WEBVTT - How the Hedge Fund Magnetar Is Financing the AI Boom 0:00:03.120 --> 0:00:17.400 Bloomberg Audio Studios, Podcasts, Radio News. 0:00:20.040 --> 0:00:23.280 Hello and welcome to another episode of the All Thoughts podcast. 0:00:23.480 --> 0:00:24.880 I'm Tracy Alloway. 0:00:24.560 --> 0:00:25.759 And I'm Joe Wisenthal. 0:00:26.600 --> 0:00:29.640 Joe, AI is so hot right now, in the immortal 0:00:29.720 --> 0:00:32.440 words of Mugatu, AI is so hot. 0:00:33.159 --> 0:00:33.479 It is. 0:00:33.800 --> 0:00:36.640 Yes, it is really hot. You know, you hear something. 0:00:36.960 --> 0:00:38.720 There's a little bit of slowing down in some of 0:00:38.800 --> 0:00:42.960 the progress on the models, but the recent in video 0:00:43.080 --> 0:00:46.880 results speak for themselves. There is nothing that I've seen 0:00:47.240 --> 0:00:50.680 yet that would suggest that this macro trend, at least 0:00:50.960 --> 0:00:53.120 as an investment trend, and I'm not talking about stocks 0:00:53.159 --> 0:00:56.800 per se, is anywhere close to quote slowing down. 0:00:57.160 --> 0:00:59.760 Yeah, And the interesting thing is we seem to be 0:01:00.320 --> 0:01:03.040 more and more players, some new types of players that 0:01:03.080 --> 0:01:05.679 are getting into the space. So, you know, we have 0:01:05.840 --> 0:01:09.319 AI funds kind of launching left and right. And one 0:01:09.319 --> 0:01:12.440 of the newest players is a hedge fund called Magnetar 0:01:12.800 --> 0:01:16.400 and I know them like primarily for credit stuff. I 0:01:16.440 --> 0:01:19.759 think they were big in redcap trades for a while. Yeah, 0:01:19.800 --> 0:01:22.680 and now they're launching an AI fund, a VC fund, 0:01:22.720 --> 0:01:24.839 which is kind of unusual for this type of hedge 0:01:24.840 --> 0:01:25.240 fund to. 0:01:25.160 --> 0:01:28.399 Do totally, I mean I've heard of magnetar for a 0:01:28.440 --> 0:01:32.720 long time, obviously, going back to the early twenty tens 0:01:33.040 --> 0:01:37.840 at least, And look, I'm not surprised that various investors 0:01:37.920 --> 0:01:41.720 are looking for what is their distinct way into this space? 0:01:41.800 --> 0:01:44.959 And of course, look, we've done interviews with vcs of 0:01:45.040 --> 0:01:48.280 various nature and positions in the past, and so I 0:01:48.320 --> 0:01:50.840 guess you know, there's sort of two questions to my 0:01:51.000 --> 0:01:54.840 mind anytime we're gonna be talking to someone investing in 0:01:54.920 --> 0:01:58.680 early stage or any stage of AI, which is obviously 0:01:59.160 --> 0:02:01.320 what is the thesis is what's going to win out 0:02:01.400 --> 0:02:04.880 where we'll value a crew. But then from an investor perspective, 0:02:05.800 --> 0:02:09.600 given so many entrants into this space, specifically whether on 0:02:09.639 --> 0:02:12.120 the public equity side, whether on the private side, whether 0:02:12.160 --> 0:02:14.880 on the VC side or early stage, late stage, what 0:02:15.040 --> 0:02:17.480 do they, as a fund or an investor bring to 0:02:17.520 --> 0:02:20.480 the table or will be able to see that the 0:02:20.639 --> 0:02:24.400 other billions of dollars competing for AI profits do not see. 0:02:24.680 --> 0:02:27.040 I have a slightly different question, which is for these 0:02:27.080 --> 0:02:30.520 types of investors, like how much is it about how 0:02:30.560 --> 0:02:34.200 good the technology is that they're investing in versus how 0:02:34.440 --> 0:02:36.960 much is it about getting in the right position in 0:02:37.000 --> 0:02:37.919 the capital stack. 0:02:38.040 --> 0:02:39.080 So that's a great question. 0:02:39.200 --> 0:02:41.440 I think it's going to be really interesting to talk 0:02:41.480 --> 0:02:45.720 to someone who's coming from this perspective. And without further ado, 0:02:46.080 --> 0:02:48.160 we have the perfect guest we're going to be speaking with, 0:02:48.280 --> 0:02:52.160 Jim Prosco. He is a partner and senior portfolio manager 0:02:52.520 --> 0:02:57.400 on Magnetar's Alternative credit and fixed income team. Jim, Welcome 0:02:57.440 --> 0:02:57.959 to the show. 0:02:58.400 --> 0:02:59.480 Thank you, great to be here. 0:02:59.800 --> 0:03:03.400 So how does someone on a hedge funds fixed income 0:03:03.440 --> 0:03:04.520 team get into AI. 0:03:05.800 --> 0:03:09.679 Well, we have a long history of investments in private companies, 0:03:10.080 --> 0:03:13.040 really dating back to an increased focus after the Financial 0:03:13.080 --> 0:03:16.400 Crisis when spreads and yields got tighter and the private 0:03:16.440 --> 0:03:21.360 markets seem more interesting. And we've often partnered with platforms 0:03:21.880 --> 0:03:24.760 where we thought we could grow the platform and generate 0:03:24.840 --> 0:03:28.040 an interesting asset, either a pool of cash flowing assets, 0:03:28.200 --> 0:03:31.320 or help grow the company and participate in that growth 0:03:32.040 --> 0:03:36.760 and support them through financing and other things like we 0:03:36.760 --> 0:03:39.760 can support them through helping them with hiring or accounting 0:03:39.960 --> 0:03:42.360 or other systems they need, and just to help them 0:03:42.400 --> 0:03:44.760 grow generally. And so you know, I've been doing that 0:03:44.880 --> 0:03:47.040 a long time and we've been a number of areas 0:03:47.320 --> 0:03:51.640 like auto lending in Ireland, and then we've moved into 0:03:51.920 --> 0:03:55.520 various fintech companies. We were one of the first institutional 0:03:55.520 --> 0:04:00.040 investors in open Door before they went public. We're supporting 0:04:00.080 --> 0:04:04.040 and investing in a very interesting fintech that is financing 0:04:04.080 --> 0:04:07.760 restaurants right now, and so we felt we had experience 0:04:07.840 --> 0:04:11.880 in that space, and then that sort of overlapped with 0:04:12.680 --> 0:04:15.920 our relationship and our investment in core Weave, where we 0:04:15.920 --> 0:04:19.080 were the first institutional investor in core Weave in twenty 0:04:19.279 --> 0:04:23.640 twenty one. So we're very early in the trend of 0:04:23.680 --> 0:04:27.680 putting capital into the AI infrastructure space and that's just 0:04:27.760 --> 0:04:30.039 sort of grown as this whole market has grown to 0:04:30.160 --> 0:04:34.279 encompass literally everything. Now, you know, we continue to look 0:04:34.320 --> 0:04:37.120 for smart ways to invest, and you know, one of 0:04:37.120 --> 0:04:40.920 those ways we felt was what can we provide that's 0:04:40.920 --> 0:04:43.320 a value And one of the things we can provide 0:04:44.000 --> 0:04:47.080 besides the general help we can give a growth stage 0:04:47.080 --> 0:04:52.280 company is compute because that is the scarce resource right now, 0:04:52.320 --> 0:04:54.599 and that's where all the capital is going to the 0:04:54.680 --> 0:04:58.480 various parts of the value chain to deliver compute, and 0:04:58.560 --> 0:05:01.440 so there's a competition to get compute, and if you're 0:05:01.440 --> 0:05:06.120 a smaller company with limited capital or limited access to capital, 0:05:06.600 --> 0:05:08.839 it can be difficult to get that, and so that 0:05:09.000 --> 0:05:11.039 was sort of the value proposition we thought we could 0:05:11.040 --> 0:05:11.560 bring to bear. 0:05:12.040 --> 0:05:14.680 Joe, I have this vision in my head of vcs, 0:05:14.760 --> 0:05:18.120 like going into startups bearing baskets full of chips. 0:05:18.279 --> 0:05:19.960 Ah yeah, instead of just saying. 0:05:19.800 --> 0:05:22.680 That, like our pitch is the relationship and the coaching as. 0:05:22.839 --> 0:05:25.720 We have access to the chips or the energy plus chips. 0:05:26.160 --> 0:05:29.839 Just for point of clarification, listeners should know we've talked 0:05:29.839 --> 0:05:33.280 to Core. We've at least twice on the show, and 0:05:33.360 --> 0:05:38.240 it feels like in the AI space specifically, this is 0:05:38.279 --> 0:05:40.720 one of those names that's a very big deal, but 0:05:40.880 --> 0:05:43.479 not many people don't know it the way they know 0:05:43.640 --> 0:05:47.919 sayan in Nvidia at the very back end or Chatgypt 0:05:48.160 --> 0:05:50.920 at the very front end, but they build a lot 0:05:50.960 --> 0:05:54.480 of the data centers that are filled with Nvidia chips. 0:05:54.680 --> 0:05:56.920 I want to get more into the business model there 0:05:56.960 --> 0:05:59.000 because I have a lot of questions in the business 0:05:59.000 --> 0:06:02.240 of selling compute, etc. But talk a little bit more 0:06:02.520 --> 0:06:06.120 about you said your experience in the private side is 0:06:06.160 --> 0:06:10.480 like this expertise with platforms per se. And when I 0:06:10.520 --> 0:06:12.880 think of platforms, I think of companies that can acquire 0:06:12.920 --> 0:06:15.400 lots of other companies or a lot can be built 0:06:15.480 --> 0:06:18.760 onto them. Talk to us about how the platform specific 0:06:18.920 --> 0:06:23.760 expertise informs you're thinking with a core weave or any 0:06:23.800 --> 0:06:26.320 other AI investment that you're making now. 0:06:27.480 --> 0:06:33.640 So we've tried to put capital into companies that are 0:06:34.200 --> 0:06:37.240 trying to build their business in a particular space, and 0:06:37.880 --> 0:06:40.760 oftentimes that could be a space where they generate a 0:06:40.800 --> 0:06:45.160 cash flowing asset, like in the auto loan example, in 0:06:45.320 --> 0:06:48.880 the open door example, they were acquiring real estate, which 0:06:48.920 --> 0:06:53.240 was a hard asset. In that restaurant fintech example, they're 0:06:53.279 --> 0:06:57.560 acquiring restaurant credit. And so we've tried to support businesses 0:06:57.680 --> 0:07:01.560 that had some kind of asset or flow and work 0:07:01.600 --> 0:07:04.039 with them on a number of ways that we can 0:07:04.120 --> 0:07:06.400 add value. I think first and foremost is all these 0:07:06.440 --> 0:07:09.359 growth stage companies need financing, and I think we have 0:07:09.440 --> 0:07:13.440 great expertise from debt to equity, private to public, and 0:07:13.520 --> 0:07:16.560 we can be innovative in trying to bring you the best, 0:07:17.440 --> 0:07:21.280 most appropriate, lowest cost capital to these growth stage companies. 0:07:21.320 --> 0:07:22.800 And like I said, as well as. 0:07:22.920 --> 0:07:25.880 So just to be clear, just to understand in this context, 0:07:26.280 --> 0:07:30.239 what makes AI distinct, say from other waves of tech 0:07:30.760 --> 0:07:34.040 or what makes it distinct for say a magnetar is 0:07:34.080 --> 0:07:38.560 in part this distinct capital demand that was not perhaps 0:07:39.040 --> 0:07:41.000 as big of a deal during the SaaS wave of 0:07:41.040 --> 0:07:41.840 the twenty tens. 0:07:43.040 --> 0:07:46.080 Yes, so not not only a general capital demand, but 0:07:46.280 --> 0:07:49.440 in many cases, for many of these companies, a very 0:07:49.480 --> 0:07:54.040 specific demand to have capital to deploy with compute, and 0:07:54.120 --> 0:07:58.800 because they need this very specific scarce resource, helping to 0:07:58.840 --> 0:08:03.240 deliver that resource, and in particular helping to deliver that 0:08:03.320 --> 0:08:05.840 resource in a high quality way. Where you have a 0:08:05.880 --> 0:08:09.080 partner like core Weave that has I think there's a 0:08:09.080 --> 0:08:11.840 lot of evidence that they have the highest performing AI 0:08:11.960 --> 0:08:16.360 training cluster, and so that is really valuable to these 0:08:16.400 --> 0:08:20.040 companies that might otherwise struggle to get enough compute to 0:08:20.200 --> 0:08:21.360 further their business model. 0:08:21.720 --> 0:08:24.920 Speaking of Core We've I'm really curious how that conversation 0:08:25.160 --> 0:08:28.640 actually started because this was a new and novel thing. 0:08:28.680 --> 0:08:31.960 I don't think we had chip based loans before to 0:08:32.040 --> 0:08:35.319 my knowledge, and I keep hearing that asset based financing 0:08:35.400 --> 0:08:37.680 is going to be like this next big thing in 0:08:37.800 --> 0:08:42.560 private credit or it's the last real frontier in private credit. 0:08:42.800 --> 0:08:45.200 How did you come up with this idea this deal? 0:08:46.840 --> 0:08:50.680 Well, acid based financing is really a classic private credit 0:08:50.760 --> 0:08:54.480 tool and there's a number of examples. Just if you 0:08:54.559 --> 0:08:59.160 think about my example with the Irish auto lender. If 0:08:59.200 --> 0:09:02.480 you buy a loan for a car, so the Irish 0:09:02.520 --> 0:09:06.080 auto lenders generating car loans and those go and you 0:09:06.160 --> 0:09:10.119 buy them in a vehicle, you have primarily the security 0:09:10.120 --> 0:09:12.960 of the people paying on those loans, and so you 0:09:13.000 --> 0:09:16.800 get paid back by the cash flow of the borrowers 0:09:17.040 --> 0:09:19.760 paying their car loans back, but there's credit risk to 0:09:19.800 --> 0:09:22.720 that they could potentially stop paying, and in the case 0:09:22.720 --> 0:09:26.239 where they stop paying, then you have the cars collateral. 0:09:26.840 --> 0:09:32.880 And really that metaphor applies almost directly to GPUs, where 0:09:33.400 --> 0:09:37.360 if you're a company delivering high performance compute like Core 0:09:37.440 --> 0:09:42.600 we've has, you're contractually selling that compute to some counterparty 0:09:42.600 --> 0:09:45.040 that's going to use it in their case. You know, 0:09:45.120 --> 0:09:48.800 that's often a very large, very credit worthy hyperscaler, but 0:09:48.920 --> 0:09:52.480 not always. There could be smaller startups that have riskier 0:09:52.520 --> 0:09:58.079 business models, and in that case, primarily by funding the GPU, 0:09:58.160 --> 0:10:01.280 you're getting paid back with those controls actual cash flows 0:10:01.360 --> 0:10:04.439 on the use of the GPU. But in the case 0:10:04.480 --> 0:10:09.480 that company fails, then as backup you have the GPU itself. Now, 0:10:10.240 --> 0:10:12.560 the GPU isn't really like the car where you'll probably 0:10:12.559 --> 0:10:14.920 go out and sell it, but you get the time 0:10:15.080 --> 0:10:17.920 back on the GPU, which you can then resell to 0:10:18.000 --> 0:10:21.120 somebody else, and being a scarce asset, you can think 0:10:21.120 --> 0:10:23.920 about what value that would have in a future time. 0:10:24.840 --> 0:10:29.160 One difference that I could imagine with the GPU versus 0:10:29.240 --> 0:10:33.120 other forms of assets, say whether it's a car or 0:10:33.160 --> 0:10:35.680 say whether it's a house, is a certain here in 0:10:35.760 --> 0:10:40.319 twenty twenty four, still unpredictability about many things in the future. 0:10:40.720 --> 0:10:45.360 Will in video always be the gold standard so to speak? 0:10:45.520 --> 0:10:47.880 In AI chips maybe it looks like it, yes, but 0:10:47.960 --> 0:10:52.120 it doesn't seem guaranteed. How fast will the current generation 0:10:52.240 --> 0:10:56.120 of chips that are deployed degrade in value? I imagine 0:10:56.120 --> 0:10:59.960 there are fairly predictable sort of depreciation curves for cars 0:11:00.280 --> 0:11:03.920 that perhaps are more uncertain for chips. And then also 0:11:04.200 --> 0:11:09.840 the uncertainty of actual deployment given permitting and challenges with 0:11:10.040 --> 0:11:12.560 energy and the other operational things that have to do 0:11:12.600 --> 0:11:15.600 with a new company building a data center. Talk to 0:11:15.679 --> 0:11:19.200 us about modeling or at least thinking through some of 0:11:19.240 --> 0:11:23.079 the uncertainties with chips specifically, Well, depending. 0:11:22.679 --> 0:11:26.240 What stage you get involved, you have the breadth of 0:11:26.320 --> 0:11:30.040 all those different risks potentially. So if you're investing in 0:11:30.160 --> 0:11:34.600 high performance compute but it's a greenfield data center, then 0:11:34.640 --> 0:11:36.360 you have to think about all those things. You have 0:11:36.440 --> 0:11:39.560 to think about the delivering of the power. You have 0:11:39.600 --> 0:11:42.319 to think about the timing on all the components to 0:11:42.360 --> 0:11:45.680 get to the data center. If you're making what we've 0:11:45.679 --> 0:11:48.520 been talking about, which is sort of a GPU based loan, 0:11:48.920 --> 0:11:53.760 then usually that loan is based upon a running GPU 0:11:53.960 --> 0:11:57.320 and an existing high performance compute data center, so you 0:11:57.360 --> 0:12:00.559 don't really have to think about some of the earlier 0:12:00.600 --> 0:12:03.520 stage issues. You more have to think about how long 0:12:03.640 --> 0:12:07.120 is my contract, how good is my contract, What do 0:12:07.240 --> 0:12:10.719 I think the value of renting that chip out will 0:12:10.760 --> 0:12:14.560 be at the end of that contract. How much rent 0:12:14.600 --> 0:12:16.520 on that chip could I get if I had to 0:12:16.559 --> 0:12:19.040 re rent that in the middle of the contract. So 0:12:19.080 --> 0:12:22.400 it's more near term things on actually having a functioning 0:12:22.480 --> 0:12:25.560 GPU in the data center, But all those other things 0:12:25.559 --> 0:12:27.679 have to be financed too, and there's going to be 0:12:27.880 --> 0:12:32.200 innovative and large amounts of capital dedicated to financing those things. 0:12:32.559 --> 0:12:35.640 Setting aside the financing for a second, how hard has 0:12:35.679 --> 0:12:41.200 it been just to find physical space in data centers, well, it's. 0:12:41.040 --> 0:12:44.559 Been extremely scarce, and a lot of that is driven 0:12:44.720 --> 0:12:50.160 by the search for power. The data centers required for 0:12:50.280 --> 0:12:53.760 the new AI chips are much different than the old 0:12:53.840 --> 0:12:57.880 data center. So it isn't really cost efficient in most 0:12:57.920 --> 0:13:00.120 cases to go and take an old data center and 0:13:00.160 --> 0:13:03.840 try to retrofit it because the amount of power just 0:13:03.920 --> 0:13:07.000 a loan that has to go there is you know, 0:13:07.360 --> 0:13:11.120 transcending an order of magnitude more per rack of GPUs now, 0:13:11.160 --> 0:13:15.040 and so that's just you just can't really retrofit that efficiently. 0:13:15.080 --> 0:13:17.520 It's better to build your own building. And so it's 0:13:17.600 --> 0:13:22.760 really come down to things like permitting availability of power 0:13:23.559 --> 0:13:26.200 and time to get all your components, and you know, 0:13:26.240 --> 0:13:28.960 all these things have their own lead time. So it 0:13:29.000 --> 0:13:32.479 had an interesting back and forth to Brian on curing transformers. 0:13:32.520 --> 0:13:35.839 You know, all these little you know, nuances come into 0:13:35.920 --> 0:13:37.920 play when you have to build a data center. And 0:13:38.000 --> 0:13:43.280 so because power is really the limiting factor most of all, 0:13:43.679 --> 0:13:47.200 you're seeing a lot of moves towards where the power is. 0:13:47.760 --> 0:13:50.600 And it was recently an article on Bloomberg. I think 0:13:50.640 --> 0:13:53.000 about a company in Texas that owns a bunch of 0:13:53.080 --> 0:13:55.560 land that's now worth forty billion dollars, right, And that's 0:13:55.600 --> 0:13:59.960 because they're near all this renewable power. But that isn't 0:14:00.000 --> 0:14:03.520 the only thing. It's incredibly complex to operate this high 0:14:03.559 --> 0:14:06.679 performance compute. So then you have to think about if 0:14:06.720 --> 0:14:08.760 I try to build my data center out there where 0:14:08.800 --> 0:14:12.559 the power is. Can I get everything out there, including 0:14:13.600 --> 0:14:15.000 operational expertise? 0:14:15.080 --> 0:14:15.240 Right? 0:14:15.280 --> 0:14:17.360 Can I staff my data center with the kind of 0:14:17.360 --> 0:14:20.680 experts I need to run this kind of highly technical, 0:14:20.760 --> 0:14:24.880 high performance compute. And each generation is just getting more complicated. 0:14:25.160 --> 0:14:28.160 We're going to have liquid cooling on the next generation 0:14:28.240 --> 0:14:31.840 of Nvidia chips, probably immersion cooling right after that. It's 0:14:32.000 --> 0:14:36.120 very complicated, very expensive, and very difficult to scale. Much 0:14:36.160 --> 0:14:39.280 harder to do in a large size than it is 0:14:39.320 --> 0:14:39.640 to do in. 0:14:39.640 --> 0:14:40.560 A small size. 0:14:40.800 --> 0:14:45.800 Maybe Magnetar can finance a small modular nuclear reactor. No, seriously, 0:14:45.840 --> 0:14:49.280 because if you're financing the compute and securing that on 0:14:49.400 --> 0:14:51.920 behalf of companies that you want to invest in, you 0:14:51.960 --> 0:14:54.520 could go one layer down finance the energy. 0:14:55.480 --> 0:14:57.520 And we're certainly interested in that, and we have a 0:14:57.640 --> 0:15:00.760 history in investing in energy. We have investment right now 0:15:01.080 --> 0:15:04.040 and a developer of utility scale solar power in the 0:15:04.120 --> 0:15:07.720 US who has least some of that solar power to 0:15:08.080 --> 0:15:12.160 various hyperscalers. So that is certainly a space we're interested in. 0:15:12.640 --> 0:15:16.120 I was just in Miami meeting with a company that 0:15:16.240 --> 0:15:19.880 has a novel heat sink battery technology that they want 0:15:19.880 --> 0:15:22.000 to deploy to data centers that they're talking to a 0:15:22.040 --> 0:15:26.720 bunch of data center type companies about launching that product there. 0:15:26.800 --> 0:15:29.160 So there's a ton of interesting things, and just like 0:15:29.240 --> 0:15:32.160 every other part of this ecosystem, it's going to require 0:15:32.160 --> 0:15:33.400 an immense amount of capital. 0:15:33.760 --> 0:15:37.320 I guess, just since we're sidetracked on the energy component 0:15:37.440 --> 0:15:42.680 for now while we're here novel battery technologies, there's a 0:15:42.720 --> 0:15:44.880 lot of them out there. There's a lot of startups 0:15:44.880 --> 0:15:48.600 that have something novel and energy, and often one of 0:15:48.640 --> 0:15:51.320 the things that they talk about is this chicken and 0:15:51.360 --> 0:15:55.280 egg problem where they need capital, They need sort of 0:15:55.360 --> 0:15:58.600 financing of some sort or another to build this stuff, 0:15:58.840 --> 0:16:01.120 but the lenders don't really want to give it until 0:16:01.120 --> 0:16:03.280 there's demand, and no one's just going to promise to 0:16:03.280 --> 0:16:06.720 buy it until it's shown that it can work. Can 0:16:06.760 --> 0:16:08.240 you talk a little bit, I mean again, I know 0:16:08.280 --> 0:16:10.720 there's a little bit off track from GPUs themselves. But 0:16:10.840 --> 0:16:14.840 since you were talking about similar yeah, talk about the batteries. 0:16:15.000 --> 0:16:17.160 Can you talk a little bit about that dynamic as 0:16:17.200 --> 0:16:19.120 it affects solving the energy side of the equation? 0:16:19.800 --> 0:16:20.440 Yeah, for sure. 0:16:20.480 --> 0:16:22.760 And it has some overlap with the way you look 0:16:22.840 --> 0:16:25.440 at an AI company too. You know, if you think 0:16:25.480 --> 0:16:29.200 about the core things that we really want to look at, 0:16:29.760 --> 0:16:36.000 it's technology team and traction. So does their technology really work? 0:16:36.080 --> 0:16:38.440 That's first and foremost. You know, what is this product? 0:16:38.480 --> 0:16:42.320 Does it have some kind of advantage? And then traction 0:16:42.800 --> 0:16:44.400 like time to market. 0:16:44.200 --> 0:16:45.560 That's super important. 0:16:45.840 --> 0:16:49.200 I was just talking to isokon pool side and like 0:16:49.520 --> 0:16:52.000 to him, like those are the two most important things. 0:16:52.200 --> 0:16:55.440 Speed to product, speed to market, because it's a race, 0:16:55.920 --> 0:16:58.960 and even if you have the greatest technology, if you 0:16:59.040 --> 0:17:01.560 take too long, someone's going to be using something else. 0:17:01.600 --> 0:17:05.240 And that's certainly true in the energy space where energy 0:17:05.320 --> 0:17:09.200 is of critical importance. So I think that for these 0:17:09.240 --> 0:17:13.160 startups on the traction side, they really need some strategic 0:17:13.280 --> 0:17:17.520 partnerships because their cost of capital is very high. 0:17:17.560 --> 0:17:20.919 Strategic partnership is kind of like an existing company that 0:17:21.040 --> 0:17:23.760 has a demand. It also has a lot of cash 0:17:23.920 --> 0:17:26.080 and could theoretically be a buyer of their. 0:17:25.880 --> 0:17:29.679 Solution, yes, and really on the other side too, So 0:17:29.840 --> 0:17:34.800 for example, because their cost of capital is so high, 0:17:34.880 --> 0:17:36.679 there's certain things that it's hard for him to do. 0:17:36.880 --> 0:17:39.000 And one of the things that it's really hard for 0:17:39.080 --> 0:17:41.199 all these startups to do, and this was true and 0:17:41.320 --> 0:17:44.680 the recycling industry and other industries, is build a plant. 0:17:45.440 --> 0:17:48.760 Like very expensive, time consuming to build a plant. You 0:17:48.800 --> 0:17:52.160 don't really want to raise bc capital to build a plant, 0:17:52.440 --> 0:17:54.840 and so it's important to have a partnership on the 0:17:54.880 --> 0:17:57.920 manufacturing side too. And that was really like the first 0:17:57.920 --> 0:18:01.119 thing this battery startup that I just visited talked about 0:18:01.520 --> 0:18:04.000 is like getting that because you've got to be able 0:18:04.000 --> 0:18:05.960 to deliver your product and you have to deliver it 0:18:06.000 --> 0:18:09.360 on scale, and ideally you don't want to be wasting 0:18:09.400 --> 0:18:12.080 time building your own plant on that and then like 0:18:12.119 --> 0:18:14.440 you said, on the other end, you want to have 0:18:14.480 --> 0:18:18.080 a partnership with the users of the energy, which is 0:18:18.400 --> 0:18:20.920 all the people that either have data centers or use 0:18:21.040 --> 0:18:24.960 data centers or customers of data centers, and you want 0:18:25.000 --> 0:18:29.560 them to ideally put together an attract a financing relationship 0:18:29.600 --> 0:18:33.600 where you know, in some form or fashion they're front 0:18:33.640 --> 0:18:37.280 loading their payments to use so that you can use 0:18:37.320 --> 0:18:40.120 that capital to actually build a product that they meet. 0:18:56.480 --> 0:18:58.879 So Joe and I went to San Francisco a little 0:18:58.880 --> 0:19:01.680 while ago and we saw some cool things. I had 0:19:01.680 --> 0:19:03.960 my first ride in a way Moo, and we saw 0:19:04.000 --> 0:19:07.480 some cool battery related technology. We also saw a lot 0:19:07.480 --> 0:19:11.560 of vcs. Everyone very excited about AI. Obviously, they were 0:19:11.560 --> 0:19:15.160 also talking about the difficulty of chasing deals right now, 0:19:15.560 --> 0:19:20.200 how do you compete with those traditional vcs or are 0:19:20.240 --> 0:19:23.160 you just not competing with them directly because you're taking 0:19:23.200 --> 0:19:25.760 the slightly different GPU backed approach. 0:19:26.680 --> 0:19:27.760 You know, I think it's both. 0:19:27.840 --> 0:19:30.560 I think you're competing with them and to an extent, 0:19:30.760 --> 0:19:33.879 partnering with them. And that's the thing we had to 0:19:33.920 --> 0:19:37.000 ask ourselves before launching the fund, is what are we 0:19:37.080 --> 0:19:40.399 bringing to the bear that's value added? And in this case, 0:19:40.440 --> 0:19:44.680 we're bringing to bear the compute. And so often these startups, 0:19:44.800 --> 0:19:47.720 even if they're backed by a strong VC, can have 0:19:47.800 --> 0:19:49.840 a bit of a chicken and egg problem, which is 0:19:50.720 --> 0:19:53.600 they need compute to develop their product, and they need 0:19:53.640 --> 0:19:56.600 capital to buy that compute. But if they don't have 0:19:56.640 --> 0:20:00.240 the compute lined up and the price locked in, then 0:20:00.280 --> 0:20:02.600 the capital might be hesitant to go in because they'd 0:20:02.600 --> 0:20:05.159 be like, we could put our capital into you, and 0:20:05.200 --> 0:20:07.159 then it could take you an extra six months to 0:20:07.200 --> 0:20:10.440 get your compute, and by that time some competitors passed 0:20:10.480 --> 0:20:14.240 you by or the technology has changed. And on the 0:20:14.320 --> 0:20:16.800 other hand, because they're a startup, they don't really have 0:20:16.840 --> 0:20:20.240 the credit worthiness to just contract the compute. They most 0:20:20.359 --> 0:20:23.160 likely have to pay up front, and so we bridge 0:20:23.200 --> 0:20:26.720 that gap. And so if we go into a fundraising 0:20:26.840 --> 0:20:29.280 round where there's a bunch of vcs putting cash in, 0:20:29.800 --> 0:20:34.240 if they know that we're putting compute in alongside them 0:20:34.600 --> 0:20:37.119 and that the second the round closes that compute will 0:20:37.160 --> 0:20:39.760 be available to the company, that makes it easier to 0:20:39.880 --> 0:20:42.880 raise the cash part of it. So we are competing 0:20:42.920 --> 0:20:45.480 and we need that value added to be part of 0:20:45.520 --> 0:20:48.359 the equation. But also I think it helps them to 0:20:48.480 --> 0:20:51.840 raise from traditional vcs because we take that one risk 0:20:51.880 --> 0:20:52.520 off the table. 0:20:52.840 --> 0:20:56.480 How big is the market of companies that need compute, 0:20:56.520 --> 0:20:59.760 because there are plenty of AI companies that just build 0:20:59.800 --> 0:21:05.720 on top of an existing model like GPT or anthropics model, 0:21:05.720 --> 0:21:10.439 et cetera. How many companies are actually out there and 0:21:10.520 --> 0:21:12.919 who like not who are they specifically, but what are 0:21:12.920 --> 0:21:16.639 the types of companies for whom actual access to compute 0:21:17.080 --> 0:21:19.240 is an important part of their business. 0:21:20.320 --> 0:21:22.520 Yes, well, you know it starts, of course with the 0:21:23.400 --> 0:21:28.439 LM companies. You're using massive, huge, huge amounts of compute. 0:21:28.760 --> 0:21:30.200 But then if you look. 0:21:30.080 --> 0:21:34.840 At the rest of sort of the AI stack, there's 0:21:34.880 --> 0:21:37.040 a couple areas where you're going to need compute, and 0:21:37.119 --> 0:21:43.240 one is all the small model custom model companies, and 0:21:43.560 --> 0:21:45.479 small commute a lot of different things. So you can 0:21:45.840 --> 0:21:49.000 have some very small companies that are using a very 0:21:49.040 --> 0:21:52.000 targeted model, like say in a vertical stack, you might 0:21:52.040 --> 0:21:56.560 have a robotics company that is specifically training a model 0:21:56.760 --> 0:22:00.520 to run a robot in a particular situation, and that 0:22:00.520 --> 0:22:03.520 could be anything from a warehouse to doing surgery, right, 0:22:04.080 --> 0:22:09.679 and they need compute to train that model or another 0:22:09.720 --> 0:22:12.360 one which is huge and dominated by an existing big 0:22:12.359 --> 0:22:17.280 players autonomous driving, but there are other autonomous driving companies 0:22:17.320 --> 0:22:21.480 that are trying to be deployed at other automakers that 0:22:21.680 --> 0:22:23.320 need compute to train those models. 0:22:23.520 --> 0:22:24.600 Or weather models. 0:22:25.280 --> 0:22:28.200 There's some really good companies that we've talked to doing 0:22:28.280 --> 0:22:31.879 weather models. They need compute to train their model, and 0:22:32.000 --> 0:22:35.560 so that whole model layer, and then even on the 0:22:35.640 --> 0:22:40.720 app layer, they might be custom elements of small models 0:22:40.720 --> 0:22:42.480 that they have that sit on top of the big 0:22:42.720 --> 0:22:44.760 lms that they need some amount of compute for. 0:22:45.560 --> 0:22:46.639 So there's quite a range. 0:22:46.680 --> 0:22:49.680