MANG VC "Round Trip" Phenomenon in AI | Apoorv Agrawal

Apoorv, thank you so much for joining me today.

Thanks for having me. It's great to be here.

All right, let's dive straight in. You wrote an amazing post recently about how there's a new VC in town, uh, MANG. It stands for Microsoft, Amazon, NVIDIA, and Google. Let's start with some baseline numbers and the premise. Can you explain this phenomenon with MANG VC and also how much of capital did they deploy in 2023?

Absolutely. So, you know, look at a 20,000-foot view. There are two factors that are driving the rise of Meng as VC. First is their keen willingness, and second is their special ability to drive the generative AI build. So on the first one, their willingness, look, AI has the potential to be bigger than the internet. Platform shifts like this come about once a decade, maybe. Two decades ago, it was internet and search. Google won that. A decade ago, it was mobile. Apple and Google won that. So if you are Microsoft sitting today in 2024, your score on those two big shifts is 0 to 2. You really don't want to miss this opportunity to stack another S-curve of growth on your top line. You know, Microsoft had their earnings yesterday. They shared a couple of great stats. 1.3 million GitHub Copilot paying customers. Their Azure AI customers tripled quarter on quarter. AI is having a massive impact on their developer and infrastructure business, and I'm sure they want to send this AI to the rest of their applications. You know, the 400 million Office subscribers, for example, right? So that's the first part, the willingness. The second part is they have a very unique ability to do that because, look, everybody has an intuitive understanding of how software ate the world, so to say. Software businesses can start and scale pretty cheaply. You need a couple of engineers, some basic infrastructure. Once you hit product-market fit, you can deliver software rapidly. Supply chain is straightforward. The marginal cost of distributing another copy of your software application is basically zero. So if a software business takes off, it can take off quickly and it's all profits. Generative AI, on the other hand, does not work like that. To start, you require expensive GPUs or a large computing budget to build and train an AI model that'll do anything at all. Then scaling it is also quite expensive. You need a ton more computing power to serve each new customer in the form of inference. Net-net, building an AI business is more like building a physical good than it is like building a digital good. And so, Meng has a special role here because they are the biggest providers of this compute. So, that is, call it, the 20,000-foot view of why MANG VC is on the rise. Now, to put this in perspective, the numbers are large.

It's amazing how a set of factors led to this absurd rise in the MANG VC phenomenon. And also, let's talk numbers. You're about to talk about the capital that they deployed, the sheer amount of capital that they've deployed.

Yeah, look, the numbers are large. The total capital raised in their 2023 investments was $25 billion. That's about 8% of all North American venture capital, up from just 1% the year prior. You know, as I, as I made this post on Twitter, one of the— somebody on Twitter asked me, hey, if you filtered this down to AI, that number is going to be a lot larger. I did the math. They were right. About a third of all AI investment dollars came from MANG last year. And so a lot of their investing is concentrated in companies like OpenAI, you know, $10 billion from Microsoft, Anthropic, $6 billion from Amazon, Google, CoreVive, NVIDIA. You know, just to put these numbers in perspective for our audience, Tiger, one of the most prolific venture capitalists during 2021, when the interest rates were basically zero, invested about $10 billion a year. Meiheng VC is more than twice of that in AI right now.

That is staggering. It's funny, like, if you go back a few years, there's a bit of a— when you think of like traditional VC versus corporate VC, there's a bit of like, hey, corporate VC is not— there's a bit of a stigma associated with that. A combination of factors, but now just the sheer volume and velocity with which MANG VCs are deploying capital, it's astounding. And one of the big points you made in your post was how the launch of ChatGPT, it separated the AI world into like two time horizons, like before ChatGPT and after ChatGPT. So how did that launch influence the investment strategies of, of the Mang we see?

Yeah, look, the alarm— well, alarm bells went off before ChatGPT. Mang was investing in really everything from mobility startups like Waymo and Rivian and Cruise to consumer businesses like Uber to deep tech like Magic Leap and SpaceX, software, everything. But after ChatGPT, the pace, the frequency, and the size of investments has concentrated in large language models. Actually, if you look at the image that I had in my original write-up, that graphic is staggering. More than 10x the dollars gone in far less time. Now let's ask the question behind the question. Why is this happening, right? What did ChatGPT do here? You know, I think it became clear that AI is going to be a mass phenomenon. It didn't— it did not take long for ChatGPT to gain popularity. It went from 0 to 1 million users in 5 days. It went to 100 million MAUs in 2 months. It's the fastest growing consumer application anybody's ever seen. And the first alliance that was formed was the OpenAI-Microsoft alliance. As that formed, a bunch of other alliances started forming to keep pace. And so just as the, you know, the BCAD is the divider in the Gregorian calendar, ChatGPT is that dividing line in the AI calendar.

Amazing. And this brings us to the really the heart of the issue here is the structure of these investments that raise big questions. And it's a problem because of this revenue round-tripping effect. So maybe to start off, can you explain what that round-trip effect is?

Look, as I describe this, let me just say that I don't have perfect visibility, obviously, into these large companies. So I'm using my intuition of what's happening here. But if anybody out there knows more than I do, and as I make these statements, you're like, hey, Apoorv, you got it all wrong, let me know. We'll correct it. We'll talk about it. So Yeah, let me explain the round trip phenomenon, right? So maybe we take the case of Microsoft's $10 billion investment in OpenAI. Of that $10 billion, only a fraction of that was delivered in the form of cash wired to OpenAI. The vast majority was in the form of compute credits that OpenAI can spend on Azure. Effectively, Microsoft is using their balance sheet to grow their income sheet. No actual cash is flowing in the credit transaction. And this is the round trip phenomenon. On the other hand, just to give you the counterfactual, if that $10 billion was raised by OpenAI through, let's say, 5 leading VCs on Sand Hill, OpenAI went out and invested that on Azure in an arm's length transaction. That would be totally fine. Go train GPT-4. And so this is, this is the issue. This is also actually not the first time this is happening. This has happened before with companies like AOL and Yahoo investing in startups with advertising credits, Illumina investing in their spinoffs utilizing sequencing technology, telco cable companies striking network capacity deals with each other. And so, you know, the other thing I'll mention is for those who are interested, Brad Gerstner and Bill Gurley have a very good detailed commentary on this in their latest BG2 podcast.

That's actually a really good point. And just to bring that together, a large company like Microsoft They invest in startups and then the startups pay, use that money to pay Microsoft as a customer. And because Microsoft is a company whose stock price is related to revenue, so more revenue would mean the stock price goes up. So what you're saying is money just kind of circles and comes back and yet the stock price keeps going up. And again, Bill Gurley, as you said, has famously written about it. Brad Gerstner has talked about it, about how you could just juice the numbers. So again, this is again only your opinion. Obviously with additional data we can, we can come back to this, but how do you think this is— why is this allowed? Like, obviously it's legal, they're doing it, but how is this allowed? And also, if there's a framework to stop this, like what can be done about it?

Look, I think it's hard for me to comment on the legality and the intent behind it, but I'm sure if inside of Meng, the bigger reason why they're driving this is the belief and the proof points, as we discussed earlier, that AI has the potential to be bigger than the internet. I'm sure those are the top, like, 9 reasons out of the 10 reasons they think about these investments. And second, you know, I will say that, you know, you, me, our listeners, We are better off as consumers, as investors, as entrepreneurs with MANG participating in the AI build. To your specific question, uh, look, FTC is looking into the issue as of last week with the investments from Amazon, Google, Microsoft into Anthropic and OpenAI. I'm just as curious as you are. We'll find out. Jury's out. And so expect more to come here. But perhaps the more relevant thing for us, our listeners, and our entrepreneurs is understanding the market distortion. That this is creating, right? And I'll say the, you know, financial investors make returns by finding great businesses. Meng VC makes returns by finding great customers. The equity returns for Meng are a bonus. And that leads to the higher AI valuations. And so from my vantage point, there's two truths that are true at the same time. One is that AI will be super valuable. In the long run, a gold rush is coming. And second, it's also very expensive right now, right? Valuations have gone up as a result. And it reminds me of something, you know, it's in the short term, we tend to overestimate the impact of technology. And in the long run, we tend to underestimate the impact of technology.

You make a really good point about how for a mining company, the equity is almost like a bonus. They're like, yeah, sure, we'll pay more to buy that stock in that startup because they don't really care about that portion. What they do care is, is the startup, do they have enough money in their bank account to spend on compute? And hey, look, we can sell compute. So again, not to say that they kind of sat together and created this evil master plan. It's not that, but it's mostly like, look, it's happening. Why don't we give it a little boost? Like the startups, they want to buy compute anyway. So why don't we have this strategy to invest in startups? Let's be a little bit price insensitive and like, let's go. So I think that's a fair argument. Now, you also mentioned in your post about how mine companies, they have evolved into modern-day digital tollbooths, meaning earning fees from a vast number of customers. Let's talk numbers here. For every dollar of inferencing that the customers pay their LLM provider, how much goes to these hyperscalers? And also, how are they using all of this cash within their businesses?

Absolutely. So look, I think it's hard for me to know what the exact dollar split is. That will depend on the commercial relationship that they have. And, you know, I'll borrow something that Warren Buffett uses, the analogy of tollbooths, to refer to a class of businesses that have such a critical position that allows them to have a very high pricing power, very high margins with very few substitutes. And Meng today fits that description. Let's call it a digital tollbooth. And so the way, you know, it works is anytime a developer creates an AI application, the first place the cash register rings is Manc. What does a developer need? They need data. That's on the hyperscalers. They need compute. That's on the hyperscalers. They need AI models. That's also on the hyperscalers. So even before this application earns a single dollar of revenue, if at all, the cash register rings. First at the MANG. It's a toll booth. And, you know, my estimate, this will show up in the gross margins of these businesses. And there's been a lot of debate on gross margins. And I think gross margins is a good place to focus than the top line for these businesses. But on the hyperscaler side, this has led them to earn supernormal profits, right? In all of 2023, MANG combined, earned about $275 billion of operating income combined in 2021. To put that in perspective, that is more than most countries' profit pools.

Yeah.

Their market cap is $8 trillion combined. That would make it the third largest economy after the US and China.

Right. Right. That is, you know, if you think about that, it's astounding how much market power and pricing power is. It's insane how much dominance they have here. And also, you talked about margins, and I want to get your thoughts on a certain dynamic. So historically, Ultimeter famously invested in amazing SaaS companies, data companies, and now obviously AI companies. So if you look at the margin profile of a pure SaaS business versus a data business like Snowflake versus a modern like AI business, obviously they're different. So can you talk about the differences in the margin profiles and also how is the corresponding market value being created? Meaning, should the multiples be higher? Should they be valued differently? Is the future so much brighter that AI company with the same amount of revenue but much lower margin, should they be valued higher? How do you think about the dynamic between these three pillars?

Sure, sure, sure. Let me actually, you know, this is, this is a fascinating discussion. Let me present some facts. So if you look at the AI stack, you know, it's in broadly 4 parts. You've got the compute semis at the bottom, so the GPUs, the CPUs. Above that, you've got the AI model, you've got the cloud layer, and then you've got the applications on top, right? So these are the 4 stacks. This is some quick back-of-the-envelope math, but if you look at the gross profits that were generated by each of these layers, right? So at the compute layer, semis layer, the amount of gross profits that have been generated are in the order of like $50 billion or so, right? NVIDIA is the most dominant player in the stack. At the next layer, which is the model layer, OpenAI is rumored to be at $1.6 billion of annualized run rate. Now we don't know what the gross margins are on that, but suffice it to say it'll be lower than that, right? At the cloud layer, you know, we got some information yesterday. Azure is doing about $74 billion of analyzed run rate, of which about $3 to $4 billion of run rate are being driven by AI. Let's say that generates a single digit billion dollars of gross profit. So the cloud layer is generating about, let's say all of them combined, maybe $4, $5, $6 billion of gross profits. And on the application layer on top, look, it's anybody's guess is as good as mine. That's another, that's a small number, hopefully a positive number. And so recapping, it's staggering. The highest amount of gross margins that are being generated today have happened at the semis layer, followed by the cloud layer, and then the model infrastructure and the applications. Now, this is sort of the way technology progresses, right? You first go through the build, as was the case in cable, as was the case in internet. And so this sort of makes sense. And I suspect we will see a lot more gross profits being generated in the other parts of the stack going forward, or that's the promise.

And if you think about valuing, a modern AI business. It's still, obviously it hasn't been around for like 5 decades. There aren't like 1,700 textbooks written on it. So if you think about valuing an AI business, net-net, when they go public, no matter who you are, your margins, your earnings are going to dictate your valuation. So is that the right way to look at it? Or would you make the argument that the future earnings, because it's so bright for the next 100 years, they'll print cash, versus other sectors where they won't do that. That's why we should value AI businesses higher, even though their margins are not so great, like not as compared to SaaS. So how do you think about valuing a modern AI business with lower margins?

This is the $100 billion question of 2024. And so look, I will, I'll give you a couple of frameworks, right? The first thing is, as we have seen in literally every technology cycle, we will overvalue the impact of this in the short term. We will undervalue this in the long term, right? Back in, let's say, let's say you and I were investors in the automobile revolution in the 20th century, we were in 1900s. There was about 2,000 automotive companies. If you fast forward to 50 years later, there were 3 companies that had more than 80% market share. If you and I were automotive investors, we'd have gotten the trend right, but the probability with which we would have picked companies would be 3 divided by 2,000. That's a tough job. Same with the internet, right? Search— there was AltaVista and Lycos and all these search engines and internet companies that were booming. But it was only until 2004 when Google went public, I think it became clear that we have a winner. And the, and, and, and the Dynamics in that market were winner-take-all, and that accrued value over time, right? The same, I suspect, is what will happen here. We'll probably have 2 or 3 dominant model providers, and the dust will settle. Now, who those are going to be, hard to tell. We might— we have some candidates. And to your second question, How to value them? Look, I mean, ultimately the value of a business is the sum of their future free cash flows. That's just the law of gravity. And now there is some debate on what those cash flows will be, at what rate they'll be growing, what is the profitability of these businesses. And I think I'm a curious observer just as you are. And so I think as we look to valuing this business today, I would keep those historical patterns in mind, or at least we are keeping them in mind.

Right, amazing. And I think that's a, that's a fair answer because a lot of these frameworks, um, and market dynamics, they're just being formed. So the goal is to keep, uh, keep an open eye and keep observing how the dynamics will shape out. Now, moving the conversation a little bit to the effect. You mentioned market distortions, right? So, when a, when a MANG VC comes in without any regard to the price or anything like that, it distorts the market. Maybe bad businesses get funded, businesses that burn cash get funded. So, if you are advising maybe a portfolio company or maybe a founder friend, they're like, hey, they're growing, the business is doing good, they're making good ARR, is at a good level, Now a big MANG VC shows up for the next round and they're like, hey, I want to invest, I don't know, a quarter billion into your next round. So what are the potential benefits and drawbacks for that particular founder? What are all the things that person should consider before raising capital from MANG VC?

Yeah, this is a great question and very topical for a lot of the founders that we are speaking with. So let me walk through sort of the differences of that fundraise from the lens of an entrepreneur raising capital. Right, first thing, let me frame the numbers. So just a quick stat, by dollars, the dollars into AI are large, right? $25 billion, 8% of all venture capital, but actually by number of companies that they're investing in, that's less than 1%. So for a vast majority of entrepreneurs, building your business with a typical Sand Hill venture capital firm continues to be a great option, a great capital partner to build an enduring business. For the special enclave where MANG is a strong option, I think it's a, it's a great option, right? And that actually might be your only option given the scale of capital required. And for those entrepreneurs, I have 3 things that I would recommend keeping in mind. The first is valuation. Look, it's okay to raise a large round, even at a large valuation from bank, but you've got to start thinking long term. No longer can you plan your fundraisers in 18 to 24 month cycles. You've got to plan 5 years in advance. You've got to think about the next round and the next round. And if you have aspirations to go public, you've got to think about whether you want a large strategic on your cap table. In the public markets? Who will be, who will be there for your next round, right? So that's one, the valuation risk. The second is think about the revenue concentration, revenue risk, right? If in addition to compute partners for you, is that creating too much concentration coming from just one partner, from one hyperscaler? Does that make it hard for some of your customers who work with perhaps other hyperscalers to use your service. And the third thing I would help folks consider is exit risk, right? Financial investors make money when the startup goes public or sells at a higher valuation, and that's our primary north star. In the case of Meng, they're not purely financial investors. Their strategic directives are to be close to their customers, and there might be some misalignment when there's an opportunity for liquidity. So those are the 3 things I'd keep in mind for entrepreneurs who are thinking about that.

Right. And you made a very interesting point. Obviously it's known, but that the concentration of capital from, from MANG is in a very specific group of companies, basically the companies that spend a lot of money on compute, and then it makes sense to invest in those companies. And again, maybe another dimension of the market distortion that you talked about. Now, what does that do for downstream growth stage VCs? Let's say, Mang VC, sometimes they step in early and say, hey, I'll lead your $50 or $100 million round. And then there are growth stage VCs who are like, we'll write a check pre-IPO and you go IPO the year, like 1 or 2 years later. So, Can you also talk about how that just works? Because market goes up and down, maybe you raise it as a nice big number from a bank VC, and next year, like, a real sharp growth stage VC comes in and says, hey, look, I looked at your business and it's worth less than that. If equal would be great, but less is, you know, it's a terrible, terrible time to raise. So how do you think about the downstream effects? Of that?

Absolutely. Look, this is a very topical question. Not too long ago in 2021, we had a distortion like this. A lot of great entrepreneurs raised large pools of capital at large valuations. And look, there's no free lunch. We are seeing the downstream impact of that this year. There are 3 outcomes. One is everything goes well, top into the right, and you work back into that valuation with the new reality, with the new interest rates. You're fine, you're golden. For the vast majority, they're in the second bucket where the valuation that was held, that was marked in the 2021 or 2022 round is higher than the valuation today. And so you've either got to work into that valuation before you raise a round at the same price or more, or you've got to accept a down round, maybe some structure. A lot of great businesses have done that. Like Stripe. The, the perhaps the third, you know, the third path, which is perhaps the, the least desired outcome, is you're, you're not able to get there in the time that you need to raise capital, and the, you know, record number of startups shutting down or getting acquired. And so look, I don't know what will happen in this distortion, right? This is being written as we speak. There's a couple of rounds that are being done by investors after a manground was priced in some of the largest LLM companies. And so the jury's out, uh, we'll see how it goes, but we have a data point from not too long ago, 2 years ago, right?

And from the founder's point of view, right, let's say they haven't done this before and they're, they're about to compare, they're about to raise a round, and on one hand you have, hey, mang VC They're giving me a higher price and they're not giving me that much of a hard time versus this like sharp VC. They've been around for, for quite some time, but they're going to price it appropriately and they'll ask you a whole bunch more questions. So how, again, it feels like a founder has to act against their own economic interests. In the long term, we know how this plays out, but in the moment it feels like, let me take the easier option, higher valuation, lower diligence. And just run with it. So maybe one last point to mention here, how would you advise them to act against what seems to be like in the moment, like the better option?

Yeah, look, this is, I hate to say this, but it depends. It depends. If compute is the largest expense for you and you are in that special enclave of companies where MANG has a very strategic value to add. I think it makes sense for you to take that round, and as many have, while considering the trade-offs that we just discussed. I think the other— the thing that I will say is, look, as a founder, you know, I know a bunch of founders where the price at which you raise your last round ends up becoming the difference between a great outcome or selling below your preferred stack, meaning that if you raised a large pool of capital and the sale price is less than the total pool of capital raised, the founders don't make any out money. However, if you raise a disciplined round at a reasonable valuation, the exit actually creates a lot of wealth for founders, for everybody on the table. And so I think sequencing matters. I think the idea of maximizing price and minimizing dilution in this current round, while might seem founder-friendly in the short term, it's not in the long term. There's a very good Silicon Valley HBO clip on this that I recommend folks watching. It's— they use humor to deliver a very powerful message.

Yeah, I remember the scene where they meet in that bar to talk about it. It's a hilarious scene. All right, maybe one last question before we go to the rapid-fire round. Now, how do you foresee the role of MANG, MANG-VC, in the next 2 years? Meaning, will this trend continue? And also, how long can this possibly continue? Like, the best— your best educated guess on how long this can last.

Yeah, look, I think AI is going to be a very big driver of GDP over the next decade or two decades. It might even be bigger than the internet. If we believe that and we understand the operating model of these AI businesses, you will need compute resources both to train and get started, but also to serve during inference. And so based on everything I understand, I think their position gets stronger over time, right? We got some very good data points from Microsoft. Yesterday. Azure has reaccelerated at the scale of $74 billion of ARR, now growing at 28— 6 percentage points of that 28% of growth was driven by AI. So about $3 to $4 billion of ARR coming from that. And so we, we, we have data points that the position of Meng is getting stronger. I expect them to play an important role. Another stat they shared is that they've got 1.3 million GitHub Copilot developers paying them, you know, at $10 a month. That's about $150 million of ARR in under 2 years, right? And so not only at the infrastructure layer, but also from a developer tools perspective. And I think on the application perspective, the jury's out. Everybody's keenly waiting to see how the Copilot experiments are going, but they're growing. And then finally, you know, on the Azure AI infrastructure, right, they disclosed that they had 53,000 enterprises using Azure AI. That has tripled quarter over quarter. Look, if you begged a developer 3 years ago to use Azure for their AI workloads, you'd have a hard time. Today, the tides are shifting. And so I would say, look, in summary, the position is incredibly solid. I expect them to continue to be dominant in driving what feels to be one of the biggest platform shifts of the next decade.

That's amazing. And I did see your tweet about the graph about how you and a bunch of people are talking about how the Copilot revenue, the Copilot platform from Microsoft, yeah, it's growing, it's growing real fast. It's amazing how Microsoft can take— again, not to— it's not the same analogy, but they say, okay, Zoom and Slack, that seems to be working. So let's just run with it. We'll build Teams and then we'll just make the graph go like this. It's Microsoft scale and distribution is absurdly good. And it's crazy.

Look, if Copilot was a startup, that startup is cranking and they've got the 3 ingredients of AI. It's got the data, it's got the compute and the distribution at incredible scales.

It's just amazing. All right, with that, we're at the rapid-fire round. I'll ask a series of questions and would love to hear your answers in 15 seconds or less. You ready?

Let's go.

All right, question number 1: What's your favorite book?

Man's Search for Meaning by Viktor Frankl.

Amazing. All right, next question: What has been an important but overlooked AI trend in the last 12 months?

The gross profit dollars being generated in AI are so high at the semis layer, about $50 billion. Cloud's about $5 billion, an order of magnitude lower. And the apps are even an order of magnitude lower than that.

And it's, it's funny because it's counterintuitive. Usually hardware is supposed to be super low margin, whereas like the software application is supposed to be very high margin. But here it's completely Flipped. It's very interesting. All right, next question. What's the one thing about venture capital in AI that most people don't get?

In the short term, we are going to overvalue the impact of AI. In the long run, we will underestimate the impact of AI, and that shows up in the valuations.

Right. All right, next question. What separates great AI products from the good ones?

People don't wake up to use AI products. People wake up to live their lives, and the greatest AI products will become a part of people's lives.

Amazing. Uh, next question: What have you changed your mind on recently?

I used to think tweeting and social media is a distraction and gets in the way of investment research, but turns out putting your thoughts into the wild is quite valuable. Get out of your bubble, pressure test the ideas, and invite diverse opinion. Yeah, I learned a lot putting out the MANG piece.

Yeah, and I strongly agree with that. The more I do it, the more I see the effect. And I think you said it right, like it's about pressure testing your ideas, because inside your head, at least in my head, I'm like, oh my God, that's the most perfect idea ever, let me write it down. So the process of writing itself, I'm like, oh, that sucks, I can't— I can barely express it. And once you write and then you publish it, and then people are like, oh, It's very interesting how in your head it seems like the most amazing thing ever, but once it's out there, it's good. But also through this journey, you'll find the— you'll find gold. Like, some ideas are like really good, and you'll get a lot of— it's a power law. One of these will work, and you'll gain a lot out of those ideas.

All right, 100%. You know, in this case, like, you know, I wrote about Meng, but a bunch of investors and friends and mentors reminded me that this has happened before, right? I wasn't around in the telco era, but this has happened in telcos. This happened in the advertising space with AOL and Yahoo. Paul Graham's got a great blog on that. It happened with Illumina. And so, you know, as a fact for, you know, young investors out there, that there's a lot of pattern matching to be learned.

Right. 100%. All right. Next question. What's your biggest or wildest AI prediction for the next 12 months?

Ooh, spoiled for choice, but I'll go with, Don't forget Apple. Look, they have an incredibly good vertically integrated setup. Siri was not the product we wanted or we dreamt of, but they know that. And I think they're going to come out with an incredible product in the next 12 months.

All right, final question. What's your number one advice to founders starting out today?

I don't know if this is the number one advice. This is the number one most overlooked advice, right? I advise founders to listen to the earnings calls of your public peers. It'll take you 1 hour a quarter. It'll be one of the highest ROI investments you'll make. You'll learn about the landscape, go-to-market strategies, the future roadmap. You know, there's not a single earnings call where I haven't learned something new. In fact, I'd suggest founders to listen to these calls together with their leaders. Maybe start with Meng.

That's amazing. That's a really good piece of advice because I do feel like there was a big chunk of time where earnings call, when you're learning and you're looking at the world, earnings call is not exactly the place you would go, at least as an engineer. But I think it's a ton of information and market numbers, ideas, information, future. You can get a lot out of that. It's very interesting. So, Apoorv, this has been a brilliant episode. Thank you so much for coming onto the show. And again, as always, I think it's true of all Ultimeter folks, but like you show up with amazing data. The depth of research is incredible. And every time I interact with someone from Ultimeter, my belief is reinforced that they are all absolutely top-notch people. And I know it's a lot of work to gather this data, synthesize it, and present it in a nice, interesting, engaging way. So thank you so much for all you do. And again, thank you for coming onto the show.

Thanks. Great to be here. Look forward to doing more.