Josh Wolfe on AI Infrastructure Glut

prompted you to send this note out to founders earlier this month?

Honestly, it was a, a confluence of a few factors. The first was capital markets. The second had to do with some of the political interventions, particularly around tariffs. The third was physical infrastructure, something you just mentioned as related to the data center build and the huge CapEx spend. And the fourth was layoffs and really the market's reaction to those layoffs and specifically Block. Um, the capital markets, I think the first piece was this divergence between equities and bonds. And we are micro-investors. We are trying to bet on startups. We're trying to bet on the most amazing founders, but ignorance of the macro is no virtue. And so we've always paid close attention to just market signals. And so you had a dynamic where you're reaching all-time highs in equities, particularly those leveraged to AI and tech, and that is the zeitgeist of the moment. And then you had a low point on the 10-year yield, and typically that occurs when people are buying bonds. And usually you can say with a sort of brushstroke that bond investors are usually a bit more risk savvy than equity investors. Equity investors typically narrative driven, bond investors more cash flow driven. And so sometimes the signal weights a little bit more heavily when you see that divergence. So that was number one. Since then you've started to see, you know, I think we published this roughly March 2nd, so we're call it 10, 12 days post. Markets last week I think were down 2, 3%. The VIX went from 17 or 18 to 29. Some of that, because of the confluence of war, although we wrote this right after the first strikes that weekend in Iran between US and Israel. And so the markets felt a little bit shaky. And when your parents are talking about AI and everybody you know is talking about NVIDIA, it's that classic signal that just, we're probably not at the bottom. Okay? So that was the first thing. Right. Second thing was the physical infrastructure. And there were two pieces here. One, if you go back to even the election here in New York and you go back to the Virginia governor election, the key thing that people were campaigning on was affordability. And all of these states around the country were doing everything they could to attract data centers and AI because they thought that it was gonna create lots of jobs, and it has and will in many cases. What they didn't anticipate was that electricity prices would be spiking. And so now you have something like 300+ moratoriums on data center builds in across 30 states, and that's a new signal. Now, whether or not those things pass, whether they are for electoral, uh, fodder to be able to show that a, a candidate is helping to reduce the cost of your daily living. All TBD, but it was another signal. And then you just mentioned the Oracle project, Oracle OpenAI teamed, funded and, and fueled rather by Crusoe in that Abilene, Texas plant. Maybe Meta takes it over, but you're starting to see these very large commitments upon which very large capital raises were justified scaling back and saying, wait a second, maybe we don't need this compute. So that was sort of the big second physical infrastructure one, which will lead to a piece of advice that we gave for, for our founders. Third was tariffs and just the nature of that. Supreme Court shot it down immediately. The Trump administration said, that's great, we don't care about the judicial branch, we're using another mechanism. And the implication there is if costs rise, not just on an individual company, but the dependencies concatenated over different suppliers, two suppliers, and having that risk mindset to think about that, we thought was, was extra important. The fourth was the market's reaction, particularly to the widespread layoffs from the note from Jack Dorsey. Market was up and what you saw was suddenly the equity markets were saying, we're going to reward you if you can cut costs, particularly in human capital and human labor. And sure enough, the job report came out last week. I think it was expecting 50,000, ended up being something like 90,000 down. People are losing jobs right now at the rate of about 750 or so a day. So the job picture is not that rosy. I think you have 60% participation in the labor force. Unemployment's up to like 4.4%. Again, we're micro VCs, but understanding sort of all these macro dynamics, it just said, let's be cautious. I have no idea if the markets are going to crash. I have no idea if we're going to be in an ascendant bull market. What I do know is it is wise to take precautions because the cost of doing that is virtually nothing, but the cost of not doing it is very high.

So you, you, you wrote this and I thought the interesting thing is you kind of echoed what you just said, which is that, hey, I'm, I'm not trying to be trying to call the top here. I'm just trying to say, be careful, run some sensitivity analysis, you know, get prepared. What was the reaction from founders? I mean, I'm curious what emails you got back.

So we have a, a big founder list of what we call the Lux family. And I gotta tell you, everybody was like, thank you so much for sending this. Super thoughtful. Really appreciate it. A bunch engaged with the various parts of our, our, our infrastructure here. So whether it's finance, operations, accounting, people that are involved in capital markets and bank lines of credits, because the basic advice we had was exactly this. We have no idea if something bad's gonna happen, but it's the same colloquial advice I give my kids on a rainy day or a cold day. You need an umbrella, you need a sweater. It's better to have it and not need it than need it and not have it. Same thing for companies. So check your runway, double test your assumptions. This is all basic common sense.

But have, have they tightened their, their bu— I mean, their budgeting forecast? I'm curious, in the last 10 days, have they tightened their budgets as a result of the sensitivities that they ran?

No, nobody, nobody was catalyzed into extreme reaction yet. But I do think that people are basically saying, okay, wait a second, we're paying attention to growth, growth, growth. We're paying attention to hiring, we're paying attention to shipping. Let's make sure that we have no regrets, that there's no SVB-like environment that happens. You know, it happened 2 years ago where we're suddenly vulnerable. Let's just think about where are our exposures. And the number 1 exposures that you have control over is how much cash you're spending. And if it's in pursuit of growth and it's working, great. If not, maybe you wanna rein some things in, make sure you have cash for a rainy day. Number 2 is look at your covenants on SVB debt or, um, any of the other venture debt lenders and make sure that your terms are good, that nothing's gonna get pulled on you, swap lines, et cetera. Uh, number 3 is thinking about that concatenated risk in your supply chain. If you're dependent on a supplier and suddenly they're dependent on a supplier and something with the tariffs or any kind of interruptions creates a spike in somebody 2, 3 layers down, you're gonna come to the board and say, we had no idea this was gonna happen. Shame on you. I always like to say failure comes from a failure to imagine failure. That is a creative act. The idea of imagining what could go wrong so that you can do something by throwing time or money at your talent to prevent it. So that was the other big thing. And the other piece, which is a very human piece, all of our companies, they don't do deals with companies. Of course they announce so-and-so does a deal with Oracle or with, uh, Nvidia, but you're doing a deal with a person. It is a human that is signing a contract that's greenlighting you, giving the thumbs up. If that person gets laid off and you're seeing Oracle lay off thousands of people or purporting to, You're seeing Salesforce laying off thousands of people. If your key contact, if your key champion is suddenly gone, no fault of your own, and they're laid off, you're in trouble. So the advice here is make sure you have deeper relationships than one single node of risk. Make sure you know 3, 4, 5 people in the organization, because if the person that you're working with leaves, you're not entirely screwed.

I wanna ask you a, a, a related question on, on the topic of a story that we published this morning. So we, we wrote about, uh, well actually late last night, we, we published a story about Anthropic looking to work with private equity firms. And I know that you follow the credit markets and you also follow the private equity sector pretty closely. And I'm curious from the conversations you're having what the vibe is amongst private equity firms right now, because I mean, we have this SaaSpocalypse that is ongoing in the public markets. I imagine it's not a great time to be buying because you just have no idea how the enterprise values of these companies are going to shake out. You also have the companies that they may have bought during the ZIRP era and taken on all this debt and loaded it up. All that debt is probably coming up on refinancing at some point in the next couple of years, and it's a totally different environment. What is the, what is the vibe right now amongst private equity and the people you talk to?

Well, you have to also look at the stock prices of many of the big PE players that were hit because of the presumption, or in some cases the verification of exposure to many of the software names. Um, Apollo was probably the least hit because they had the least exposure to it. I think everybody is looking, and they're well diversified. None of these platforms are, are going outta business. Even the growth equity firms like Vista or Thoma Bravo are gonna be in really good positions because they'll also be able to take fresh capital and buy things for cents on the dollar, recapitalize things. So I'm not worried that this is like some bubble disaster in SaaS. We've long prophesied that there was gonna be a shift away from SaaS and enterprise software into deeper tech and harder tech, a total self-serving view that has hence, you know, for, uh, proven true. But I think the private equity guys are savvy enough to be able to pounce on opportunities that are super low priced. Uh, and you'll probably see a lot of consolidation. So there will be people that do have really messed up capital structures, debt that does need to be refinanced, that might happen at higher prices, but there might be other companies either in their portfolios or adjacencies that aren't as indebted, have cash, have growth, and you'll end up seeing a lot of mergers. So I don't think that there's any sort of linear prediction that you would say here that the private equity guys are screwed in any way. I think what Anthropic is doing, uh, with Blackstone and others, or, or is purporting to, is actually very wise. These are huge aggregators of hundreds and thousands of businesses. And the name of the game in private equity, particularly if you're doing a true levered buyout, is how do you squeeze efficiencies out of the business? You can lay people off, you can get, uh, pricing power in a aggregate buying, you can get pricing power if you have an indispensable product and can raise your revenue. but there's a lot of things that people are typically looking at as levers. Same thing with like an activist investor coming in. But if you can use AI as one of those tools, now there are people that are using AI just to lay people off because it's part of the narrative and people will accept it, and you're gonna see a lot of that. But there's people that legitimately say, we can actually reduce our spend on SaaS, uh, instead of having 60 or 70 vendors doing everything from HR to payroll to, um, lead generation and marketing, we can do some of that internally. And that's why you've seen the SaaSpocalypse. And in other cases, uh, I think Ed Roberts is very smart to say if we can go to one buying center, give them some discounted rate, it's a virtue, uh, uh, for them as a sponsor to these companies. And, uh, they could probably get, you know, another 500 enterprise clients, which has been the vector that they've decided to focus on rather than the consumer. It's a good hedge also, uh, relative to everything that's happened with DOW.

Now you talked about data centers and one of the things that I have been— that we had Bradley Tusk on the show a couple weeks ago and You know, he was talking about the pushback locally that comes from having these data centers and electricity costs are rising. And yet we have hyperscalers that, I mean, there's, there is no reluctance to continue to invest in data centers. And he basically suggested that we have this election cycle coming up. Politicians are very much interested in getting reelected. And so it would do them good to be able to say, hey, maybe we shouldn't build as many data centers because of the rising electricity costs. Do you think that there is any kind of a reckoning coming here on that level?

Uh, yeah, as I, as I noted in my opening comments, and Brad's a great guy, by the way, uh, very politically savvy and tech savvy, um, and a friend. 300 moratoriums right now, you know, or 300 bills for moratoriums across 30 states. That is, that is not like a small pattern. That is a wave. That is a movement. And it's been validated again because whether it was Mamdani in New York or the Virginia governor they won on a message of affordability. So that is going to be, I think, the main message going into the midterms. Affordability, you know, at a national level, federal level, we're at war, Straits of Hormuz closed, oil's up, we are releasing reserves. People are going to try to temper that, but it's going to be a force that is putting a lot of pressure. Some people will say, and, and, and by the way, I think some of the hyperscalers may actually be relieved because they've made so many capital commitments, but they won't publicly admit it. They'll say, ugh, these politicians, They're trying to thwart our growth plans. But the reality is I do think that in some companies, in some domains, what the individual company does rationally, collectively is irrational. And the amount of spend, the amount of CapEx, the amount of build for these multi-gigawatt data centers, it, it to me does not make sense. And I know there are people on the other side, good friends that run large funds that will say, no, no, no, this is not like fiber optic cables 25 years ago, that there is no dark fiber this time. There are no dark GPUs. And that is true, but each of these major large labs are competing for 100% of my attention. And I give them roughly each 20% of my market share. I open 5 different browsers and I'm typically using them or running QuadCode or Codex. And I'm just not that optimistic that all of this compute is actually going to be needed. And so I think they're relieved to see some political pressure actually push back and end up canceling some of these projects.

Okay, but cast it forward then. What happens? I mean, something gets canceled and then what, we have the overcapacity? Issue again?

Look, I, uh, when NVIDIA reported 2 weeks ago, 3 weeks ago, uh, to me it was like peak expectations because the fundamentals were amazing and the stock was down 1, 2, 3%. And that was all you need to know. I mean, expectations are fully baked into the fundamentals, uh, if not above and beyond. And that's what markets do. You know, it's, it's the difference between expectations and fundamentals. So what, what happens? I believe there's going to be a narrative shift. It's something I've been talking about for 2 years. And it really happened, uh, because I read a paper from Apple of all places, most of which, uh, Apple was being dismissed because Siri and Alexa have been horrible. They've been laggards in this. Apple's CapEx is virtually nothing in the AI game. And so people discounted, but, but you read this paper and it said, hey, we could do large language models on device using flash memory. And the same insight that I had back in 2016 with a company Zoox in self-driving cars about NVIDIA being the soul of the new machine when it was a $15 billion market cap. We shared with our LPs and some of our public market friends and said, I think you're going to have a big boom in the memory players, not just for the attach rates of high bandwidth memory for the NVIDIAS and the GPUs that are going into the stacks, but for on-device inference where 50% of your inference will be on your device, not having to go to the cloud. You'll go to the cloud to search as you might replace Google in some aspects of that, but you're going to do 50% of your cached emails, chats, text, photos, videos, health data, on device. And that means that the winners there might be SanDisk and, um, uh, Samsung and, uh, uh, SK Hynix and, uh, uh, that has— and, and Micron in the US. So that has been the big push that, uh, that I think you've seen over the past 2 years. And these stocks have been ascendant, uh, since we made that call. And so that's what I think happens next. I think you're going to get a glut in the data center and you're going to end up with a possibility and a scarcity, uh, on the localized edge inference. Great.

Well, Josh, there's a lot to watch for there. I want to thank you for coming on. That is Josh Wolfe, partner and co-founder of Lux Capital, here on TITB.