Episode #17: DeepSeek, Importance of Running a Fundraising Auction, Growing Seed Round Sizes

If you want to make technology broadly available, essentially told us that eventually it has to become cheap and ubiquitous. In many cases it was an architecture innovation that allowed people to dramatically scale these things out. And I think time will tell whether Deep Seq's reinforcement learning approach was the architecture innovation we needed to dramatically lower costs and make AI more widely available.

It's the week of January 27th. Welcome to the Learning Corner, a Precursor Ventures podcast.

Welcome to the Learning Corner by Precursor. I'm Mia Farnam and I'm joined by two of our Precursor team members, Charles and Marina, where we will be talking through the articles and Goodreads that caught our team's attention over the past week. Obviously, Chinese company Deepseek has been the talk of the tech ecosystem for this week as it debuted an AI app that rivals OpenAI's ChatGPT and claims to do so on a fraction of the costs than our more well known foundational LLMs have burned through. There's been a tidal wave of content flowing around Deep Seek in the last handful of days, but one that caught our eye came from Stephen Sinofsky's hardcore software newsletter titled Deep Seq has been Inevitable. So in this piece Stephen shares the POV that a company building a low cost AI model was bound to happen eventually. It's really just a matter of time and who would be the player to do it. So lots of different directions to go here. More generally, the Nvidia market cap drop, questions of Deep Sea being truthful in costs building the model, trickle effect onto efficiency concerns around the US LLM companies and more. Really all topics that are listeners can get pretty robust rundowns on through other sources. But Marina, you have an awesome background in the semiconductor space, so I would actually love to start here with you and asking how you reacted to the news.

I'll start with Nvidia because that's where I'm getting the most questions. I think the hysteric about Nvidia is a bit overblown. However, I think the stock drop of 17% or whatever it was, I think that was warranted on a relative basis. I think 70% is a pretty fair drop given Deepsea's new model and its impact directly on Nvidia's sales and that has to be taken into account. Deepsea basically just proved that companies building don't have to blindly buy more and more Nvidia GPUs to make their models better, and they don't also have to buy the newest and most expensive version every time, which is a huge part of Nvidia's business model. So of course that means they're going to get slightly less revenue. I have to mention the flip side with, you know, more accessibility and more affordability and building these models now, that ultimately also means more competition and more people that will start buying Nvidia GPUs to build even a more efficient model and there'll just be more usage. And just because deepsea didn't necessarily use the newest version of Nvidia's chip because they didn't have access to it, supposedly that doesn't mean that Companies in the US can't refine Deepseek's model and use Nvidia's more pronounced H100 chip to improve their models. So I think the market has to take into account the huge change in Nvidia's business model and what the customer profile is going to look like, but I think there are some wins and losses. But focusing on early stage tech in general, I think the Deep SEQ news had me pretty optimistic for the future. I mean, something we've talked about, a precursor a ton, is that it's hard to invest in this space right now because of how rapidly things are changing in AI every day. It's hard to guess what the future will look like. And this week that change was Deep Seeks model and I'm excited for it. As you guys know, I've thought a lot about the massive amounts of energy demanded as we add more and more compute for AI and what this means from more from a sustainability standpoint. I know there are companies building in different directions, whether they're building a new chip to compete with Nvidia or trying to change the infrastructure surrounding that chip. And I've always been rooting more for the latter. And deepsea did that and took a massive step in that direction and is using significantly less chips and invest less power. And that makes me excited. And personally I hope that the next person or company can iterate on Deep Seek's version and make it even more efficient. That all said, of course there are questions on national security of Deep Seek and if the administration will do anything about it, but in the short term I am really happy about the news and I think it's a step in the right direction.

Thanks Marina. Charles, anything to add there from your perspective?

The reason I enjoyed this article is I too felt it was inevitable in the sense that we had kind of a monolithic view of how AI was going to develop. And I think in tech, in so many, many instances, when you have a consensus monolithic view of where technology is going. There's always somebody who takes the other side of that bet. And if the original bet was really expensive chips, tons of training data, big data centers, to me it was inevitable someone's going to say, well, I don't have access to those resources and I'm facing a different set of constraints, which is I can't get access either financially or politically, to the current generation of hardware. I don't have unlimited access to capital, I don't have unlimited access to power. So I'm facing a different set of constraints. And I think that's what makes tech to me so interesting, is that most people build in response to constraints. And I think with Deep seq, you see a different set of constraints can produce a different set of outcomes. And my hope is that the approaches that people use to building AI models continue to diversify and that we get more people running more experiments, because I think we're still really early in this tech, and I think the industry would benefit from having more people pursuing different approaches at this point, because we still don't really know what works best.

Steven says in the article, quote, the history of computing is one of innovation followed by scale up, which is then broken by a model that scales out when a bigger and faster approach is replaced by smaller and more numerous approaches. Anything else that you might want to add to that kind of hypothesis?

I think the Internet has lots of predicates, whether it's mainframes and PCs. If you want to make technology broadly available, essentially told us that eventually it has to become cheap and ubiquitous. We got that with the Internet, we got that with mobile phones, we got that with PCs. These things started out really expensive and eventually, ironically, in many cases, it was an architecture innovation that allowed people to dramatically scale these things out. And I think time will tell whether Deep seeks reinforcement learning approach was the architecture innovation we needed to dramatically lower costs and make AI more widely available.

The second piece is a LinkedIn post by Aaron Harris titled Always Run an Auction. In this post, Aaron advises founders to create competitive dynamics during fundraising by running an auction, essentially positioning their round as an opportunity that investors need to move quickly on. Obviously easier said than done, but Aaron argues that competitive tension and building momentum is critical for getting the best possible terms and conditions for a raise. He shares practical advice on timing, messaging and managing conversations to create urgency, while cautioning founders not to overplay their hand or alienate investors. So Charles Aaron emphasizes the importance of building competitive dynamics during fundraising. How do we advise founders in our portfolio to Approach this. Additionally, this might seem well understood by investors, but for founders, what are the key signals that indicate an investor is truly engaged versus just, you know, taking a meeting? Or what's the difference between being preempted and actually just getting an early phone call to check somebody out?

Yeah, I met Aaron when he was doing the series, a program for yc and I feel like he gave a lot of the companies I knew really good advice. And so I'm glad he made this post. This one plus First Round wrote a post a long time ago about fundraising and waves. Those are probably the two that I share with people. And over time I've become sympathetic to the plight of founders. I think most founders do not enjoy the fundraising process and anything that can make it shorter and faster is usually desired. I think it's really important to run a good process and I would say one of the biggest challenges I have is convincing founders that they should do that. Because if you read this article, running a good process is actually a lot of work. It's a lot of work to try to keep everybody on the same schedule, to try to be organized enough that you can move people through phase stages. But I think the outcome of running a good process is you get really good price discovery and figure out what does the market think your company's worth. As the article says, if you think about demo day as the ultimate auction, you get hundreds of investors in a single place trying to make decisions in a really compressed time period. That's about as much leverage as you can get as a founder. But most people don't have the benefit of being part of a program that has an oversubscribed demo day. I've walked lots of founders through this process and they go, yes, this sounds great. Then they get an inbound email from an investor that they think is interesting. Like, we're getting preempted. I'm like, you're not getting preempted. You got an email, somebody heard what you were doing and sent you an email. And for a long time, venture didn't do as much outbound as they do now. Now I think most firms do a ton of outbound. So getting that email doesn't mean what it used to mean. But I think for a lot of founders, the prospect of somebody A being interested in your company and B, being able to raise around without having to do all of the process oriented stuff that Aaron says, it's very attractive. But I've seen things go wrong. One is you spend a bunch of time with one firm and you don't get a term sheet and then you're like, great, I have to start this whole process over having lost time and momentum. Or B, you get single threaded with one firm, they give you an offer, it's not great, but you have the offer in hand and you're like, do I want to walk away from this offer or do I want to go run a full process? And you know that usually comes down to the risk tolerance of the founder. And so I think when you run an auction, you're getting feedback along the way about how interested people are in your business. And I think that's a better set of facts from which to make a decision about, hey, should I take this offer from firm or firm B than one conversation and one term sheet.

Being on the operating side, I remember when you're getting inbound from an investor and suddenly like you blink and then you're in a diligence process kind of by accident and you find yourself spending all of your team's resources to fill out large diligence questions, building out cohort analyses. And I think investors are really good at selling that there's something at the end of this when in fact maybe they've made a decision two weeks prior that this isn't quite it, but they're going to round out the process in the end to tie it off in a nice bow. And so it's, it's always fascinating to hear what are the signals that show maybe pump the brakes or this is not actually the diligence process that you have to exert so much of your resources on. Marina, I wanted to ask you, what are maybe some of those signals that founders can look for?

I think it's great if you have that inbound interest and you want to continue to build on it, like Charles said, like the article said. I think where I try and give a lot of advice to founders is to be very careful in the words that you use when you try to build momentum and try to create this FOMO with others, other investors. I think you still have to be very careful not to lie or especially not even to over exaggerate to the point where an investor would consider it a lie. So just always remember, VCs all know each other. So if you say one fund is in diligence or another fund has committed, we're going to talk to those people and just confirm that it's true. And the second that that fund says anything to the contrary of what you said, it's a red flag. And VCs sort of look for any reason to say no. So the last thing you want to do is to hand them that reason yourself. I typically try to give some grace to first time founders because all this lingo is very confusing and made up. But the most common mistake I probably see is a founder saying they have a commitment. We try and tell founders to be very careful when using that word. We tell them, you know, you don't really have a commitment until you have a signed term sheet. Even if a VC tells you verbally, yes, we are going to invest just until you get the lead. Just be on the safe side. Don't call it a commitment until you have that signed term sheet. If you do want to communicate that sort of momentum and create fomo, you could say something like, we have X amount, soft circle, just depending on terms and a lead. But the last thing you want to do is lie. And maybe it's not in any malicious way it sounds when someone says verbally, yes, we're going to invest, that that is a commitment. But it's very easy for investors to take that back. So just be very careful when trying to create momentum. But it's obviously important that you do so because you want to create FOMO around your round and you want to get it done as quickly as possible.

Even the term soft commitment is hard. And I know for founders it's something that it's maybe one of the only intangible, tangible things you can hold to when you're looking for the lead. But I found that's a tough thing when founders share me the soft commitments, because those could fall apart at any moment. Charles, question for you. Would there be advice for founders to maybe always have an updated diligence room in the back end in the case that somebody outbounds? How do you manage that?

I've certainly seen companies who've told me, well, we're not fundraising. I'm like, okay, great. Do you have anything I can look at? Yeah, we have a full data room. And I'm like, well, this sure looks like the level of preparation that a company that's fundraising would do. And I think the other part about running an auction is it forces you to do a bunch of the work upfront to put together the data room and to be ready to go. But you choose sort of when you fire the start cut. What I tell people is there are certain things you should kind of all like your cap table should generally always be up to date. Most of your incorporation docs really not that hard to keep those up to date. To me the things that are harder are like, if you don't have an investor ready, set of cohort analyses for going back 24 months. I don't know if it's worthwhile to keep that stuff handy if you're not already using it to run the business. And you might have it in a version that you could present to your board or to your team, but maybe not for external consumption. In most cases, when people tell me, oh, we're thinking about raising, I'm like, well, what's the plan? And if the plan is one random firm sent us an email and I like the partner, I'm going to talk to them. I was like, to your point, Mia, you can find yourself sucked into a process that you didn't intend to get in. Oftentimes what I try to tell people is, listen, we're either going to fundraise or we're not going to fundraise. If we're not going to fundraise, you should tell these folks, hey, right now, we're not raising. We're busy operating. We have a plan to fundraise. I will be in touch with you. And I was like, just because somebody emailed you and said they want to know a lot about your business doesn't mean he was a founder or obligated to do anything. You don't even need to acknowledge. I mean, it's polite to acknowledge the email and say, hey, thank you for your interest. We're not fundraising, but you don't even owe them that. So I get it. And I think the reason VCs send so many of these emails is because it works. Because I think it's flattering as a founder to get an email from an investor saying, you, you don't know me, but I'm interested in your business. And the real question is, why are they interested in investing? Are they looking at investing in your competitor and they're trying to make sure they have a good picture of you? Are they just trying to do a market map? You don't always know. So I, I try to help people figure out, how do we qualify this interest before we go spend a bunch of time answering questions and doing work for them.

Marina, I think you and I, we lean into outbound pretty heavily. I think one of the best questions I hear founders ask, which I think is a good indication that they know how to test out what investors are actually interested in, is what gets you excited about the space that I've been building across and what's kind of your perspective on that space? Because then it shows that they're really interested in you and not just which investors do all the time. Send out a ton of outbound like 16 notes in four hours to as many founders as they can find online. Our final topic is from Crunchbase News titled seed rounds got larger through the downturn. Why is that? So while much of the venture funding ecosystem has contracted during the downturn, seed rounds have really bucked the trend. So growing larger even as the broader market slowed down since 2020. There's a few things to note here to continue the conversation over. So while seed deals by size are getting larger, they're also getting harder to come by. It's a little bit more competitive. Seed round sizes of more than 5 million make up a large share of seed rounds. And since 2022, those larger seed rounds have received more additional investment than seed rounds of 5 million or less. The larger rounds could be due to the higher benchmark required of founders in order to graduate to series A, resulting in, you know, longer seed stages and requiring more Runway. What's very clear is that there's a variance on valuations at seed, and that's really high still. So investors are still all playing the stage differently. Charles, I know there's different vehicles of which to carry out seed rounds, whether it be priced or safes, but the median seed round size, according to a couple of articles, was around 1.5 million in 2011 and has since crawled north of 5 million in today's market. So from your perspective, what might be the pros and cons of that round size and the growth that it experienced?

I think there's a bunch of factors at work here. I think one of them is purely mechanical. I've seen companies in our portfolio, companies in my friend's portfolios. It used to be you did one seed round and that was it. But I think it was Hunter who said a while ago, like, seed is a face. And now what I see oftentimes is a company will do a million dollar safe as a pre seed, a $3 million first seed, a second seed round as a price round. They will roll the whole thing up and convert all the safes and they will announce it as a $6 million seed round. People will go, oh my gosh, that's a huge seed round. I'm like, well, it's only, only a third of those dollars are actually new money into the company. And I think one reason to do that is purely mechanical, just to account. Second, I think most founders want to project strength. And it sure sounds better to say I raised a $6 million seed round, even if not all of that money is new, than to say that I raised $2 million seed round of new capital. And so in the same way that I've seen journalists push for more answers on revenue and valuation when they report on fundraising, I really hope they start pushing people to disclose, well, how much of this seed round was new capital versus the conversion or roll up of previous capital. I don't think it's in most founders incentive to share that number, but it would be good to know. I think that's part of it. Second, I think we've been in this weird thing since 2011 where every year I feel like it was the series A bar has gone up. I'm like, well it goes up every year. Life is hard. And I think some firms reaction to, well, if the series A bar is higher, you sort of have two choices. Give companies the same amount of money or slightly more and ask them to be more efficient, or find founders who are capable of clearing that high bar with less money, or give companies more money with the belief that if we give them more money, it increases the probability that they will clear that high bar. And I think that is an untrue statement. And probably some of the most heated arguments I have with seed fund investors are I think it's a crime to give these companies 36 months of Runway pre product market fit. It's just too much money. It creates bad habits, they tend to spend too much that you lose the urgency of finding the answer. And I'm a big believer that until you have product market fit, you should keep the burn really low. And I find that everyone says, yeah, we're going to raise 36 months of army, don't worry, like we're not going to lean into spend. And six months later you're like, why are there 12 people at this pre product market fit? Well, you know, launches in six months. So we hired a marketer so that we could be ready for launch, but we don't even have a product. What do you mean launch launches in six months. And I just think the answer to a higher series A bar is not to give companies more money up front. It's a demand, higher performance. And you all know me, I've leaned very hard into you need to make this money work and to clear the bar, no matter what the bar is. And I've had almost universally bad experiences with these companies that raise really large seed rounds before they have product market fit or even a notion of what they want to build.

There's Just so much pressure placed on founders when they raise these larger seed rounds. Marina, last question for you. They happen though, raising these big seeds. What are things that really strong investors can provide as tidbits or tips for founders who've done that? Especially maybe if they're first time founders to just make sure that they're staying ahead of the ball. Because I've found that investors who join those rounds, most of them won't even pipe in throughout the burn until the very end when they've realized that nothing really has materialized and then they're blaming it all on the founders.

I always think back to Charles, that piece you wrote a while ago about the last 250k, I think that applies pretty strongly here. Summary. If you only have four to six months of runaway left, you act very differently than if you had 24 months. You focus on specific things. You have a much smaller team that can execute a lot faster. If something doesn't work, you move on and you break things and you move fast. And I think that really applies regardless of how much capital you have, like your pre product market fit, you don't know what works yet. So test different customers, test different models, do many experiments and then you know, that's not work quickly move on. And that's sort of a benefit with a smaller team is that you can all collectively move on pretty quickly. So always a proponent of breaking fast. I mean it applies to deep seek as well. I mean one of the reasons that they figured out this model and the US necessarily didn't is because they had all those limitations around resources and capital. They were just having such a different focus of okay, we have such little to work with. How do we still compete with everyone else that's out there with the limited capability? And they were able to figure it out. The US wasn't focused on operational efficiency, see, because they had endless amounts of capital. So it's a very different two ways to think of it. I think just, you know, focusing and breaking things. If it's not working, if this team isn't focused on the right thing, let them know and move beyond that.

A great way to tie it all back together. We are all out of time for today. Thank you Marina. Thank you Charles for joining in today. Thank you to our listeners for jumping in. Please like and subscribe and we will talk to you all next week.

Thank you. Bye everyone.