37Nobody needs to manage API keys.

38:38Nobody needs to manage separate billing across like 100 different services, figure out what you

38:43can give access to and not access to.

38:45We take care of that.

38:47So it's a Steve Jobs way of doing it.

38:49Yeah. End to end integration.

38:50And how do you think about local models?

38:53I have started running Kimi 2 .5 on a Mac studio.

38:58It's not as good as Claude or Gemini or Croc, but you can probably do

39:02about 80 % there for free.

39:04Yeah. Essentially. Yeah. And so that's quite compelling considering some of my other bills, Claude

39:10and stuff were getting expensive.

39:12So do you have one of those you started testing on your local Mac studio?

39:17I assume you have a Mac studio and you're doing this yourself.

39:19Yeah. Or now I don't know if you saw Dell and Nvidia announced a giant

39:24workstation. Yeah. Is it a 3 ,800?

39:27Something like that. Something like that.

39:29It's 750 gigs of RAM.

39:31So what do you think about the desktop going back to workstation slash server?

39:36Yeah. Status? I think it's very promising.

39:40My prediction is that it'll initially start off as a subagent.

39:44So whatever you need to go, like your tax returns, your personal photos, your emails,

39:51your calendar, all that stuff, those local apps, your personal notes, very personal notes, you

39:59could make sure that the models that access those calls.

40:02tokens will be running on your local hardware if you want to if you're that

40:07privacy conscious and more complicated stuff that accesses your data that's already on the server

40:14side example your google calendar yeah your gmail this is personal data still but an

40:21ai runtime can access that through your connector your google calendar connector your google workspace

40:26connector and that could run on the server side because anyway the data is on

40:30the server it's not even lying on your device so that sort of hybrid orchestration

40:35is where we are headed to i don't think it's a dichotomy between fully local

40:39versus fully server uh it's all about choice and anyway when you're on your phone

40:44uh you wanna you don't care actually which server that workload's running from because it's

40:49not going to be able to run on your phone anyway the chips need to

40:52exist on a mac studio or a mac mini and or on the server or

40:56this new dell that's coming out and i i really think the idea of spending

41:01ten thousand dollars on a powerful desktop will appeal to people if it lowers their

41:06right five hundred dollar a month yes claude bill yes this is an incredible savings

41:11plus you get the benefit yes of privacy and not educating the language models on

41:16your personal data yes and it's going to be it's going to be like you're

41:19buying a refrigerator your your internet mode of like the cost for these will eventually

41:25go down yeah but it's not going to feel like you're wasting your money uh

41:30every every home has a lot of other sensors yeah that runs your home that'll

41:36also be part of this orchestration loop yeah so so that's where it gets exciting

41:41because now you can just dictate something to your phone and that can control your

41:46entire home so that's the dream that everybody has and all that orchestration loop can

41:52run on your local hardware no problem and i'm curious what you think of the

41:56operating system what's eventually going to be the operating system of this workstation ai is

42:03the operating system like earlier in the traditional operating system you execute programmatically now you

42:10start with objectives not specific instructions right you come up with a high level objective

42:15go build this website for me that you know takes all the transcripts of and

42:20podcasts and tracks the stock price just before the podcast and after yeah and chart

42:24it for the max seven yeah and and chart it over time you can so

42:28that's the objective but individually it's running a file system a code sandbox access to

42:34the internet it's having like its own html tools and like yeah so i think

42:38that's basically where you know models systems and files and connectors are all coming together

42:44you would think of that as an os except you're operating at an abstraction about

42:49that where you're thinking in terms of objectives yeah and does it need to eventually

42:55become its own operating system in your mind it could be like people could think

43:00about us it's like yeah i have my perplexity computer running all the time whether

43:04it essentially it runs on linux machines right now every server -side computer is a

43:09linux machine yeah so i think mark and recent tweeted this right after our release

43:14that turns out linux computers was the right idea desktop desktop linux computers are finally

43:19going to work yeah i mean they're stable yeah they're customizable and you're not at

43:24the mercy of apple's desire to contain the experience exactly or microsoft surface area as

43:32for hackers exactly you build something rock solid and it does feel like linux might

43:37actually become the correct the eventual winner it may not need to have a front

43:41-end, that's the thing. You could access the Linux machine on your phone, which could

43:47be running iOS or Android, it doesn't matter.

43:50The actual valuable runtime is running on Linux on the server.

43:54You've done great as a consumer company, a lot of love there.

43:59Now I'm starting to see corporations with computer starting engaging.

44:05In fact, you'll be happy to know this, last week I took two people in

44:09my back office and I said, stop working on OpenClaw.

44:12Your job is to do the back office automation at our venture firm only using

44:17perplexity, perplexity computer. And they were like, oh, okay, it doesn't talk well in Slack.

44:25It doesn't have an agent in Slack.

44:27I was like, it will.

44:28I'm going to see Arvind.

44:30I'll talk to him about that.

44:31So we need a really strong Slack connector.

44:34It's already out. It is.

44:35Okay, great. A computer exists as a Slack bot right now.

44:38Okay. That you can add to your Slack workspace on the enterprise plan.

44:41And our entire company works like that.

44:43People are talking more to computer on Slack than other people.

44:47In our first volley, we were sending reports in, but it wasn't interactive.

44:51That's perfect. So now you've got your company going in two different directions.

44:57This incredible consumer run you have.

44:59How many people are using the product every month?

45:01Several tens of millions. So tens of millions of people.

45:04That's very much similar to the trajectory of the Google and Yahoo consumer business.

45:09Now you've got corporate. How are you doing on the corporate side?

45:12Thousands of companies? It's the fastest growing business for us.

45:15It's growing faster than the consumer and revenue.

45:18And things like computer unlock entirely new possibilities.

45:22For example, we've saved more than $100 million for our Enterprise Max customers who are

45:27on the highest tier of Enterprise.

45:29Explain what that is. What does it cost?

45:30$200 a month per person?

45:31So there are two tiers.

45:33One is the Enterprise Pro, which is $40 a month.

45:36And there's the Enterprise Max, which is $400 a month.

45:40And on a computer, after you run out of your credits, you would pay for

45:45the tokens. You pay for the usage.

45:48Are you making money on the $400 a month, $5 ,000 a year one?

45:51Or at this point in time, are people going so crazy?

45:55One thing that Perplexity has is every revenue we make, unlike certain other wrapper companies,

46:00every revenue Perplexity makes has positive gross margins.

46:03Got it. Because we're not just selling tokens.

46:06Right. Most of our revenue is recurring because people are paying a subscription fee.

46:11And because we route through multiple different models, we're very efficient in terms of how

46:16we spend on the tokens.

46:17Because we have all this advantage with drag and orchestration and search, we don't actually

46:22need to blow up the context window of the models.

46:25Yeah. As a result of that, we have positive gross margins on all the revenue.

46:29Every single penny we make, we make profits on that.

46:33Overall, the company is still yet to be profitable, but we're working towards that.

46:36You've had the opportunity to exit.

46:38A lot of rumors. Apple, other people were like, hey, this is a great team.

46:42How many people on the team now?

46:44About 400. Yeah. You've got a very coveted team.

46:47You obviously understand consumer. You obviously understand business.

46:50It's a product -driven organization.

46:52Reports are you declined. But the world's getting hyper -competitive here.

46:57How do you keep up as a 400 -person organization when you've got Sam Altman

47:02over here raising $100 billion?

47:05And then you have Elon putting data centers in space and merging with SpaceX and

47:10Twitter. You have Google with unlimited resources.

47:13Amazon getting in the game.

47:15And obviously, Gemini, a very strong product.

47:18And Google. Google. Really good at consumer.

47:21I think we'd all agree Facebook and Meta haven't figured it out yet, except maybe

47:26for serving us better ads, but they haven't figured out the consumer case yet, but

47:30they'll copy it. They always do.

47:32How do you look at the playing field?

47:34Because the degree of difficulty, this isn't playing checkers or this is like playing against

47:40the 10 best chess players in the world.

47:44That's what you have to do every day.

47:45So how do you think about it?

47:47Long -term and independent company.

47:48Do you think you'll need to join forces at some point?

47:51And why didn't you take the deal?

47:53These deals were incredible that you got offered.

47:55So one advantage we have that all these companies you mentioned don't have is the

48:01multi -model orchestration. We're like Switzerland.

48:04We don't have to have one horse in the race.

48:07If GPT wins, Gemini wins, Claude wins, Lama wins, it doesn't matter to us.

48:13Or even open source models can win.

48:16No problem. And you have them on the service.

48:17Yes. We have DeepSeek and Kimi.

48:19We have Kimi, we have Nemotron, and we have a lot of usage of Quinn,

48:24Alibaba Quinn. Yeah. Silently under the hood.

48:28So for us, that advantage of being able to take the best in each model

48:33and give the user the orchestra of everything they can do, I don't think any

48:37of the companies you mentioned can do that.

48:39Right. Nor would they. Nor would they.

48:41It makes no sense for them.

48:43It would be an admission that all the data centers and CapEx they've built out

48:46still couldn't produce them the best model.

48:50And Dario, CEO of Anthropics, said recently in an interview that models are specializing.

48:57Towards the beginning of last year, people thought models are going to commoditize.

49:01But towards the end of last year, models started specializing.

49:05Even within coding, Claude Code and Codex have very different capabilities.

49:12Our iOS engineers love using Codex.

49:15Our backend engineers love using Claude Code.

49:17Yeah. So even within a specialization like coding, models have their own unique specialties.

49:23And there are many other use cases outside coding where different models are good at

49:27different things, which means the orchestra conductor that has no one model to the horse

49:32in the race can win by providing a very unique value and service to the

49:37customer that each of these amazing names that you mentioned cannot.

49:41And so you're buying tokens wholesale from them, and then you'll charge customers to do

49:48it? Or do you think it's all...

49:49We're going to take care of all their orchestration.

49:52Yeah. So you don't have to manage tokens across different models?

49:55Because I authenticate a couple of my different accounts, my pro accounts, into Perplexity.

50:00But does it... I don't have enough knowledge to know if you're abstracting that and

50:05people can just search across them and it's part of their Perplexity subscription?

50:08No, we're not bundling subscriptions into other AIs.

50:12Yeah. We just ping the models directly.

50:14Got it. What you get in us is the Perplexity or orchestration.

50:18Got it. The harness. Right.

50:20So when models are kind of specializing, there's a bigger value in the one who

50:26knows how to build a great harness that can take the best in each model.

50:30Does it auto -route today or do you still have the dropdown somebody's got to

50:34pick? It definitely auto -routes the best model for each prompt, but we also give

50:38users the flexibility to pick whatever model they want.

50:40What do you think of...

50:42I've seen a bunch of startups hack this together, but doing the same query across

50:46multiple... We built a thing called Model Council.

50:49Model Council, yeah. Yeah. So that's one of the modes and Perplexity where I saw

50:54Jensen say in one of the interviews that he puts the same prompt...

50:58five different AIs and sees what each of them says.

51:00Yes, everybody does that, yeah.

51:02But then you still have to apply a biological compute to read every answer and

51:07then figure out where they differ.

51:08It's like talking to five lawyers about your trust or your...

51:12Five different doctors. Five different doctors and they're trying to figure it out.

51:15Exactly. It's dumb. So the model council is a feature we built where it will

51:19not just give you the answers of each model, but it will tell you exactly

51:22where they agree, where they disagree, and where the nuances are.

51:24And that's in the interface?

51:26Yeah, it's there. I didn't know it was there.

51:27It's there. I mean, you release product at a pretty great cadence, huh?

51:32Yes. Where did you learn that?

51:33And what's your philosophy of shipping product?

51:35Our philosophy is like speed is our moat.

51:38Like, you know, again, one of the things that big companies cannot do is move

51:41at the speed we do, serve customers at the speed and quality.

51:44It's very hard to maintain quality, speed, and trust at the same time.

51:48Yeah. Like Apple takes a long time to ship anything because they're very worried about

51:52people not trusting them. Yeah.

51:55And so some companies are bureaucratic and they just take forever to ship something.

51:59They don't maintain what they ship.

52:00They may make a big deal about an event, but nobody even knows how to

52:04go and use that feature.

52:05Yeah, they get abandoned. Exactly.

52:06So perplexity has those advantages of being very small.

52:09And towards the end of last year, we found that like AI coding tools have

52:14made it much faster for us to ship things, which is honestly one of the

52:18reasons why we built a computer because now even non -engineers are shipping code here

52:22by just bringing a Slack bot and asking it to fix bugs.

52:25So the iteration has just been like exponential.

52:29The moment I had where I became claw -pilled was when I was working with

52:35it and I was like, hey, I want to build my network.

52:37I know these 20 people in Japan.

52:39I had dinner with them during my recent trip.

52:41I want to know who they know.

52:43So check out LinkedIn and other things and who they're associated with and make me

52:46like a mind map of it.

52:47And then the next trip, I want to meet with the next circle of those

52:52connections. So I started asking, it's like, okay, I got the results.

52:55I was like, great. And it said, where do you want me to put them?

53:00And I was like, well, where can you put them?

53:02And it said, well, I can put it in a Google sheet.

53:04I can put it in a Notion table.

53:05I can put it here.

53:05I can give you a PDF.

53:06I can give you a CSV file or I could write you a CRM.

53:10And I was like, yeah, sure.

53:12Make me a CRM system.

53:13And it made a CRM system.

53:15And I think that becomes, I think maybe one out of a thousand people working

53:20with AI have had that experience.

53:22Maybe it's one in 10 ,000 where your agent says, I'll make you bespoke software.

53:27Yeah. Have you had that yet?

53:29And do you see that as a part of computer that when a person needs

53:33a spreadsheet, you don't launch Excel or Google sheets, you just pop up a spreadsheet?

53:40Yeah. Well, we have a board meeting tomorrow.

53:42Okay, I'll come. I'll pitch it to the board.

53:45Sure. Our computer made the memo.

53:49Oh, wow. Yeah. And we had a partner meeting to pitch a partnership idea.

53:55And earlier we would have a design team do the whole deck.

53:58Yeah. Computer just one -shotted it.

54:01I had a press briefing with a bunch of journalists.

54:03My comms person would usually...

54:05Sorry about that. Oof. Brutal.

54:07And then my comms person would usually give me a memo.

54:10A dossier. What to say.

54:11Yeah. Computer one -shotted it.

54:13So... It's crazy. It's crazy.

54:15The context is so good because the memory is getting better.

54:19Yeah? Yeah. So it's like, I know that journalist from the last time.

54:23Yeah. I know the board meeting.

54:25Yeah. I have all the previous decks.

54:27Yes. When did that happen?

54:30I think it happened with Opus 4 .5.

54:33The Anthropic. 4 .5, yeah.

54:34It was 4 .5. That was an inflection point.

54:36But... models were started being amazingly good at orchestration and reasoning and tool calls and

54:43cloud code brought in this new idea in ai that everything can happen inside a

54:48sandbox a console a terminal with access to tools where tools are just command line

54:55tools yeah they don't even need to have graphical user interface so when you did

55:00that and when you organize around files and sub -agents and skills and clis the

55:06models started becoming becoming very good at handling the context so the context window no

55:12longer became a problem it just put whatever necessary into the context whenever it wanted

55:16to and dumped them away when it wanted to yeah and that made it like

55:20suddenly so good at doing very long orchestration tasks yeah it's it's pretty crazy i

55:27have every episode of this week in startups all the transcripts and then all of

55:32all that was one of the tasks i did by the way i can send

55:34it to you yeah i asked it i want you to download every all in

55:37podcast yeah uh since the beginning and i want you to take a mention of

55:43all the public companies they mentioned during that yes i want you to have a

55:48histogram of the accounts and i also want you to chart it across time and

55:52then i want you to analyze the impact on the stock price and the sentiment

55:56of what we said exactly and it did like it clearly said are we moving

55:59stocks around google's stock going up yes prior to that you guys were talking a

56:04lot about google yes and clearly i said i made a bet publicly on the

56:08thing i said i am buying a bunch of google because i believe even though

56:12they're behind it's because they're too precious you were kind of mentioning a company that

56:16might be too precious at times it doesn't release yeah i was like that's that

56:19company they need to release more yeah and uh i told sergey i was like

56:23like give us the good stuff yeah he started giving us the good stuff it

56:28literally gives you the time stamps of every single and then i can go click

56:31on it and actually hear exactly that moment yeah sweet yeah so that's when that's

56:36when i was like damn like this i would have had somebody do this as

56:40a week -long project it would have been 10 hours a week of researcher i'm

56:45experiencing the same thing when i do research notes i've created my own uh like

56:52mega prompt yeah and it will go and like tell me where you worked before

56:58and who's in your circle who your competitors are who your friends are blah blah

57:02blah and then go far i try to find old podcasts is one of my

57:06secrets if you're an interviewer watching i try to find what was the person talking

57:10about five years ago ten years ago and then over ten years ago and i've

57:14gone into interviews now with michael dell and talked about things he was talking about

57:17in the 90s yeah and it finds me some ancient stuff like you would pay

57:22a researcher a producer you know seventy thousand dollars a year eighty thousand dollars a

57:27year to do this and they would have done a third of the job in

57:31ten times longer yeah it's really gotten weird just in the last six months what

57:36do you think the next six months looks like i think the the dream that

57:39what we are going to try to do is help businesses run as autonomously as

57:44possible you know everybody talks about this ai is going to create this one person

57:48one billion dollar company some people say it's already happened because people pay researchers like

57:54one billion but it's not truly moving the gdp by one billion it's not truly

57:58creating new value so the best way to do that is to actually help a

58:02small business people who would otherwise drive ubers for like yes extra passive income to

58:08like buy like a mac mini set up perplexity personal computer and run their business

58:13on that or like run around the server doesn't matter uh and actually make real

58:18money yeah hundreds of thousands or even millions a year and uh grow it have

58:24computer go and run your ad campaigns on instagram or google i mean integrate with

58:28scm and seo tools find new users and uh integrate with stripe charge them ship

58:35new features have your own like intercom integration for customer support and like have this

58:41all working well you can be sipping wine in napa that's the dream that you

58:44know so it feels awesome to say everybody thinks yeah it's already there it's not

58:48there yet someone has to do that hard work yeah that's what we want to

58:52do yeah it's a great vision because when i watched startups 20 years ago there

58:58were so many checkboxes they had to do i have to find an office space

59:01i gotta put up a bunch of servers i gotta hire an hr firm i

59:06gotta hire a pr person all this stuff and now i talked to young founders

59:11they got a three -person team they've come out of a16z my program launch accelerator

59:16whatever it is y combinator and i'm like okay you raised a half million you

59:19raised a million who are you hiring and they're like um i don't know if

59:23we need to hire anybody i'm like if you could hire somebody would you hire

59:26they're like well i do my own hr i have this partner and there i'm

59:31like how are you doing hiring anyway and they're like well i put out an

59:34ad and then uh it sorts and ranks the candidates and then it emails the

59:40top 10 ask them a bunch of questions and then i meet with the last

59:42two and i'm like that's what a recruiter did like the entire recruiting job has

59:46been abstracted and like a tool like computer is gonna make that even faster work

59:52to do a lot of connectors a lot of specific workflows people don't want to

59:58like learn how to write like you know essay long prompts you know it needs

1:00:02to be so quick and fast and autonomous you just set it up and done

1:00:05and you have an idea you can turn it into a business and start making

1:00:08money yeah it's it's an incredible future uh and it feels like it's right here

1:00:13do you how do you think about job displacement because you're actually making the tool

1:00:18that enables people yeah to be a solo entrepreneur and get to a million in

1:00:22revenue but it's also the same tool that doesn't require them to hire and we've

1:00:25had this debate a million times on the podcast do you i'm wondering if like

1:00:31me you have moments where you're like oh my god this is really terrifying yeah

1:00:35a lot of people are going to lose their jobs really fast yeah and then

1:00:38oh my god you can learn any skill you want and all the things that

1:00:42were hard are now easy yeah i go back and forth i'm 70 80 yeah

1:00:47super positive about this but i do worry about like 20 of the time i'm

1:00:51a little worried yeah where do you sit i mean america has always been about

1:00:54like entrepreneur entrepreneurship right like yeah we've been about like trying to build new things

1:00:59discover new things go explore uh i think this whole like henry ford came and

1:01:05built factories and brought in jobs and things like that and like put people into

1:01:09a box but uh i think the reality is people most people don't enjoy their

1:01:15jobs they're doing it no they hate them exactly so there's suddenly a new possibility

1:01:20a new opportunity to go use these tools learn them and start your own mini

1:01:24business and if it pays for your needs for your or multiple years and lets

1:01:29you have a high quality life and good work -life balance and true feeling of

1:01:34agency and ownership and passion to like get your ideas out there i think that

1:01:38is even if there is temporary job displacement to deal with that sort of glorious

1:01:43future is what we should look forward i i think you're exactly right if there

1:01:48will be some displacement but then there's also gonna be so many opportunities open up

1:01:51and it requires the individual to not be passive exactly they have to be rugged

1:01:56individualists they have to be resilient yeah and they have to be resourceful and i

1:02:01think once you start playing with these tools that's what happens exactly you you all

1:02:05of a sudden feel like it brings out the best in you if you truly

1:02:07are in a good space yeah yeah i and then today uh comet for ios

1:02:13is out yeah i'm a comet super fan i required everybody you were nice enough

1:02:18when i i emailed you i was like can you send me some licenses you

1:02:20you don't may not remember you sent me a bunch of licenses i said everybody

1:02:23put this on because it was 300 a month when you first came out with

1:02:26the common browser now it's free i think for all users highly recommend it highly

1:02:32recommend getting a pro account it's only 20 bucks a month to get into perplexity

1:02:35which is a joke so you can get on board for nothing less than a

1:02:39dollar a day but what does ios allow me to do and and how does

1:02:43it connect to computer because that's another thing i'm having yeah cloud code uh computer

1:02:50there's not a good enough integration with this mobile device yet yeah so computer is

1:02:56already on the perplexity app so you can just toggle the computer and start using

1:02:59it uh comet's uniqueness and perplexity for the company uh and and and the strategy

1:03:05is the fact that you can control the browser so the browser also becomes a

1:03:10tool for computer yeah just like your google workspace and all these other things uh

1:03:16until the whole world is organized around clis and tools yeah there's still a lot

1:03:21of tasks we have to do manually on the web on the browser open tabs

1:03:25fill up forms click on things upload stuff all that stuff if you want to

1:03:29automate you need a browser you need an ai that can natively control the browser

1:03:34so that is comet and that's why no matter how many other tools in the

1:03:39market exist like open claw or like clock co -work yeah executing tasks on a

1:03:44browser on the server side along with all the other things is something uniquely perplexity

1:03:49can do yeah my dream is that you'll create an android app that roots my

1:03:55android phone yeah and that you just take over and see everything because one of

1:03:59the blockers i have now is a some of the websites have gotten a little

1:04:04persnickety yeah i don't want to mention too many but reddit linkedin yeah and like

1:04:10they're just i i am a great reddit user i'm a great linkedin supporter but

1:04:16sometimes like i need to get my inmail yeah from my linkedin and i just

1:04:21need to you know find seven people a company i is there going to be

1:04:25a solution between the linkedin and reddits of the world and the claws and perplexities

1:04:31is yeah how is that i mean the negotiation going you don't have to speak

1:04:35about any specific ones unless you want to yeah but it feels like there's got

1:04:39to be a solution and i'm willing to pay for it as a user i'm

1:04:42willing to play reddit to allow my bot to show up and behave properly yeah

1:04:46well i i cannot speak about any particular company but yeah we are happy to

1:04:51work with anyone right so um i think with comet our idea is to give

1:04:56people the flexibility to set things up on their own yeah and uh any um

1:05:01official apis that anyone's willing to offer we're always happy to put that as part

1:05:05of computer here's what i think should happen let me see if you agree um

1:05:10and this is for steve huffman at reddit i go on reddit i do a

1:05:17pro account for 20 bucks a month and when i do that i can authenticate

1:05:21whatever tool i want um to do a series of well -behaved things a certain

1:05:28number of times a day yeah so it's not unlimited i'm not going to scrape

1:05:31the site but i would like it to just let perplexi or computer go and

1:05:37just tell me hey what are people saying on this week in startups and all

1:05:41in subreddit summarize it for me so i get the customer feedback and i would

1:05:45literally name my uh agent and i would say i it won't post on my

1:05:51behalf it won't vote on my have just needed to do a couple of little

1:05:54read -only things this would be an easy solution or linkedin i would like if

1:05:58you i have i already pay linkedin like 50 bucks a month like they should

1:06:01just let the 50 a month one work with computer yeah absolutely i mean okay

1:06:06this is for satya nadella let linkedin work with perplexity and the other players and

1:06:12we'll pay you extra perfect it's a revenue stream don't you think api access for

1:06:17customers is a revenue i think so i think so i think i think fundamentally

1:06:20giving users a choice and setting it up as a win -win for both the

1:06:25business and the user yeah is where the world should head to yeah and and

1:06:28i would say the same thing applies to any any website in the world like

1:06:32if you want an ai to use it on your behalf it should be okay

1:06:35for because that's what the user wants i mean i have a paid new york

1:06:39times subscription like let me go in there and do you know whatever 100 searches

1:06:45a day a week a month whatever they choose but that would make the subscription

1:06:49that much more sticky exactly uh all right arvin love the product anybody at home

1:06:55it's just tremendous go learn computer and get the comet browser it has changed my

1:07:02business for the last two years love the product and uh we'll have you back

1:07:06soon when you launch your operating system and come up with your own server and

1:07:10desktop server but business is the focus yeah yes all right great scene thank you

1:07:14we have an amazing guest arthur manches here the ceo of mistral ai how are

1:07:20you doing sir great thank you for uh and so you're here at uh nvidia's

1:07:26big conference big announcement you're going to be working with nvidia to build models uh

1:07:33to open source them what is the uh big announcement here well we're announcing that

1:07:38we are going to be training the next generation of frontier models with uh with

1:07:42nvidia um it's something that we've been doing uh before with nvidia with mistral nemo

1:07:47something we did like 18 months ago and the point for us is really to

1:07:50be able to produce the best open source models out there so that we can

1:07:54actually use those assets to specialize them through products that we do for our customers

1:07:58like forge that helps us customize the models for the enterprise we work with in

1:08:03engineering in physics and science uh in making them better at certain languages when we

1:08:07work with governments etc and and mistral obviously based in uh france you're the leading

1:08:14ai company there what's it like running the company and building a large language model

1:08:19in europe obviously there's regulations and all kinds of considerations privacy the french are known

1:08:25for protecting privacy in the united states we're known for taking it away how is

1:08:29the landscape there and what do you have to deal with there that maybe you

1:08:33wouldn't have to deal with in america what's the pros and the cons let's say

1:08:36first we have 25 percent of our business in the us uh and 25 percent

1:08:40of our researchers are actually here so i actually spend a lot of time here

1:08:44as well as in france as well as in the uk in singapore where we

1:08:47are so of course it's it's different markets uh it's markets where you have language

1:08:52which is a topic uh where there's much more manufacturing manufacturing is a bigger piece

1:08:57of the cake than it is here uh and i'd say the our strength has

1:09:01been to also work with european companies that are a bit lagging behind uh and

1:09:06that wants to adopt the technology to to leap forward and we've been able to

1:09:10do that to a forward deployment engineering engagement for our Forge product, for our studio

1:09:14product that allows to deploy agents that do end -to -end automation.

1:09:17But on top of that, the thing that we have announced today, like Forge, is

1:09:21something that is actually being used today with customers in the US because they come

1:09:26to us with needs for post -training, for making models specifically good at financial services.

1:09:31And what's happening is that we have this product and we can bring the models

1:09:34to specialize them as well.

1:09:35And so your belief is specialized, verticalized models healthcare, finance, engineering, different verticals will win

1:09:44the day or a global model will win the day that does everything?

1:09:48Well, you need general purpose models to do the orchestration parts, et cetera.

1:09:52But at some point, enterprises sits on a lot of intellectual property, on a lot

1:09:57of signals coming from physical systems, from factories, from tools.

1:10:01And it's actually not trivial to connect those systems, to connect those data to models

1:10:07that are closed source. If you have open models, you can actually add new parameters,

1:10:11you can make a lot of deeper things that you cannot do with closed models.

1:10:15We can also, and that's something that we do, not only do we work at

1:10:18the model side, but also at the orchestration side, we sit with subject matter experts

1:10:23to understand their needs and we build business applications that are fully bespoke to their

1:10:26needs by modifying the models, but also modifying the harness on top, et cetera.

1:10:30So we believe that eventually, building on open source technology is a way to save

1:10:35costs, is a way to have better control, because you can see the thing on

1:10:38every cloud that you want, on your hardware if you want, you can deploy it

1:10:41on the edge if you want.

1:10:42And eventually, from a customization perspective and from leveraging your decades of IP that you've

1:10:48been accruing in financial services, in heavy manufacturing, like companies like ISML, for instance, they

1:10:53do benefit from working with us because we take their data and we build models

1:10:56that are specifically good for their purposes.

1:10:59this training data, using experts to come in and refine a model.

1:11:04Most people don't know this business that well, but this has become a very large

1:11:09part of the industry. Obviously, Scale .ai was doing it.

1:11:12They went to Facebook, lost a lot of the customer base who didn't want to

1:11:16send their data, I guess, over to Meta.

1:11:19We're investors in a company called Micro One that's doing pretty well in this space.

1:11:22There's other folks doing it.

1:11:24Explain to the audience what you're doing specifically for companies and how this training works

1:11:30in a verticalized way and then how you silo that data because if you're working

1:11:34with one customer in aerospace or fintech, they might have a need set, but they

1:11:39may not want that training to go to a competitor.

1:11:42I can use a few examples.

1:11:44I think overall, the data segregation is super important and the way we have solved

1:11:49that is through a portable platform.

1:11:50So our technology is a set of services, a set of training tools, a set

1:11:55of data processing tools that I can take and that I can put on the

1:11:58infrastructure of my customers. So suddenly from an IT perspective and when we talk to

1:12:02the CIOs, they realize that from security perspective, the flow of data doesn't go, there's

1:12:07no data flow coming back to Mistral because everything stays there.

1:12:10Now, the way we then use that technology that has been deployed is that we're

1:12:15going to be working with the teams that is doing image scanning and default detection

1:12:20with ISML, for instance. And we're going to be sending forward deployment engineers, scientists, they're

1:12:26all PhDs, they know how to train models and they spend some time with the

1:12:29subject matter experts that can explain how an image is being detected, how do you

1:12:33detect defaults, et cetera. And based on that, we're going to work out what kind

1:12:37of data needs to be used to train the models that it's going to solve

1:12:41the task in itself. And so we send the technology, typically we send a little

1:12:46bit of scientists because you're going to be used to be used to be used

1:12:48to be used. you do need that expertise transfer and that knowledge transfer in between

1:12:52our teams and the vertical experts.

1:12:54And then we make sure that eventually our team no longer needs to be there

1:12:57to retrain the models, to get more data access, et cetera.

1:13:00So that combination of data segregation, expertise transfer, knowledge transfer is the one thing that

1:13:06makes us quite unique and allows us to serve the most critical use cases, the

1:13:10most critical processes in industries that actually need to take their data and put it

1:13:14into models for it to work.

1:13:15Yeah, this seems to be, once the entire open web, what was available legally, gray

1:13:23market, et cetera. I wouldn't have your comment on that controversy, but we kind of

1:13:28exhausted what's in the open crawl, yeah?

1:13:31We have. And it's time to actually either make synthetic data or actually use experts.

1:13:38Do you believe in synthetic data and where does that work and where does it

1:13:41fail? We use synthetic data as a way to warm up the models.

1:13:45It's a way to actually be quite efficient at the beginning.

1:13:48If you have a large model and you want to train a small model, you

1:13:50will use your large model to process and to produce a lot of synthetic data

1:13:55at the beginning. But eventually, you do need to have human signal.

1:13:59So the human signal is something that is always a bit costly to acquire because

1:14:03you need to talk to the experts.

1:14:05They need to give feedback to the machines.

1:14:07And so at the beginning, synthetic data allows you to do the compression, to further

1:14:11compress the models. At the end, you do need to go and get data that

1:14:14is produced by humans. So yeah, it's a way to have, it's mostly an efficient

1:14:20way of training models to have bigger models that are used as teachers for smaller

1:14:25models, but it's not enough.

1:14:26And so you also need human signal.

1:14:27Arthur, we've seen an incredible explosion.

1:14:31We're sitting here on AO52 after OpenClaw, the year of our Lord, 52 days.

1:14:39When you first saw OpenClaw and saw the reaction of hackers, founders, startup CEOs, just

1:14:47the amount of energy and it racing to the top of GitHub with the most

1:14:51number of stars and likes and all these contributors, what did that say to you

1:14:56as an executive in the space who's been grinding on this for many years?

1:15:01What does that OpenClaw moment mean?

1:15:03Well, it resonated a lot with what we were doing with our customers because pretty

1:15:08quickly, enterprises realized that if they wanted to make some gains with artificial intelligence, generative

1:15:13AI, they would need to automate full processes.

1:15:16And to automate the full process as an enterprise, well, you can use OpenClaw, but

1:15:20it's going to be, it's actually not really enough because you have data problems, you

1:15:24have governance problems, you can't observe the process that is running and you can't control

1:15:28it. In many cases, when you run a KYC process, if you're HSBC, for instance,

1:15:33one of our customers, you will want to have deterministic gates that are going to

1:15:38always do the same thing in a way that is observable and that you can

1:15:42guarantee the CAO that it's always going to go through these gates.

1:15:45And that's not something that OpenClaw is providing because it doesn't have the kind of

1:15:50primitives that you need to work on collective productivity, observable productivity, and to work on

1:15:55mission critical systems. On the other hand, the autonomy it gives and the autonomy it

1:16:00brings to people that are just individuals that are hacking together things is a way

1:16:05to also show to enterprises that if you set up the right control plane, if

1:16:08you set up the right sandboxes, if you connect to the right data sources, if

1:16:12you make sure that your access controls are well -respected, then you can actually unleash

1:16:16the power of agents doing things for your employees and that's going to work.

1:16:20Work on the platform because otherwise you will not be at ease when you're sleeping.

1:16:24It is definitely something you have to be thoughtful about.

1:16:27When I installed it, I gave it, just for my agent, root access to my

1:16:32Google Docs and my G Suite, my Notion, my Zoom, and my Notion, and G

1:16:40Calendar, everything. And then I realized, wow, I can, with my enterprise edition of Gmail,

1:16:46essentially, I can just summarize for my entire 21 -person investment company, every conversation going

1:16:53on in Gmail, and then correlate it with every conversation in Slack.

1:16:57And then I realized, oh my gosh, there's compensation discussions going on.

1:17:01There's a person on a PIP who we put on a performance improvement plan, perhaps,

1:17:06or something like that. Oh, I have to make sure nobody else can access this,

1:17:10because the power comes from giving it access to data, but with great power comes

1:17:17great responsibility. And I think people are learning that in real time.

1:17:20Yeah, it's a big problem, because the enterprise data is not a single thing that

1:17:24you want to put into a single system that is going to be accessible by

1:17:27everyone. And so you need to have this layer that actually understands what is in

1:17:32the data. You need to have a semantic of what can actually be proposed to

1:17:36HR, or what can be proposed to engineering.

1:17:41And typically, compensation is one of these things.

1:17:43You want to make sure that the compensation data does not flow back to all

1:17:47of the enterprise, because you're going to have a lot of problems, if that's the

1:17:50case. And so what you actually need, which is hard to do, is what we

1:17:53call a context engine. So a mapping of where the data sits, that comes with

1:17:58a certain number of metadata that is telling you that this data is actually not

1:18:02accessible to this part of the company.

1:18:03And if you actually have someone in engineering that is asking for something related to

1:18:08comp, the thing is actually going to tell you, look, you actually can't access that

1:18:12data. So that's hard. It's actually hard.

1:18:15You need to rethink entirely the way your IT systems are being connected.

1:18:19And at some point, you also need to think about your management, because your information

1:18:23flow is completely different today.

1:18:26If you're connecting agents together with your data sources than it used to be.

1:18:30And suddenly, maybe you don't need that manager whose only purpose was to take information

1:18:34from the bottom and put the information on the top, etc.

1:18:36So there's some IT problems to solve.

1:18:39And you need the right primitives, you need sandboxes, you need airbag, world -based access

1:18:44control, and these kind of things.

1:18:45And you have change to do.

1:18:47You need to rethink your entire customer service department, because suddenly, you actually don't need

1:18:52that much transfer of information operated by humans.

1:18:55All right. You have to go.

1:18:56You got a flight to catch.

1:18:57I have. It is so great to see you, Arthur.

1:18:59Continued success with Mistral. Thank you very much.

1:19:01Cheers. I'm really lucky to have Daniel Roberts here.

1:19:04He's the co -CEO and co -founder, along with his brother of Iren, and they

1:19:09are a publicly traded company.

1:19:10They started in BTC. Welcome to the All In Interview Program.

1:19:14Thanks, Jason. Pleasure to be here.

1:19:16Yeah. And so you started in Sydney.

1:19:18You and your brother was seven, eight years ago, and you got in early on

1:19:25Bitcoin. And all these Bitcoin miners wanted to have data centers, huh?

1:19:31Yeah, that's directionally right. So the thesis we saw was this explosion of the digital

1:19:35world, the growth in the online.

1:19:37And at some point, the real world was going to struggle.

1:19:40So we set about to build out large -scale data centers.

1:19:43Yes, the first use case was Bitcoin mining.

1:19:46But as we said to our seed investors, use that to bootstrap the platform, generate

1:19:50cash flow, layer in higher and better use cases over time as they emerge.

1:19:54Here we are today with AI.

1:19:55We are swapping out all the Bitcoin for AI chips.

1:19:58When did you first start seeing the demand in the company shift from, hey, Bitcoin

1:20:04miners, what we need? some H100s, whatever it is, to, hey, we're this non -profit

1:20:11open AI. Hey, we're this research lab.

1:20:13We need some AI compute.

1:20:16When did that start hitting?

1:20:17Look, we had a bit of a false dawn, I would say, back in 2020.

1:20:20We signed an MOU with Dell to start bringing out customers and compute.

1:20:24But in hindsight, it was too early.

1:20:25So we went back to Bitcoin, kept bootstrapping the platform.

1:20:28Look, I would say about two years ago and month by month, the demand just

1:20:33continues to escalate. And you were in so early that when you were looking at

1:20:37data center space in the United States, you were one of one looking at the

1:20:43space, one of two or three people looking at the space.

1:20:46They were trying to sell you on space, yeah?

1:20:49Yeah. So we actually developed the data centers ourselves.

1:20:51So we go and find the land, we go and get the permits, we go

1:20:55and apply for grid connections.

1:20:56And we were doing it at a scale that just amazed people at the time.

1:21:00Like 750 megawatts is our flagship Texas site four years ago was unheard of.

1:21:05In the middle of the desert, we're building these big data centers.

1:21:08The traditional data center industry going, what are you guys doing?

1:21:11We're saying, we believe in the future digitization, high performance computing, and obviously now today

1:21:16it's paying dividends. Yeah. I don't think anybody could have predicted when ChatGPT came out,

1:21:21OpenClaw recently as a turning point.

1:21:25And then Microsoft, Google, and everybody embracing this.

1:21:29And that's your big partner, Microsoft.

1:21:32Yes. Microsoft's one of our early partners.

1:21:35We signed a $9 .7 billion contract with them late last year.

1:21:38But as I was explaining to you before the show, that's 5 % of our

1:21:42capacity. So things are busy at the moment.

1:21:45And when you do these build outs, the big conversation today is no longer the

1:21:53number of GPUs putting in.

1:21:54It's just power. Power is the constraint today.

1:21:59Yeah. Look, for many of the industry, it is.

1:22:01But for us, because we started eight years ago, tying up all this land and

1:22:04power, it's not. So we've got four and a half gigawatts.

1:22:08For context, that's almost as much power annually as the Bay Area uses in its

1:22:13entirety. Wow. It's huge. So for us, the hurdle or the constraint is really time

1:22:19to compute. And that's emerging across the industry as well.

1:22:23And time to compute means tradespeople coming to West Texas, living in a trailer that

1:22:31you set up to then break ground on a data center, build foundations, build water

1:22:37cooling systems, like this is hard manual labor going on.

1:22:42Yeah, exactly. And this is the whole real world challenge to respond to these digital

1:22:46exponential demand curves that are unconstrained by the real world in terms of their appetite.

1:22:51And it just compounds. You need thousands of people out in these locations that haven't

1:22:56supported it. You put stress on supply chains.

1:22:58We're seeing what's happening with the memory, every aspect of it.

1:23:01So it's just permanent whack -a -mole, permanent solving fires to try and bring online

1:23:06this compute. And you get to spend time there.

1:23:10What's it like when you set up a town or you bring 1 ,000 people

1:23:15or 2 ,000 people to what's a pretty much remote small town?

1:23:19I'm assuming that when you bring 1 ,000, there might only be 500 living there

1:23:24right now. So what are those towns like?

1:23:27It sounds to me like something out of like the gold mining era when people

1:23:31first, you know, uh, went and, and were prospectors.

1:23:35Yeah. Prospecting town. Pretty, pretty much.

1:23:38I mean, the barbecue is great.

1:23:39That was the draw card.

1:23:40But apart from that, uh, look, we've always had a policy of hiring locals, supporting

1:23:44the local. community uh this year we're hitting a million dollars in community grants cumulatively

1:23:49yeah that's things like local playgrounds supporting the fire departments but we will hire locally

1:23:54once we can't find that trade locally we will expand the radius by 20 miles

1:23:58and hire out of that and so on and so on for us that's very

1:24:00thoughtful yeah and and these folks are coming say an electrician or a construction worker

1:24:06they're coming having built houses or you know uh maybe building um corporate offices and

1:24:16now they come for a tour of duty here and the salaries go up massively

1:24:20but they got to leave their family for a three -month tour or something yeah

1:24:25yes and no because typically where we locate is where there's heavy electrical infrastructure where

1:24:30there's heavy electrical infrastructure is typically where old manufacturing and industry has closed down so

1:24:37we go in leverage that sunk capex rehire retrain local workforces and bring a new

1:24:43industry to town in these data centers has that workforce now been completely depleted and

1:24:49we need to train another generation a younger generation to be generation two belt and

1:24:55really embrace the trades a hundred percent we're partnering with universities trade colleges absolutely and

1:25:01you go to a trade school you got you go to a college people are

1:25:06getting degrees in philosophy and english literature they're going 50k a year in debt 200k

1:25:12a year in debt what's the starting salary for a trades person working on a

1:25:16data center doing electrical or construction or hvac what's the ballpark range oh look i

1:25:23won't talk specifics but they they are going up the price is going up depends

1:25:27on the level but yes there is a rush for a good i'm hearing 150

1:25:31to like 300k am i in the ballpark the lower end directionally you're right yeah

1:25:35yeah i mean it's incredible when you think about it there's concern about hey ai

1:25:40shaking jobs and then on this other side of the ledger can't find enough talent

1:25:45to to to service it talk to me about energy sources and how you think

1:25:51about that uh president trump chris wright the administration that kind of started with hey

1:25:57clean beautiful coal year two they're like all sources matter nuclear obviously that gas is

1:26:04plentiful in that area we obviously got a lot of oil people don't know this

1:26:07about texas in the united states the number one uh source of solar installations yeah

1:26:13yes talk to us about energy so our philosophy has been sustainability from day one

1:26:18we have used 100 renewable energy since inception what 100 how is that possible we

1:26:25use hydro in british columbia we use wind and solar in west texas in west

1:26:29texas where we're located there's around 45 to 50 gigawatts of wind and solar yeah

1:26:34the transmission line to export that down to the load centers in dallas and houston

1:26:38is 12 gigawatts oh so you go and locate to the source of low cost

1:26:43excess renewable energy monetize it into this digital commodity exported at the speed of light

1:26:48as token great arbitrage and the wind is producing a lot but it's harder to

1:26:53get from those areas where people are willing to put up i mean people don't

1:26:57understand how big west texas is it is an incredible amount of land and you're

1:27:02coming from australia we're also on the west side people don't understand exactly how much

1:27:08just pure nature land there is yeah undeveloped so much land and the issue is

1:27:12distance you've got to spend billions of dollars on this transmission connection infrastructure to move

1:27:17that power to where people actually want it you can build wind farms you can

1:27:20build solar farms but if you build it in the desert and no one can

1:27:23use it, then what's the point?

1:27:25So the whole opportunity for our industry is to go to the source of that

1:27:28power and monetize it. So the data centers follow the wind turbines, the solar installations.

1:27:36How do you think about batteries and are you able to put those online?

1:27:39Because obviously you're going to have periods where, hey, it's not a windy day.

1:27:42In Texas, we have very few days when it's overcast.

1:27:45So that problem's pretty much solved, but you're going to have 50 days where the

1:27:49sun's not beating down. So how do you deal with the demand and softening that

1:27:54duck curve? We don't need to.

1:27:56The utility does that on our behalf.

1:27:58So this is why these grid connections are so scarce, so hard to get and

1:28:01so highly valued because once you get that grid connection, the utility underwrites all of

1:28:07that variability. They guarantee you 24 -7 reliable power.

1:28:11Got it. So on their side, they're figuring it out.

1:28:14Something goes down and they could fall back even though you're 100 % committed to

1:28:19renewables if they need it to fall back to gas or whatever.

1:28:22They have that ability out there.

1:28:23So you have that as a backup.

1:28:25A lot of talk about or a debate.

1:28:28Are we getting ahead of our skis?

1:28:31Are people slowing down? There was some talk about the OpenAI project, maybe downscaling a

1:28:36little bit. Is OpenAI a partner as well or?

1:28:39Can't comment. Can't comment. Okay.

1:28:41So we'll read into that whatever we want.

1:28:45But are there pockets where people are saying, hey, let's slow down or is it

1:28:49still gangbusters? It's right up the end of the spectrum.

1:28:53It's gangbusters. We cannot meet demand.

1:28:56That's why the whole industry now is around time to compute.

1:28:58There are no idle GPUs in the world sitting in a data center.

1:29:02And what's your take on when software makes, and this is a big discussion from

1:29:11Jensen himself during his two and a half hour keynote yesterday.

1:29:14We're sitting here Wednesday. I think he did his keynote on Tuesday.

1:29:16He was talking about, hey, software is going to make it 50 times more, lower

1:29:21the cost of tokens, 50x.

1:29:23And then you have transport also contributing to that.

1:29:27But when do you think the curve goes from parabolic to simply growing at a

1:29:33ridiculous level? Is there a slowdown coming or how are you planning for the future?

1:29:38Look, I think it's actually the opposite.

1:29:40I think it feeds on itself.

1:29:42So I'll give you one example.

1:29:43You go into ChatGPT today and you generate an image.

1:29:45You enter the prompt. It's like the dial -up internet days.

1:29:48It is. Right? It takes minutes.

1:29:50You're like, I better get this prompt right.

1:29:52Yeah. Finally, two minutes later, it comes.

1:29:54Now, I'll give you an example.

1:29:54If we 10x the amount of compute available, which is an enormous task from where

1:29:59we are today, and those images take 5 to 10 seconds, are we going to

1:30:03generate more or less images?

1:30:05Oh, many more. This is J.

1:30:07Evans' paradox. This is the theory of induced traffic.

1:30:10You build a couple more lanes, people start to think, well, maybe the distance from

1:30:15Bondi Beach to the central business district in Sydney terms would be an acceptable commute.

1:30:20Love the analogy. Yeah. So what do you think about, or what are you seeing?

1:30:26I mean, we're here at NVIDIA.

1:30:28Obviously, they make the leading -edge chips.

1:30:30They just bought Grok. So now you've got two of the leading -edge chips coming

1:30:35out of the same company.

1:30:37But custom silicon becoming a big discussion.

1:30:40Has that started to land in the data centers yet?

1:30:42Obviously, Google, don't know if they're a customer you can tell us, but they're making

1:30:46custom silicon. Amazon is making custom silicon.

1:30:50Meta is making custom silicon.

1:30:52Talk to me about that revolution, and is it actually making it to the data

1:30:55centers yet? Look, to various degrees, it is.

1:30:58They're promoting their products. They're trying to tie up data.

1:31:01They're trying to tie it to the data center center center center center center center

1:31:01center center center center center center center center center center center center center center center

1:31:01center center center center center center center center center center Yes.

1:31:01in the capacity. So yes, there's multiple silicon looking for homes.

1:31:05I think it's fair to say NVIDIA has a massive head start.

1:31:08The ecosystem they've incubated, the standards that they're setting.

1:31:11So I would say the safest pathway to build out at scale early is to

1:31:15follow the NVIDIA roadmaps. But absolutely over time, we are seeing these chips emerge.

1:31:21And in terms of desktop computing, I don't know if you saw the announcement that

1:31:27Dell and NVIDIA are making a really powerful desktop, 750 gigs of RAM, a lot

1:31:34of power. You're going to be able to run some local models, open source with

1:31:39OpenClaw and open source coming from Kimi and a bunch of the models out in

1:31:46China. Has the hacker group, which I think you started in like I did probably

1:31:51in similar time periods. People are starting to get really obsessed with having a $10

1:31:55,000 or $20 ,000 desktop setup and running this local.

1:31:59What do you think of that trend?

1:32:00I'm curious. Yeah, I mean, the breakthroughs we're seeing in software, the way it's distributing

1:32:03power to every man and woman in every house and their ability to code and

1:32:08use products like OpenClaw, the generation of demand and appetite for computers at a local

1:32:13level all the way through to these mega data centers.

1:32:16It's absolutely real. And as we see the emergence of agents using more and more,

1:32:20as we see autonomous vehicles and other automation, robotics, it's absolutely going to compound.

1:32:25And what about nuclear? The Trump administration really seemed to flip the switch on a

1:32:34growing belief that, hey, wait, nuclear is pretty great.

1:32:39It's clean. It's the original renewable in a way.

1:32:43And these new modular reactors have nothing to do with Chernobyl, Fukushima or Three Mile

1:32:49Island. They're much safer. They're a completely different architecture.

1:32:52Have those started to land yet?

1:32:54And are you, since you followed correctly in the great state of Texas where I'm

1:32:59from, you followed correctly that time, are you following nuclear?

1:33:03I think you have to.

1:33:04I think the reality is it's going to take a decade, a bit longer by

1:33:08the time big projects can come into commissioning.

1:33:10But now is the time to start that conversation, put in place policies, mobilize capital

1:33:15and start that ball rolling.

1:33:17Yeah. Do you have a data center going up near nuclear?

1:33:21No, not at the moment.

1:33:21Not at the moment, but you're actively tracking that activity?

1:33:26Yes. Yeah. This seems pretty inevitable.

1:33:30Yeah. Feels like it. And if that happens, what impact does it have on your

1:33:34industry? If you could, because obviously it's happening in China and people always put the

1:33:38Bitcoin miners, they were like the canary in the coal mine near the hydro dams

1:33:43and near the nuclear where there was excess capacity.

1:33:46What impact do you think this has if you could actually have small modular reactors

1:33:50next to data centers? Well, I think it just opens up the market and enhances

1:33:54the US's competitive advantage in this space.

1:33:57Like AI is inevitable. Robotics is inevitable.

1:34:00The reality is the correlation between human progress and energy consumption is really, really high

1:34:06over a very long time period.

1:34:08So if we can find a way to unlock new generation, clean generation as nuclear

1:34:11and locate that more at the source and enable more compute on a distributed basis,

1:34:16all those use cases we just discussed become easier, more fluid, faster.

1:34:20And then you get that positive flywheel around Jevon's paradox and demand.

1:34:25Talk to me about the architecture today of Ethernet and data moving between data centers

1:34:33within data centers. That backbone is going through a paradigm shift as well.

1:34:39Yeah. Yeah, it is. And Jensen coins the term, the data center is the new

1:34:43computer. Yeah. So you need to step back and you say, right, this big building

1:34:47is essentially the old desktop PC we had under our desk at home.

1:34:52You go, right, how does that work?

1:34:53So all the cabling, the latency, the number of hops between each GPU, how they

1:34:58talk to each other, the fabric around InfiniBand, Ethernet, it's absolutely critical because every millisecond

1:35:04matters in terms of performance of that cluster.

1:35:06Yeah. And where do you think or what do you think of Elon's vision?

1:35:15It's obviously a longer term vision of putting data centers in space and there's a

1:35:20couple other people working on it as well.

1:35:22Yeah, I mean, it's very hard to argue with Elon.

1:35:25He's been very right on a number of things for a very long time.

1:35:29I think sitting here today, it feels exceptionally difficult given the cost of moving things

1:35:33to space, the challenges around radiation.

1:35:36There's a huge amount of engineering challenges, but that's never scared Elon before.

1:35:40Yeah, he's uniquely qualified and he's inevitably right, but sometimes he's late.

1:35:47He might be late to the party.

1:35:48He might be late to the dinner party.

1:35:50He might show up at dessert, but generally he nails it.

1:35:54But how much of an issue is getting the data out of the data center

1:35:58to consumers today? Is that not something people are worried about when you're building something

1:36:03out in West Texas? All that data, fiber, all that's been taken care of, or

1:36:09does that become a gating issue at some point?

1:36:11So this was one of the big myths that we had to bust when we

1:36:14started this business, because everyone said data centers must be located close to population centers,

1:36:19metropolitan areas. Latency is really important.

1:36:21And we say, yeah, that's right.

1:36:22Latency is important. But the reality is in the US, Texas especially, there is fiber

1:36:28everywhere underneath the ground. Lots and lots and lots of it.

1:36:31And when you look at latency from our site in the middle of the desert

1:36:35in West Texas, down to Dallas, the big carrier hotel, six millisecond round trip latency.

1:36:41What's six milliseconds? There's a thousand milliseconds in a second.

1:36:46Yeah. We're talking six. It's adjacent.

1:36:49Yeah, it's not even, it's definitely not material.

1:36:53Listen, continued success and you're hiring a lot of people.

1:37:00Yeah. Yeah. I think we've got 129 job advertisements up at the moment.

1:37:04All right. So everybody go to the IRAN website and listen, company's doing fantastic.

1:37:10Thanks for spending some time with us here at All In at GTC.

1:37:15Thanks, Jason. Appreciate it. I'm doing all in.

Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN