VC10X - The AI Bottleneck Keeps Moving - Ashmeet Sidana, Founder & Managing Partner, Engineering Capital

Speaker A: So at any given time, you are combining components, storage, memory, compute, software, et cetera, which have different characteristics to make a system. And the problem is the real issue here is that those underlying components grow, evolve, change at different speeds. The answer is we are going to have to rebalance again every few years. That's just what has to be done.

Speaker B: Some people have started the debate that, okay, maybe investing in software right now is not as valuable as investing in hardware. Do you buy into that? How do you view that?

Speaker A: Here's the startling number. 63,000 times you can get a performance improvement if you move from Python directly to writing on the intel x86 chip. This is without touching the hardware. This is without changing anything in the chips, in the systems, in the motherboards, in the interconnects, et cetera.

Speaker B: Data isn't just the new oil, it's also the new asbestos due to its potential toxicity as enterprise model training becomes heavily polluted by synthetic or poisoned data. How must engineering architectures change to filter out these liabilities?

Speaker A: What we've got is now this very mushy, messy collection of data which is being further poisoned by. By data being generated by those AIs themselves.

Speaker B: Hey everyone, this is Prashant and I'll be host of the VC10X podcast. And today we have Ashmeet Sidhana with us. Ashmeet is the founder and Managing partner of Engineering Capital, a, uh, seed stage venture fund he runs as a solo GP in the Bay Area. Ashmeet is a Stanford trained engineer and former product leader at VMware. In this conversation, we get into why software still offers a very lucrative investment opportunity backed by a number that will stop you in your tracks. Why data is no longer just the oil, it's the new asbestos dose. How data sovereignty is reshaping the global cloud landscape and what it actually takes to back a technical founder at day zero, before there's a product, before there's revenue, before any proof. Ashmit has been oversubscribed on every fund he has ever raised. And he'll tell you exactly why. Running alone is a structural advantage, not a risk. So without wasting any time, let's dive straight in. This episode is brought to you by Podcast NX, my podcasting agency for VCs. We help VCs start and run branded podcasts that can serve as their entire content engine. Turn the conversations you're already having into a machine. One recording becomes your presence on LinkedIn, Spotify, YouTube and email. Don't just build another VC firm. Build a firm that can stand out to LPs and one that founders can brag about having you on their cap table. Check out our services@podcastnx.com I'll put the link in the description below. Now enjoy the episode. Hey, Akshmi, it's so good to have you on the VC10X podcast. How are you doing?

Speaker A: Thank you, Prashant. Very happy to be here.

Speaker B: It's a pleasure having you on. Uh, let's get to the first question that we have here. So, as clusters grow to hundreds of thousands of chips, raw processors have jumped ahead, leaving networking to catch up. What specific breakthrough needs to happen in Interconnect Layer to stop idle hardware from bottlenecking?

Speaker A: LLM training I think to understand this, it's important to understand the basic constraint with which all of these decisions get made. So at any given time, you are combining components, storage, memory, compute software, etc. Which have different characteristics to make a system. And the problem is the real issue here is that those underlying components grow, evolve, change at different speeds. So when people design a system, they say, okay, I'm going to build a system with so much memory, with so much compute, with so much networking, et cetera. They put it together and it's a balanced system. But over time that those decisions that were made, they start drifting apart from each other. You, you had so much compute, so much networking, so much storage, they were all balanced. So imagine if you were designing a car and three years later the engine had become 10 times bigger and everything else was the same size. That would be a weird car. It wouldn't work. You know, it would just not, it would just not be balanced. That is constantly happening in computers because the underlying components grow and evolve. What people call Moore's Law, you know, Dennard scaling, et cetera, is happening at different speeds. And so because those curves are so fast and because they are at different rates, this balance is always getting out of whack. And so, uh, there's no perfect answer to your question. The answer is we're going to have to rebalance again every few years. That's just what has to be done.

Speaker B: Absolutely. Quite interesting. And the ceiling of AI scaling is shifting from silicon availability to megawatt capacity. Right, Right. When we look at the physical limitations of grids and cooling systems, where does the data center stack break first as we attempt the next 10x scale?

Speaker A: Yeah, once again, that's a temporary constraint. You know, power generation is an extremely well understood problem. Obviously it's, uh, over 100 years old and frankly, the total size of the utility grid, at least in the U.S. i'm not aware, uh, I don't study other countries. Um, has been pretty much flat for the last few decades. Now, for the first time with AI, we were seeing an increase in the total demand for, uh, electric power. And that is an easy problem to fix. That's simply a question of capital expenditures and a little bit of time. Because these are large physical projects, they take some time, but, uh, in a few years they'll catch up and that will no longer be the constraint. And the constraint will again go back to the real issue, which has always been this interbalance between the different components and those components being out of whack with each other. So that's the real constraint.

Speaker B: Absolutely. And even on the power front and how these data centers are being powered, uh, do you see that limitation being hit sooner than later? Uh, and also the kind of power that is coming in, because I've heard that they need consistent power to fuel these data centers. And some of these sources that are currently active might not be as consistent or trustworthy. Right. So people are saying we'll need to tune into nuclear and stuff like that. So we'd love your view on that.

Speaker A: Yeah. So, uh, traditionally the power sources that were the cheapest also happened to be the ones that were the most consistent. That's just luck. That's not by design. So which were the power sources that were the cheapest? Hydropower, coal power and nuclear power were traditionally the three cheapest sources of power. And those are very consistent because they literally generate, uh, electricity on demand as much as you want on a continuous basis. Um, Moore Dam was built, uh, thanks to the Great depression in the 30s. It is still generating over a gigawatt of power continuously. So, um, what has happened more recently, unrelated to the AI wave, is that intermittent sources of power, primarily solar and to some extent wind, have become cheaper than these other more consistent sources. So now you have a problem. Now, if you want the cheapest power, you don't get the most consistent power. And so you have to make this trade off between these two sources of power. For data centers, you know, which are especially AI oriented data centers, demand is continuous, it's flat, it's continuous. You always want them running. And so you want a consistent source of power. But, but that's no longer the cheapest. And so people are making different trade offs around it. Again, these are very well understood curves. Um, engineers can balance between them given the set of constraints that you want. And I frankly don't see a ton of innovation over there. Um, because the Physics is very well understood. The mechanical engineering is very well understood. The electrical engineering is very well understood. And we're already at 90, 80, 95% efficiencies in most cases. So there's not that much juice left to extract, you know, from there. The innovation, where the real juice exists is on the AI side, the software side, the chip side. That's where there are massive amounts of inefficiencies.

Speaker B: Absolutely. And one debate with the recent how the markets are behaving right now and the sudden drop in Google, the Alphabet stock that we have seen, some people have started the debate that, okay, maybe investing in software right now is not as valuable as investing in hardware. Do you buy into that? How do you view that?

Speaker A: Uh, I don't buy into that at all. And if, uh, you want proof of that, I would recommend that you go online and search for the Turing Award presentation by Professors Hennessy and Patterson. So Hennessy and Patterson are icons in the computer architecture industry. Professor Hennessy was my professor at Stanford. He was the president of Stanford. Um, they got the Turing Award, which is the most prestigious award in computer science, um, about seven or eight years ago for computer architecture. And they have a wonderful slide in their presentation where they show the amount of inefficiency that exists when you run a Python program. Python is the second most popular programming language in the world. So, for example, Facebook is written in Python, one of the very, very large, uh, applications. Um, when you take a Python program and you do matrix multiplication, which is the most common compute, uh, that is used especially for AI. A lot of AI applications are written in Python. And you run that on a standard intel x86 chip, but you do it manually, by hand. You bring in as much efficiency as possible. There are several PhDs behind this. Here's the startling number 63,000 times. You can get a performance improvement if you move from Python directly to writing on the intel x86 chip. This is without touching the hardware. This is without changing anything in the chips, in the systems, in the motherboards, in the interconnects, et cetera. And so why do we accept such a m massive amount of inefficiency? That's a very interesting question. And it also, of course, answers the question of where there is so much opportunity to be done. So for the foreseeable future, there's massive amount of opportunity to be had in software. Uh, it will continue to increase because the innovations are still there. Yes, there are some innovations on the chip side. We are seeing new architectures develop. You know, the tpu By Google was a very interesting new architecture. Cerebras had a wafer scale architecture. Um, Nvidia has tried to come up with the NPU, you know, intel has, uh, NPUs, et cetera. Those are interesting architectures and they do make some incremental improvements. Um, but they are dwarfed by orders of magnitude, many, many orders of magnitude by what we leave on the table for good economic reasons on the software side of the house. And that is where I believe that we will see tremendous innovation, tremendous investment, tremendous value creation that will continue to occur.

Speaker B: Absolutely. Quite interesting. And you anticipate the rise of 5 to 7 independent cube audit cloud regions driven by data sovereignty. For a global enterprise, how can the software layer evolve to abstract away the architectural friction of structurally isolated hardware?

Speaker A: Yes, so that's a very interesting technical challenge. Uh, again, a place where I see technical approaches that will create opportunities for new startups, for new innovation, venture capital, et cetera, to occur. Let's first understand what is data sovereignty. Data sovereignty refers to the concept where a country would be able to control all of their own data and the entire tech stack that it runs on. It's not dissimilar to, for example, in the defense industry, when people build munitions, you know, they build a fighter jet or they build, uh, a smart bomb or whatever, they want all the components to be built locally because they want to control it in the case of a war. In the case of data sovereignty, you want to control the data and the underlying technology. Um, complete data sovereignty is, I believe, impossible to achieve. It's really very, very hard to achieve that. And you are seeing that tension right now between China and the U.S. where the U.S. has been putting restrictions on China, yet China is able to build things like Deep Seq, even though the US is actively working against, to prevent China from doing that. So absolute data sovereignty is impossible. But, uh, you will see degrees of data sovereignty, degrees of independence, degrees over there. And, and we are clearly, because of geopolitical reasons, moving in a direction where the desire for sovereignty is higher. And therefore we will see these sort of islands get developed. You know, we will see islands of sovereignty and people will make an attempt towards doing it. Uh, this is the march of human civilization. You know, we were in an era of globalization and an era of integration for several decades. That era has come to an end. And so data sovereignty is now an interesting problem which becomes yet another economic impact on how companies, uh, design products and how, uh, products get shipped.

Speaker B: Absolutely. Quite interesting. And you have championed the thesis that data isn't Just the new oil. It's also the new asbestos dose due to its potential toxicity as enterprise model training becomes heavily polluted by synthetic or poisoned data. How must engineering architectures change to filter out these liabilities?

Speaker A: Yeah, so, uh, you know, data is the new oil was the common thing. Once machine learning, um, came online, I started talking about how data is the new asbestos when we saw things like, uh, toxic applications like social media, et cetera, develop where people were abusing the data that they were collecting, and they were able to do that. And so data started becoming toxic when there was a regulatory backlash on that type of data. Um, so, for example, Facebook had Cambridge Analytica, and then they saw all of these privacy regulations come up against that. Today we are in an era where scale really matters. So it so happens, didn't have to be this way, but it so happens that the artificial intelligence architectures that are working today, namely transformers, work better when you give them more data. And so people are trying to collect more and more and more data. And so you talk about data scaling and these giant, you know, collections of data that people have produced to be able to feed these forms of AI. Great. But when you do that, you lower the quality of the data that's coming in, and that's just inevitable. And so what we've got is now this very mushy, messy collection of data which is being further poisoned by data being generated by those AIs themselves. So now the loop is getting completed where you are getting AI slop that is getting fed in over here. That said, there will always be what I call golden data. Data that is precious, data that is incredibly valuable, which traditionally has tended to be very, very small. So let's start with the academics. Okay, I work in computer science. One of the most important papers in computer science was written by Alan Turing, hence the Turing award, uh, between 1933 and 1936. In, you know, every student in computer science reads that paper. Even today, that paper is incredibly important because it lays out a foundational theorem of computer science. Okay? It defines the halting problem, it defines during machine, et cetera. And so that's golden data that will always be incredibly important. Doesn't matter what form of intelligence is being applied to it. You can go online, you can read all of GitHub, you can read all the data, you can read all the source code of everything outside which is available open source, millions and billions of lines of code. But Turing will always be golden data. And so I believe modern architectures will be able to separate these types of Information today. It's a very brute force approach. We just throw all this data into, um, machine learning and training. And there's really no distinction made between, uh, Alan Turing's paper, uh, and some, uh, online blog post that someone writes. Uh, but that distinction will start coming in and I believe intelligence will be able to separate those forms of data also. So we are headed towards that. I have portfolio companies that are working hard to make that, uh, happen. Not at the level of Alan Turing. I don't think they get a Nobel Prize, but, you know, very deep technical insights. I have a portfolio company, Concentric, which is able to separate, you know, pii personally identifiable information from non pii completely automatically. Um, so that's an example of, you know, where we are using AI to solve the problems being created by AI.

Speaker B: Absolutely. So you're saying that AI can also be used to detect AI slop versus what is actually high signal and first source.

Speaker A: Yeah, I don't believe AI is able to do that today. Uh, I think a lot of those claims are overblown. But there's no law of physics which says that we will not eventually be able to build AIs that are as intelligent or more intelligent than us and be able to make these separations. So I do believe that we are headed to, you know, and it's not too far away. I hope to live long enough to see things like that, um, where we will see intelligences of that level.

Speaker B: But right now, would you say that it is more on the humans to, you know, separate that all the AIs

Speaker A: that you are seeing today are intelligent only because even more intelligent human beings trained those AIs. Um, they have become so intelligent because they have a wonderful team at Anthropic and OpenAI and Google and Microsoft. We're all working very hard to do this. And there are hundreds and thousands of people, PhDs who are advancing the state of the art. But compare that to someone like Alan Turing or let's take Indian examples. Ramanujan. Right. I mean, he wasn't trained by someone who was more intelligent than him. Yes, I know he worked with Hardy and, you know, obviously they came up with some work together, but he was able to write a paper by himself. Where does that intelligence come from? We have not yet solved that problem in computer science. I hope one day we will solve it. But today it is not a solved problem.

Speaker B: Absolutely. And the modern AI stack is structurally optimized for the transformer. If a radically superior architecture emerges tomorrow, how much of the specialized infrastructure being built today becomes Instantly obsolete. And what remains? Future proof.

Speaker A: What a great question. The answer to that depends on what that new architecture is. And so nobody knows the answer to that question, because nobody knows what that new architecture is. And depending on what the characteristics of that new architecture are, uh, you know, the current stacks, parts of that may or may not become obsolete. So for example, I talk about something called the death of the X86. The X86 is a very intelligently thought, very smartly designed chip which for decades ruled as the source for where real computation got done that has been completely obliterated today. None of the interesting computations are done on an X86. All the interesting computations are done on different architectures, uh, primarily ARM, but there are other architectures like TPUs, et cetera, that are used. So, uh, depending on how intelligence evolves, what part of our current stack will become obsolete? Depends on the nature of the innovation that happens in these new architectures. So we just don't know the answer to that. And this is true in every industry.

Speaker B: Right.

Speaker A: When agriculture evolved, we used to work with animals. We built plows in a certain way. The moment we started working with tractors, we built plows in a very different way.

Speaker B: Right.

Speaker A: We still have to plow the land because we still want to grow wheat. And I still want to have my roti and paratha every day. Uh, but, uh, so I still want the wheat, but the way I plow the land has changed. And so what becomes obsolete depends upon what innovation occurs. And what is the nature of that innovation? You cannot predict that that's the nature of innovation.

Speaker B: Absolutely. And as custom, silicon and arm, um, dominate the next generation workloads, where do you see the most asymmetric startup opportunities arising from the legacy CPU package?

Speaker A: Yeah, So I think there is still tremendous opportunity in the legacy stack, um, because right now we have this new form of intelligence AI, but all productive work is still done in the old stack. And so connecting these two at some level is where there is massive opportunity. And that's why you are seeing this huge growth of B2B business to business startups that are being built, funded and growing, where people are trying to connect these two things. Um, that is a 5, 10, 20 year effort. It's not going to happen in six months or a year. Uh, economists call this phenomenon diffusion. In other words, you get a new innovation, how long does it take to diffuse through the entire economy? In the case of electricity it took 40 years. In the case of television, it took 30 years for the diffusion to occur. Um, I think AI, it'll take 10 years or 20 years. It will not be two years for sure. It cannot happen that fast. Um, there's too much friction, there's too much legacy, there's too much work involved in making that happen. But whoever can accelerate that can make a tremendous amount of money. Uh, and that's why there's tremendous opportunity over there. And that's where it's highly asymmetric because you are getting the benefits of AI with the value and the money of the legacy, where all the money is absolutely quite interesting.

Speaker B: And big tech, with their big capex budgets can afford to buy private power grids and buy out entire chip manufacturing lines. How can an early stage infrastructure startup maintain capital efficiency and survive when its primary competitors deploy infinite capex?

Speaker A: Yeah, this is such a good question. Because it is so hard right now. The decks are stacked in favor of big tech. So you know, the companies like Google and Facebook and Microsoft and Intel and etcetera, they are monopolies. Okay, let's face it. I mean legally, technically, they may not meet some legal test or maybe the government doesn't go after them, but they have massive sources of captive revenue and profits that they are generating which they are using to control the markets. And uh, perhaps a better word than monopoly for them is oligopoly because they choose not to compete with each other directly. So you don't see Google going head to head with Facebook going head to head with Microsoft. They actually partner with each other, um, in many cases so that, you know, um, they don't uh, destroy each other. Um, they are too strong for that. And so that really leaves portion for the startups and in the gaps which are left over there. And as uh, I tell all my founders and CEOs, you know, you have to work harder and smarter and faster if you're going to beat these big guys. And there's no substitute for that. You just absolutely have to do that today.

Speaker B: Absolutely. And success in today's terms with the big capex, if you even achieve a breakthrough innovation, they might offer you a uh, very lucrative exit at the outset because that will be absolutely peanuts for them, but a massive exit for you right there. So even if you achieve even some breakthroughs in whatever technology you operate in, then it can be massive for you as well.

Speaker A: Yeah, that is a structural problem in the way our markets work today because we've created these massive concentrations of capital and large wealth and these people are able to spend the capex and able to then buy the startups out. Um, startups have A tendency to sell earlier. You know, ideally the public markets would be able to allow them to raise money and have independent existence. Since the government has, um, in my opinion, in a very mistaken way, made it so inefficient and so unattractive to be a public company. The attraction is to sell in the private market and to sell over there. And the numbers are big enough. I mean, I can give us an example. Just from last week, um, one of my companies, Wide Field, was acquired by Cisco. Um, I was the first investor in the company, which is typical in the stage that I work at. Um, and uh, the founders, Abhay and Karthik, had some very interesting innovation in the agentix space. We hadn't even announced our complete product. Uh, and Cisco bought the company. Um, and so obviously it was attractive to me, it was attractive to Abhay and Karthik. You know, they've made generational wealth for themselves. Congratulations to them. Um, and I'm good, you know, congratulations to my LPs. They're going to get a nice return on it. Uh, but, uh, my condolences to the broader economy where an independent company is not going to exist. You know, it's going to now become part of Cisco, which will also compete, which will also grow. Uh, but it would be so much better if we could let a thousand flowers bloom. So in general, I am in favor of a looser public market system where we have more private, where we have more public, independent companies. Most people don't realize, but the number of public companies in America has been going down over, over the last 20 years. Every year we end up with, even though there's an IPO boom right now, the total number of companies is actually going down. So it's kind of a sad occurrence. Uh, hopefully the government will wake up and hopefully they will fix it. But the government, frankly, has bigger problems right now to address.

Speaker B: Absolutely, I completely agree with you on that. Uh, and incumbents are rarely beaten on raw resources. They fail because existing revenue models or rigid organizational structures make them slow to react. Where are, uh, big tech structural blind spots in the current AI cycle?

Speaker A: Yes, this is such a good observation. And my way of saying that is what I like to tell founders is what you can achieve with a small amount of money, you can never achieve with a large amount of money. Most people think the opposite. Oh, if I had more money, I could do more. I could make things happen. And the example I often give them is of Gandhi. Okay, now that's a political example. But here was a man who took on the World's largest navy, the world's largest army, the world's largest treasury that controlled two thirds of the world's gdp. And what did he have? How much money did he have? How big was his army? How big was his capex, how big was his PR budget? He didn't have any of those things. Uh, and yet he succeeded because he invented a new way of competing against their power, namely satyagra. Right. So he was a very intelligent politician. If you are a technologist, you have to similarly invent a new way of beating them. That innovation typically comes from a very small number of people. It's usually one, two, three people working together. Even Transformers, which have revolutionized, uh, modern AI. Uh, you know, was an architecture originally. Uh, obviously Google had a lot of work with it, but the real work was done by OpenAI when they were not heavily funded, when it was a very lightly funded, venture backed startup. It is only when they started scaling that the massive dollars started coming in. And that's an artifact of the innovation that happened. A different innovation may not require that. So that's just the nature of how innovation happens. Um, your job as an entrepreneur is to find one which you can execute with the amount of capital you have, which is typically much, much, much less than what a bigger company has. So that's just the nature of the problem. And as an entrepreneur, that's the problem you have to solve if you're going to be successful. Almost always all the stories in history, if you look at the great companies that have been built, raised very small amounts of capital, you can go back in history, you know, Queen Isabella financing Christopher Columbus to go bring gold for the Spanish Empire, uh, you can go back to Rockefeller, you know, digging one well himself, uh, you know, to strike oil, uh, in Pennsylvania and building the Standard Oil company or more recently Google, you know, they only raised one round of financing. VMware, we only raised one round of financing. So very small amounts of capital is enough to build. Very interesting if you are truly innovative companies.

Speaker B: Absolutely, completely agree and beautifully explained there. And uh, when backing deeply technical founders whose first commercial product is 18 months away, what early signals tell you that a group of brilliant engineers possess the raw, uh, commercial instincts required to successfully scale a business?

Speaker A: Yes, this is such a nice question for my work as an investor because I always back founders who have zero revenues, have not yet shipped their product. But the founders who are going to make it, they talk about customers, they talk about people who will pay them money to solve a problem that nobody else can solve. And that is the best signal. Everything else is talk. I tell my founders, yes, you can raise a lot of money at a high valuation. You'll get press releases, TechCrunch articles, bloggers will write about you. Doesn't matter. That's not a predictor of success. There is a little bit of a correlation, but it's not a predictor of success. You can go hire this famous person and say, oh, you know, John Smith is on my board or John Smith is my VP of sales or VP of engineering. It is not a predictor of success. John Smith has tried many companies and failed and not succeeded at that. What is a true predictor of success is the founders who cannot, uh, stop talking about their customers and their customers problems and understanding their customers better than even the customers understand themselves. Those are the founders that succeed.

Speaker B: Absolutely. And another term that you use frequently is, uh, founders that have a technical insight, you know, as a solution where if you tell a good engineer what you're doing, the how remains completely non obvious. Right. So how do you pressure test this at day zero before there is any product or market feedback?

Speaker A: Yes. So a technical insight which you already defined, um, you know, has to be something that can be commercially applied for it to make sense. For me as a venture capitalist and in my case I am looking for that 12 to 24 month window by when they have to make it commercial. How do I pressure test that? Uh, by keeping myself with an informed mind. I'm constantly educating myself on what the technologies are, where they work, where the pressure points are. I have a large set of friends. Um, I have 13,000, uh, friends in my iPhone. Um, not all of them remember me, but I remember them, uh, and I reach out to them and I talk to them and I read and I learn and I talk. And that is the job of a good venture capitalist, is to try to figure that signal out between all the noise that exists on these technical insights. Every day, Stanford, Harvard, mit, Wharton, um, they publish academic papers. Only one out of a hundred, actually maybe one out of a thousand out of them is going to become a commercial company. I like to go listen to the PhD defense at Stanford. I happen to live near there. Um, that's how you educate yourself, keep yourself informed and then you make a bet and say, oh, I think this will work. And even then, most of the time I'm wrong. Right? More often than not I'm wrong, but I'm right enough times that the business works and uh, you know, you are able to build a successful company because in venture capital you can only lose one times your money, but you can make 10 times, 50 times 100 times your money. And so that's what makes venture capital so fun.

Speaker B: Absolutely. That's what makes this game so exciting. And you purposely kept your funds lean and operate as a solo gp, directly contrasting firms that scale to billions in aum, um, and heavy internal management overhead. What has this journey taught you about running, ah, a concentrated alpha driven portfolio. And how do you convince institutional LPs that a key man risk is actually a structural alignment advantage?

Speaker A: Yeah, the second part of your question is easy to answer. In my partnership, which is engineering capital where I'm a solo gp, I have no partners. You have zero partnership risk. Partnership risk is the number one risk that LPs take when they invest in a fund. So I'll give you a very simple example of partnership risk. There's three people. Forget the large firms. There are three people, they start a fund or there's five people. By definition, one of them is a better investor than the other. Everybody cannot be equal by definition. Which means you are going to get the average of the three, right? You're not going to get the best of the three if you invest in a fund. So you are lowering your returns. And if the best one gets hit by a truck, you're going to get the bottom of what you invested in. That's the sad nature of how partnerships work. So I like to work as a solo gp. I tell them their risk is actually lower in my case, um, you know, barring me getting hit by a truck. And we have mitigation factors for that. Uh, you know, they have really zero partnership risk, so that lowers their risk. Two, I am a pure play. There's zero attribution risk in my case. Okay? When I do a deal, I am responsible for that. In big firms, what happens is people will often hide the ball. They will not let you know who really did it. And it's the person who's the most powerful who gets to claim the credit, not the person who did all the work.

Speaker B: Okay?

Speaker A: So that often happens, uh, in firms. And that's a very sad reality of how human beings work together. In my case, there's zero attribution risk. I take responsibility for my successes. I cannot hide from my failures. And so my LPs are very pleased that I've structured it that way. The biggest learning for me was that, um, venture capital is an idiosyncratic business. So far, it is a services business. Yes. AI is creeping in. And insofar as it's a Services business, it's best to structure it to meet the needs of the individual who is going to perform the best. Think of it like an athlete. You are not going to take a cricket player who just won the World cup and tell him, you're an amazing player and we're going to put you in the swimming competition in the next Olympics. That would be stupid. I mean, he's not going to win this swimming competition. Uh, and so that's how you should think about venture capital. What game are you good at? How do you want to play the game? How do you win at the game? There are great swimmers and there are great cricket players. And, you know, in this venture capital industry, I happen to be a certain type of animal, and that's the game I play. And I'm very happy and proud with the success that I've had so far.

Speaker B: Absolutely. And you were an operator previously, and now a vc. Tell me about that transition and how you think differently, work differently, and especially as a solo gp, which is a completely different ball game. So give me your experience in that.

Speaker A: So you're right. I mean, I started on the operating side. I'm an engineer. I still think of myself as an engineer. I wrote code, I ran a company, I ran product management at VMware for ESX, et cetera. Um, so I've done jobs before. I became a venture capitalist. And I left VMware with the intention of starting another company. That was what I really wanted to do. But I accidentally ran into, uh, the good folks. Kathryn Gould, Mike Shue, Bill Elmore, Jim Anderson. They changed my life. You know, I was a young kid. Um, I was in my 30s, um, and I had had some early success in my career. Uh, and they literally asked me and said, hey, do you want to be a venture capitalist? And I was like, why are you asking me? You know, and they said, well, you know, we think you're good raw material, and we'll train you and we'll help you. And so I'm grateful for the opportunity they gave me, uh, because it's a completely different job now. My real job is a money manager. Yes, I'm still an engineer. Yes. I spend all my time on technology, uh, and helping my CEOs and founders run their companies. But at the end of the day, I'm also a money manager. I'm a financial services professional, you know, technically speaking. And so, uh, you have to kind of straddle that line. Uh, and I think, uh, I was blessed that I got those first 10, 12 years of real hard Operating experience. Uh, and then now another, uh, you know, equal chunk of financial services experience on top of that. It's a good combination.

Speaker B: Yeah, absolutely. But being a VC is a lot of work. There are a lot of things to take care of. One is that you have to raise the fund. Right. The, uh, other part is that you have to source high quality deals and meet those people, build those relationships and then, you know, take those deals forward and onto a close. On the other side, you have to close the LPs, right, and close the fund on the timeline that you've decided. So too much pressures and too much selling happening, too much relationship building happening. How do you manage all of it by yourself as a solar gp?

Speaker A: Um, I would say you are largely true. It's a highly competitive business. And yes, there are multiple aspects to the business. Like any other high performing job, there are multiple aspects to it. Um, but in particular on the fundraising side, I've been blessed. So, you know, just like my job is to find great entrepreneurs, great LPs, job is to find great GPs, to find great investors. And I have been oversubscribed in every single one of my funds since day one. So I've had a waiting list of LPs, I've raised the hard cap, I've been oversubscribed every single time. And so, frankly, fundraising has been, you know, a breeze. I mean, I really didn't have to spend much time on it. Um, but yes, on everything else, it is a lot of work. It is a highly competitive industry, and so you have to work hard. There is no substitute for hard work. Elon Musk works very hard. Bill Gates worked very hard. Mark Zuckerberg works very hard. Uh, if you want to succeed in this industry, in any industry, you have to work very hard. There's no substitute for that. I mean, I love the famous quote, uh, by President, uh, Truman. My father actually used to talk to me about this. And I was growing up in India, in a village in India in the desert. You know, we were living in a mud house. My father used to talk about this to us and he used to say, where does you know in life? It's good to be lucky, right? It's wonderful if you can get lucky. But the truth is, the harder you work, the luckier you get. And this is a famous quote from President Truman. Uh, and I believe that, I believe all these people who are successful have worked very, very hard. Yes, they also had luck along the way, just like I had luck along the way. You know, I was very fortunate to get admission to Stanford. It's a wonderful school. I ended up here. I was very fortunate that MIT denied me admission. They rejected my admission letter because if I had gotten admission to mit, I would have gone to MIT over Stanford. But I ended up in Silicon Valley. You know, it's the luck of the draw. So there's always a luck element, but there's also hard work involved.

Speaker B: Absolutely. Yeah. I love it that you're also thankful for the things that didn't happen. Uh, and a lot of people sort of don't appreciate that enough. But every. No that you get everything that doesn't work out, it's actually for something that has to work out, so. Amazing. And you mentioned VMware. Right. So one question related to that. Uh, at VMware, your team found immense leverage by pivoting from a generalized platform play to a hyper specific pain point that is a server consolidation. Yes. Why are today's AI infrastructure startups struggling to find a specific consolidation equivalent? And how should they identify it?

Speaker A: Uh, every startup struggles to do this. Every startup. And all AI startups fall into this category. Anytime there is a new innovation, VMware was an innovation. AI is an innovation. Anytime innovation happens, by definition, people think, aha, if this was a platform, it would be better because that's what the innovation is. Right. But that's not how the economy works, unfortunately. The way the economy works is people buy solutions to the problems that they have. And so connecting these two things is the job of the entrepreneur to take an innovation and find a problem for which you can build a product that you can position today at a price point that the customer is willing to pay, which can be packaged together into a single solution. A lot of P's in that, uh, that's not an accident because all those P's have to come together, um, for an entrepreneur to be successful. And that is the job of the entrepreneur. The AI startups are doing it m, you know, and they're being successful. Clearly we've seen successful in the legal space. I mean we have startups like Harvey, we have success in the software development space. I have a company myself, Code Rabbit, wonderful Indian entrepreneurs. Uh, Harjot, second time with me, Gur. Um, you know, in the code review space, they've built a massive company literally overnight. Um, because they managed to connect LLMs, AIs, the insights of how they develop code to code review, which was a real unsolved problem. So connecting that is the job of the entrepreneur. That is what Steve Jobs did so well. That is what Mark Zuckerberg figured out that is what Larry Page did, and that's why these great companies exist.

Speaker B: Absolutely. Now we'll move to the rapid fire round, wherein I'll ask you six quick questions about the investing you're doing through the fund. And you have to give six quick answers. So the first one is what are the sectors and regions you invest in?

Speaker A: Software companies in the Bay Area at the seed stage, typically as the first investor that are highly capital efficient. So it's, uh, a very strict criteria. I have a very narrow practice. That's all I focus on. Most of the technologies today are primarily AI cybersecurity data. And then I have a small practice in what I call death of the x86, which is the change in the computer architectures that are occurring. But I'm open to any technical insights as long as they meet the initial criteria that I specified.

Speaker B: Great. Uh, what's the typical stage of investment?

Speaker A: Pre, uh, seed or seed stage. Some people call it inception stage. This, uh, is typically two people with an idea and perhaps they've started some development. Zero revenues. So pre seed or seed is the stage I invest at.

Speaker B: Great. And do you lead rounds?

Speaker A: Uh, almost always. I'm happy to follow, uh, if the opportunity calls for that, but I almost always lead the round.

Speaker B: Got it. And what's the typical check size you put in?

Speaker A: Uh, 1 to 3 million is sort of, you know, the two pizza team, enough to get you started.

Speaker B: Got it. And where can founders get in touch in case there's a direct way?

Speaker A: This is a hard one because I've kind of reached a stage in my career where I don't really solicit a lot of people coming out. So my request to my founders is please get a referral to me. Um, I don't take cold inbounds because I just get too much volume, too much spam, too many people reaching out. So, um, I am connected to 13,000 people. You should be able to get a referral to me. And that's the best way to reach out to me.

Speaker B: Awesome. Last one. Where can our listeners follow you?

Speaker A: I don't have a standard podcast and I don't publish a lot online, but I am on Twitter shemeetsdana. Uh, I am on LinkedIn. You know, Sedana is my profile. And, uh, those are probably the places where people will find me most. And of course my website, uh, which is pretty sparse, but I do occasionally write. Over there is www.engingcapital.com. so engineering capital is my firm. Engineering Capital is the website.

Speaker B: Lovely. I'll make sure to put all those links in the show notes below. Thanks so much for doing this. For this, Ashmit. Uh, and I wish you happy investing.

Speaker A: Thank you, Prashant. I enjoyed it.