The DeepSeek Fire Drill

[00:00:00] Hi, I'm Eric Newcomer. And I am Madeline Rynbarger. And this is the Newcomer Podcast. Each week, Eric and I discuss the VC deals and the drama that went down. Let's do it. Here we go. There are weeks where it feels like the world has changed. And sometimes, you know, world history does transpire in a week. And other times, it feels like we can get very excited about something.

[00:00:28] And then it's less exciting than it seems. And I think that is the story of DeepSeek, the Chinese AI company that many of us hadn't heard of. You know, I'd probably heard it on tech Twitter for a little bit, but certainly wasn't a front of mind story until the Chinese tech company devastated the NASDAQ and devastated NVIDIA's stock price in particular. NVIDIA shares plunged 17% on Monday.

[00:00:51] It was the largest market cap loss for a U.S. company in one day, close to $600 billion in value. Of course, that's back up now. The NASDAQ fell 3.1% or 612 points, but now it's only down 1.8%. So it seems like the big freak out of everyone discovering this over the weekend is now people know what it is. It's out there. It's a little bit. It's calmed. It's cold. But yeah, it feels like one of those weeks where all of a sudden the game has changed.

[00:01:21] Yeah. I mean, I think, you know, some things have changed, right? I mean, I think one, and this I think has been talked about less. It's like, you know, it's Cerebral Valley in November. It's like, is there a wall? Is there a wall? You know, is AI going to improve? Well, you know, Deep Seek isn't the only one changing the game. Like, yes, if you can be smarter in a more cost-effective way, then yes, the game is definitely changing.

[00:01:45] It feels like certain that we're going to see serious AI development this year. So I think that's clear. It underlines, you know, how important open source is. I think people pay attention. You know, Lama and others were already sort of doing incredible things. And then third on the technology side, obviously, it just reminds us that China is serious. It's a major global power.

[00:02:10] You know, it could be during the Trump administration that we sort of wake up and say, oh, man, China is surpassing us on some things. And we can't just sort of pretend like American tech giants are the only source of activity in the world. I mean, I think, honestly, in self-driving cars, you're seeing a lot of Chinese cities deploy self-driving cars, even though in the U.S. we mostly talk about Waymo.

[00:02:34] So, yeah, there's definitely a lot of areas in AI where China is doing stuff that I guess our markets probably don't quite appreciate yet. Totally. I think one part of this, too, is that in the model race, leaders have been saying for months now that the next breakthrough is always, you know, just the next model release away. And competition can be really fierce, right? Like OpenAI has had the industry lead for a while.

[00:02:57] But as new developments come up, I remember, you know, when Anthropic dropped computer use, that was the agent, that product that just, you know, took everyone by storm at that point. So it's not really a given that it's just going to be an American tech giant that's going to take the whole industry. Yeah, I don't know. Coming out of this whole experience, to me, my initial reaction was vindication for the application thesis. We've been platforming people with the application thesis since Cerebral Valley, New York last June. Right.

[00:03:28] I mean, this idea that a lot of the value is going to accrue to people building use cases, whether for consumers or enterprises, it's the relationship with the customer, understanding how to sell it to people, and sort of the more traditional things around software businesses that would, again, differentiate AI companies. And that foundation models were potentially spending a lot of money to build a commoditized technology. Certainly, DeepSeek made that seem even more likely.

[00:03:56] I will say that, you know, as we're recording this podcast, now the Wall Street Journal is reporting that OpenAI is in talks to raise $40 billion at a $340 billion valuation from SoftBank. SoftBank, you know, this – So, you know, the market – Investor appetite has not slowed, sounds like. Right, right. They're going to do okay right now. Right, right. It's not as damning. You know, SoftBank clearly is still a buyer for OpenAI.

[00:04:23] I mean, now, you know, obviously OpenAI has a direct consumer product and there's part of the value proposition for them is that even if their technology gets commoditized, they are the Google, the iconic brand of the category. And so consumers will always flock to them. Right. Like you could say that they really are the first to build a killer app so far.

[00:04:45] I mean, they have the – we talk a lot about where the companies are going to come at the application layer, but they have an application that million – like 300 million weekly active users as of last year, you know, like they have it. So it doesn't – the real value coming from applications also doesn't necessarily count out OpenAI.

[00:05:02] I think the thing that is muddled a lot of takes and honestly makes it even hard for us to opine on some of it is this debate about how much DeepSeek is sort of a psyop or like how honest are they about their access to NVIDIA chips and how much money is being spent. This is really where journalism fails because you're seeing a lot of narratives have to get written where it's like classic journalism, you pick one side or the other of that.

[00:05:30] Like either they had a lot of money or they didn't, but then you know you don't know for sure and so you have to sort of thread to competing narratives. So I do think one thing – Yeah, the waffling is pretty apparent everywhere. I mean, just because we need more information. We will not participate in that as talking heads. I think it's an open question. I think what's clear – you know, it's clear that they have NVIDIA chips, but there are, you know, things beyond H100s and it's not clear exactly the mix and how high power their access is

[00:05:58] and whether it represents this sort of random company and a hedge fund or if this is part of the Chinese government's real sort of bet here and their sort of backing that we don't know about. And then there is, I think, more certainty that what DeepSeek has released can be run on a pretty cheap – in a cheap way. Like you can reproduce it. You can run the model. And so certainly doing the thing they have done can be done cheaply and that will be important for AI.

[00:06:25] What is less clear is the amount of investment it took them to do it, which is people are sort of comparing apples to oranges where they're like, oh my god, our foundation model spent so much money and they, at the end of the day, have produced something that you can do cheaply. But obviously if you just like follow everyone else's research and then do something at the end, like it'll be cheaper. Of course that's cheaper. Yeah.

[00:06:50] Jumping in and like the last mile to say and build something a lot faster with your sprint certainly looks faster than doing the underlying work, right? I mean there's been the juicy irony I think, you know, that the media class has enjoyed that it feels like open AI is like, how dare you copy our technology, you know? You can't inhale everything in the world. Only we can inhale everything in the world.

[00:07:14] The irony is not lost on those of us who create original content that has become, you know, the high quality training data that these models use that now open AI and, you know, these closed models especially are getting very angry about something being trained on top of them. Yeah. So, you know, I think open source remains super powerful. I think Anthropic and open AI are really going to have to work to show that they can stay ahead.

[00:07:40] And so this is going to put pressure on them to release new versions faster. And I think we've seen what you could fairly call defensiveness from both open AI and Anthropic. Dario at Anthropic wrote a long note on sort of a tangential point that I think we can talk about, but sort of, you know, responding to everything that was going on. Open AI has been tweeting about this. You know, this is the new world where everybody like, you don't ignore your competitors like big news. You sort of like get in the Twitter trenches.

[00:08:10] You have to jump in with your take. Yeah, exactly. What did you think of the reaction? I found the reaction from the little tech boosters actually to be one of the more positive sides of this. Alex Rampell over at A16Z called this a Sputnik moment, but also, you know, compared that to how Russians getting ahead of the U.S. and the space race led to like a massive wave of U.S. innovation in space technology. And so I think there's a lot to be said for this really opening up the application layer.

[00:08:36] If foundation models get cheaper, there's a ton of value to be added if you can build killer apps for it. So I don't really see this, you know, as like a death knell for the venture space that's gone all in on AI, if you will. I think there's plenty of funds that have much to invest in on the application layer that this actually opens up a lot more opportunity for founders and investors to really get into the space. If if this if the foundation model layer is where it is now and everything's open, you can build on top.

[00:09:02] I like I like the Andreessen Horowitz, Alex Rampell essay. I love the Sputnik moment was smart. Madeline, you've been talking to investors all week about this. What's what what's your read from them? I mean, a lot of them are kind of taking I think some of them can get credit for this. Some of them can't. But a lot of people are saying, you know, we always knew that the value really wasn't going to accrue at the model layer. You know, it's really at the application layer. We knew this the whole time.

[00:09:26] So I've gotten a couple, you know, takes of people maybe talking their own books a bit about how smart they were about this whole thing. But what firms are most exposed to foundation model bets? Well, certainly Thrive is pretty exposed. They've just given so much and invested so much into open AI. Menlo Ventures is probably is one of the largest backers of Anthropic outside of the big tech companies like Google and Amazon.

[00:09:52] A16Z obviously has backed a lot of application companies, but they have the opposite. But well, they've written really big checks into Mistral and open AI through the secondary sale from employee tender. So they're kind of wrapped up in everything. Have it both ways. Yeah, exactly. Yeah. So they're kind of on the application side and the model layer. So they're and maybe that's a too big to fail situation, but they have a lot of exposure. I know Sequoia and Tiger Global also bought in the open AI share. So share sales.

[00:10:21] So we're going to see. I mean, the big firms obviously do have a lot wrapped up in this. So maybe this is, you know, the little tech victory lap, if you will. This should almost go without being stated, but I think I need to say it. It's like AI being better and cheaper is obviously good for AI broadly construed, right? Yes. Bad for people only differentiate by having the smartest and fastest AI in town.

[00:10:48] But in the fact, consumers are more likely to use something that is better and cheaper to use. And so I do think some of the whiplash confusing reactions. Isn't this the Jevons paradox point? Yeah, exactly. Something gets cheaper. More people use it and demand goes up. I mean, I feel like that's also been just seen with, you know, different service adoption for tools throughout startup history. Yeah. So it was an interesting, it was a market test.

[00:11:16] It was definitely, I think Silicon Valley was rightly like by the dip. And I think we're seeing that borne out. It was sort of like a freak out. It felt like outside of Silicon Valley, even though Silicon Valley reacted. It was like an interesting cultural moment. Well, I think it's important to circle back to Dario's point, all really being about export controls and how the leadership here isn't necessarily a given that America will have this leadership.

[00:11:40] And obviously he's team safety with Anthropics, whole, you know, safe AI, controlled, top down kind of strategy. But he's saying, no, actually the export controls are working because they didn't get quite as many chips. And if we, you know, get, you know, strengthen control over this part of the stack and keep it really in the West, then we can stay ahead.

[00:11:59] And I feel like in this sort of open, opening up moment and, you know, big, you know, revoking of the controlling Biden AI executive order that, you know, the administration's doing. Everyone's all, all in EACC all the way, you know, like, like at this top down kind of safety mindset isn't necessarily what's going to lead us forward. Well, there are two pieces of this, like, you know, the export controls. First of all, it's hard not to read is talking his own book, right?

[00:12:29] Like certainly open AI and Anthropik would benefit from one less major competitor in if we deny China NVIDIA chips, they'll have a harder time being competitors. Therefore, we have less competition. Great for us. Very anti free market capitalism.

[00:12:45] Awesome. But, but potentially, you know, I mean, I, you know, probably agree with the export controls just in the sense that like everything with China seems to flow one way where they get to take all our stuff and then our companies can't operate there and they, they, they benefit from our work.

[00:13:03] I mean, that said, I do think the open source nature of this makes things a little different because I mean, deep seek is giving it away, you know, and like, that's the ultimate sort of good thing. Ben Shapiro had a terrible tweet where he is like the deep seek surprise is a fantastic reminder that our enemies are motivated and aggressive and that our technological superiority is not guaranteed.

[00:13:27] I guess. Okay. That's a fair take, but Bill Gurley, I thought had a very moving response, which was why are they enemies? No one that works at deep seek is an enemy of mine. The fact that they love open source makes me think I would gel with them quite well.

[00:13:44] And I mean, I think, you know, the right China message is that troubled by Chinese governments, obviously free speech policies control, and then also their industrial policies. But Silicon Valley probably has more in common with the average tech person in China than they do with the average American, you know, like that's, that's the ongoing contradiction.

[00:14:07] Like, while deep seek was able to operate fairly under the radar as like, I don't know, hardcore technologists, and then give it all away. Yeah, they're very in sync with a lot of people. I was like, that's like the classic Silicon Valley story. That's the classic Silicon Valley ethos right there. But you know, then as soon as deep seek becomes, I don't know, China's most important AI company, then it becomes about sort of nation states competing and that, you know, these things get complicated.

[00:14:35] We also worth mentioning that now Alibaba is like coming in claiming that they have like something even better. So to be interesting, again, we don't know what's happening in China. So it's like, is this all like, oh, they're ready to like, announce their sort of big hits to make a message or who knows. But I agree with Gurley fundamentally that if they're giving it away, how evil can that be? You know? Yeah.

[00:15:00] I mean, I was talking to a lot at basis set yesterday who just said the best line to me, which was, you know, people to make this too much of a big deal because they told you the recipe. Just take it and make the dish. Like they gave it away. You know, why are we spreading over how it happened? And they've given us a gift. We can build on top of it now. Right. The other piece of Dario's essay, you know, the safetyism. I mean, I probably would have agreed with him. I probably agree with him.

[00:15:27] But like, man, that feels like a lost, a lost fight. You know, it's like if we can't even keep the executive order, this sort of the idea that we're going to like hold on to AI, keep it in sort of a tightly held set of hands and make sure it doesn't get out of control is feeling less and less likely.

[00:15:45] I mean, I do think there's a world in which we look back at 100 years over this time and we happened to build AI in this very anti elite moment, which meant it was totally ungoverned and we sort of let it run wild. And we're like, oh, my God, the one time we should have been sort of more sort of top down and sort of thoughtful about how this all came out. We just happened for totally other cultural reasons to be in a like, you know, let everybody do everything wild moment.

[00:16:13] And like, so we create, you know, the all powerful AI that that destroys us all. You know, hopefully that doesn't happen, but but it certainly makes it a little more likely. It's feeling very science fiction novel. See what they did take, you know, prediction. Exactly. Yeah, you get it right. Exactly. It's classic that you get sort of destroyed by your vice, right? Your vice, you know, it's like we're in this sort of like burn it all down mode.

[00:16:37] And the thing that destroys us is the thing that, you know, we didn't quite appreciate coming that needed a little bit more of collective control. Mark Andreessen would hate that take with all of his being. But this is not the Mark Andreessen podcast. So we can espouse it freely. Well, moving on from China and DeepSeek and the crumbling of the American empire. Let's talk about fund returns. Let's talk about funding data.

[00:17:05] Eric, you got an exclusive report from AngelList this week. Wasn't the prettiest picture, right? Of the venture market. I don't know if this is moving on from the crumbling of the American empire at all. I mean, you know, basically, yeah, AngelList gave me their annual report, which I find super useful on the state of venture from their platform.

[00:17:29] You know, obviously, some of the firms on AngelList are smaller, but they see more and more of the startup market and generally have a good sense that I think matches with what everybody else sees. And, you know, one big take is that basically, if you wanted to mark up your venture capital firm, you needed to be active in 2021 and the time around that. Like everybody got their markups then and then it's been flat. If you started after 2021, you basically don't have many markups.

[00:17:56] And if you started before it, you had all your markups then and you're sort of killing time. I think the other big take for venture capital firms is that they have, you know, TVPI. They've marked up their holdings, but they don't have much DPI. They don't have distributions. And, you know, you can't eat TVPI. Ultimately, LPs want DPI. And so I called this year a crucible year. You know, it's because one of two things is going to have to happen.

[00:18:23] Either those marks turn into exits and they can pat themselves on the back or, you know, they're going to have to eat crow, at least to some extent, and say, oh, you know, we thought they were more valuable, but actually they're never going to turn into distributions. And if they don't turn into distributions, you know, that's potentially going to undermine LP's view of the venture capital model.

[00:18:49] So we really have these two forks in the road, which is some of these holdings are valuable around what the VCs say they are. And we get some exits or we don't. But it feels like it's been going on long enough that 2025 should be the year where we have that reckoning. I'm skeptical that these markups will turn into distributions. Yeah, I was going to say, what's your take on this?

[00:19:16] I mean, obviously, there were some startups that also kind of, you know, scraped along and if, you know, tried to hold off before raising again, just because due to the poor market conditions, like, are there many signals that you could see us, you know, moving into a better exit year this year? I mean, the couple of IPOs that we've had so far not been promising. Yeah, I mean, everybody thinks we're going to have better M&A.

[00:19:39] So maybe, you know, we'll see some activity there and like that could help, you know, your billion dollar company could sell for 500 million, though, with preference tax and anything, as we've talked about on other episodes, you know, that's not necessarily a good outcome. IPOs so far haven't looked good. I mean, basically, you know, if you hold Databricks, you know, OpenAI, Stripe, you know, the big winners. And so those funds, you know, firm, you know, LPs can see that they're fairly liquid.

[00:20:08] They do secondary deals. Like, I mean, we talked about in your UTIMCO piece, which people should subscribe to Newcomer for and go check out. You know, Thrive is doing well, partially because they're doing these late stage deals where they're raising again and they get to market them up. But like if you're a firm that's not in one of those big names, I don't think people are very optimistic that the random, you know, three billion dollar unicorn is going to deliver for you.

[00:20:35] And obviously a couple of them will, but I think a lot of them won't. Yeah, it seems kind of like the power law, but on steroids, if you will. Right, exactly. Right. Yeah. How little DPI is like, I don't know, on some of these funds? Like, even like the 75th percentile from their oldest vintage funds, 2017, has only returned 0.64x. So you don't even have all your money back yet, like eight years later.

[00:21:05] You know. Yeah. If you're going on a 10-year cycle, that's not feeling right. I know. Nobody wants to be on a 10-year cycle anymore. So, you know, the only other, you know, AI, obviously, lots of AI deals. There is the counterintuitive point that valuations are basically like 10% up, but you have to remember that's because fewer people are starting companies and raising money. I don't know. What was your biggest? We've done two UTIMCO pieces.

[00:21:32] What was your top take or one or two takes from those pieces? The biggest takeaway I've found, at least from, you know, doing this last year and then now doing it again a year later to look at this data, is really just that the market downturn, like, is much more visible now. And obviously, you can see exceptions, but it's people who are doing things a little bit differently, like thrive, like getting into these growth stage rounds and not performing like a traditional venture firm, if you will.

[00:22:00] And otherwise, like, you know, kind of doing these more growth stage deals to where their returns look better faster, you know? I thought your take was just going to be Union Square Ventures is a singular firm. I don't know. You look at the UTIMCO performance and they're just like, they have so many good funds. Like, you just wrote about Blue Yard, a great firm that has a really strong fund that's competitive with USV. I think they were in the creator of Filecoin. Kudos.

[00:22:29] And it seems like they're doing well. Their 2016 fund is great. Yeah. Right. But USV, it's just like hit after hit. Like, 2004 is the best one on our chart. I think, what, USV Opportunity from 2011, looking at it, might be the second best on our chart. You know, it's just like they are, yeah, they're pretty strong. It'll be great. Yeah. Well, I was going to say also, I will say New York-based IA Ventures, also incredible returns.

[00:22:57] Not quite to the level of USV on the charts per UTIMCO's internal calculations on IRR, but still really consistently up there with each of their funds. And so maybe this is big New York. Maybe that's my other take, you know? I'm actually, I'm calling it now. Bullish on New York. IA Ventures and USV, the steadiest, biggest returns in the UTIMCO portfolio. Turns out financial mind is still valuable when making investments and that just like. Yeah.

[00:23:27] Well, yeah, definitely worth subscribing. Yeah, check it out. Newcomer.co. Check it out. Cool. Cool. Our deal of the week. Deal of the week. What's our deal of the week? The deal of the week. Exciting deal of the week. Eleven Labs finally closed its own funding. You know, we're big fans over here of Voice AI, although different Voice AI companies, to be fair. But... No, I think Vali, you know, our partner in Cerebral Valley, my good friends, Voice Games. I think they use Eleven Labs. Oh, great. I think they're bullish on them. I don't think it's competitive.

[00:23:57] Then it's all love. Fantastic. Eleven Labs raised $180 million Series C round, which, you know, raised its valuation big up round to $3.3 billion. And it was co-led by Iconic Growth and A16Z. Yeah, Voice AI is hot. Lots of market opportunity. Eleven Labs is growing quickly. Doing well. Big up round. That's our show. We had, what, a couple weeks ago, Red Note taking over the U.S. App Store. Now we've got... Now it's DeepSeek.

[00:24:27] DeepSeek. What Chinese app will we be bringing to you next week on the show? We'll wait and see. All right. See you next week. See ya. Bye.