In The Arena by TechArena - Engineering Capital’s Ashmeet Sidana on the Future of AI

Episode Date: July 16, 2026

In this episode of The AI Hedge, host Marc Austin (Founder & CEO of Hedgehog) sits down with Ashmeet Sidana, Chief Engineer at Engineering Capital (an investor in Hedgehog) and former VMware executive, to discuss the future of AI infrastructure, networking, and enterprise adoption.

Hello, I'm Mark Austin, founder and CEO of Hedgehog. This is The Hedge, a podcast where we talk about AI infrastructure and hedging AI infrastructure risk. And today have Ashmeet Sedana, who is the chief engineer at Engineering Capital, very experienced veteran in infrastructure business and in venture capital. So welcome to the show, Ashmeet. Thank you, Mark. Tell us a little bit about your back.

You've been in the valley quite a long time. I have, but it was a long and winding road coming here. I grew up in India, came to Stanford for grad school, and then I've lived within two miles of Stanford ever since, including doing a startup, SSI, where I was founder and CEO, ran that for five years, VMware, where I ran product management for ESX server, obviously became the killer product for VMware, very proud of that.

And then I've been doing venture now for, over a dozen years. Great. Okay. So you've been around the block a few times. You've seen technology cycles come and go. You've seen the expectation, trough of disillusionment, long-term value creation cycle multiple times. We're in the middle of one right now.

It's AI. How would you describe this AI cycle and what makes it different than ones that have preceded it, like mobile or web or social? I would say we are at the start of another cycle because AI is so transformative. It is a new form of computation that we have invented. And it is going to have an impact on the entire world, on every aspect of the world, the economy, the politics, the technology, et cetera, et cetera. And so this is definitely larger than the Internet wave, larger than the iPhone,

larger than the desktop, PC, et cetera. And it's going to be fun. So hang on. Yeah, it's already a pretty good ride. Okay, there's this AI shift that's coming. It's changing everything. How should CEOs, CTO, CIOs of large enterprise companies be thinking about this? What's their opportunity?

What's their risk? What should their AI strategy be? Like I said, this is a new form of computation. It's a new form of intelligence. And without getting into debates about, is it artificial general? intelligence or super intelligence, whatever we've got already is pretty darn intelligent. And it's extremely useful. You've already seen the impact it's had on software development.

Similar types of impact is going to happen on every aspect of knowledge work. So what I would tell CEOs right now is that you have to embrace it. There's no other alternative. You have to embrace it. And you have to figure out how you are going to leverage and benefit from it. Because in 10 years, there's going to be no such thing as an AI company. Every company is going to be an AI company. And so there's no alternatives.

You have no choice at this point. The decision is easy. Yeah. And you mentioned that you were at VMware. And can you tell us a little bit about the origin story of ESX server and the role that played into what ultimately evolved into cloud computing? And really the infrastructure side of that, okay, how do I transform? to be an AI native company going forward.

Yeah, so VMware started as a research project at Stanford University. Professor Mendel, Rosenblum, my professor, he had been doing work on this. They came out with a developer product, now called Workstation, which was a form of virtualization running on top of Linux or Windows. So you could run Workstation on top. And the primary use case there was for kernel developers, being able to test their drivers, et cetera, on Windows in an easy way. It was really painful before Workstation came along.

That's when I joined the company. It was a test and dev tools company. We were selling it for anywhere between $99 to $300. It was a small little startup. And the opportunity was to take this R&D project, what became ESX server, and find a new use case for it. That's the company I joined as an employee. And they said, figured it out.

And the initial hypothesis was that we would go into the service provider business. We would want to create something equivalent to what became Amazon AWS, et cetera, and sell that as an enabler. I was skeptical of that. And I actually am the person responsible for killing that project inside VMware because I found a much better use case, which was server consolidation in the enterprise. The enterprise had all these old legacy Windows 95, Windows NT, Windows 2000. servers, and virtualization offered a marvelous solution where you could consolidate all of these, run them on newer hardware, which was much faster, much better, clean up your data center, and you got all these other benefits, better backup, better disaster recovery, lower cost,

better performance, et cetera, et cetera. And that's really what made VMware into the giant company that it became was that enterprise push for server consolidation. And I'll take partial credit for that. Great. Okay. So we hear a lot in these really early innings of AI about infrastructure, right, which is kind of true in most market shifts, right? The early stages of that market shift, or if you're investing, you know, infrastructure is a pretty good place to make your initial bets. So for enterprises who've sort of gone through this cycle of using VMware and ESX server,

for the use case you just described about, maybe we're moving some workloads to public cloud. And now we've got AI. And anybody watching the news knows that, okay, Nvidia is like the leading AI infrastructure company. It's a $4 trillion company. And GPUs are hard to come by. They're really expensive. But I know I need these GPUs to be able to do AI. How should enterprises be thinking about AI infrastructure in the near term and in the long term? Yeah. So AI, as I said, is a new form of competition. and it has very different demands on infrastructure than your Oracle server, your web server, your SAP server, et cetera, that we are used to running in the infrastructure.

So because the demands of AI are so different, it's going to be an entirely new infrastructure layer. That's where TPUs and GPUs and these massive data center buildsouts, et cetera, come in. So that's one characteristic. It's a new architecture. Two, it requires massive forms of computation. So huge forms of data, huge forms of networking, huge forms of compute are required to make AI work. That's where the large, in large language models really comes in. So we're talking orders of magnitude more than previous forms of computation.

And so that's why we've got an entirely new form of infrastructure. A lot of this is going to be built within the form of public clouds as in the hypers. They're not going to give that market up. They understand that they're going to be there. there's going to be new clouds, entirely new forms of clouds which are going to be created because the demands are so different. And then there are some enterprises, the Global 500, the Global 1000, who will want to run some things in-house, primarily because they have the scale, but also because there are some things like privacy

and regulation that are going to demand that they do it. So we're going to see all three forms. They're all going to have to reinvent themselves. And we're going to see an entirely new build-out occur of infrastructure over there. Yeah, so let's talk on that note. You mentioned data privacy. Let's talk about data sovereignty for a minute. And there's a rapidly changing geopolitical landscape right now. So just start with what in your mind is the definition of data sovereignty is a real thing and how are people going to implement it? Data sovereignty is absolutely a real thing. And yes, what has happened is over the last few years due to events not related to technology. These are primarily geopolitical. forces outside the context of pure technology, what has happened is the world has been de-globalizing. We obviously saw a tariff war, which has nothing to do with AI, erupt. And so

those are just examples of this de-globalization trend that we are seeing. And all indications are that is the direction we are headed in. So in addition to the privacy requirements, there are sovereignty requirements. There are going to be independent forms of architecture and clouds that will be created. And that's why I see the golden age of infrastructure coming ahead of us, where people will have to build it. And those who don't embrace it and don't build it, they will get left behind. They have no choice. Yeah. Okay. So if you are a sovereign entity or a global enterprise and you're building out sovereign AI infrastructure, what are all the layers of a stack, right? Because you've been looking at stacks over time for a long time. You said previously,

hey, look, this is a fundamentally different way of doing compute. What are the key layers in the AI infrastructure stack? And where do you think the best investments are going to be? The last one is always hard to answer. And that is my job, and that's where I spend a lot of time thinking about it. So we can talk about it. But let's start with the first one first, which is one of the layers of the stack. So the way I see it today, we have the stack consists of the power layer,

which is literally just electricity, which is what powers all of these data centers, and because the demands are so high, we are seeing a renaissance in the way people are thinking about power. On top of that comes the physical layer, the data centers. Those obviously from a sovereignty perspective have to be geographically distributed where they are. China is always going to have data centers in China that they can control.

We will always have data centers that we want to control. The Middle East has already started investing very heavily over there. has woken up and realized they're going to have to do it. So we're going to have five, six, seven different regions that are going to get created from scratch. Above that physical layer, you have the data center, the actual compute as we think of that we're going to build, which consists of the chips, which is typically GPUs, TPUs. We're going to see a proliferation of architectures.

We saw Cerebrus grow public, have a very successful IPO last week. So new architectures are going to develop. There's nothing which says that you have to run our architecture. a GPU and there's nothing which says that you have to run on a TPU or a CPU. That's not the be all and all. Innovation is occurring. New things will come out. On top of that is where the software starts coming in.

How do you connect these, the networking, the storage, where does the data actually physically reside? You're not going to run this on a Net app with all due respect to NetApp. That's not going to be the architecture where all of these terabytes and petabytes of data will reside. How will you connect these? The networking. where will you store the data?

On top of which is the LLM, that's the logical layer which is using this, the large language model. And on top of that will sit applications, which will consist of two pieces, the AI layer and the traditional layer. The traditional layer doesn't disappear. Doesn't mean you don't need a database

if you're still not using an, if you're using an LLM, you're still going to need a database, you're still going to need a web server. And so all of that infrastructure has to be built out. And because of the change in the scale and approach of AI, we will see a different architecture emerge.

Okay. So you mentioned networking in that stack, Hedgehog's an AI networking company. And we've been talking about Nvidia as well. So there are Nvidia reference architectures for Infiniban, for Spectrum X, which are sort of the two networking technologies that go into SuperPod reference architectures. And you mentioned there's going to be new AI accelerators.

You mentioned the Cerebris IPO last week. Do you think that new players like Cerebris are going to, to define their own reference architectures, their own super pods, their own network reference architectures. It's guaranteed that's going to happen. And what typically happens is that people invent a new compute infrastructure and then storage and networking are usually a half inning behind in terms of catching up with what are the demands that have come from compute and how they have changed. People slap on their old stuff onto the compute and then they realize it doesn't scale,

it doesn't work, it's not reliable, et cetera. And then by then, the networking and the storage stacks catch up. And right now, we are in the first inning of the infrastructure build out for AI. So there is going to be a lot of innovation. And yes, we are definitely going to see new storage and networking architectures get built over here. Both are going to be very interesting and dynamic areas for investment. And one of the things we did recently at Hedgehog is we contributed a reference architecture to the open compute project.

which is, it's really, it's a nonprofit organization. It was founded by Microsoft and Meta to really create an ecosystem, a community of vendors who could provide open specification, hyperscale infrastructure. I guess the question is, open networking. When you first made an investment in Hedgehog, that's really kind of where we started. Why is open networking important in this whole AI stack? Networking is the one part of the technology stack that has to be open by definition.

Where being open gives you benefits that cannot accrue to closed networks. People have been trying closed networks since the dawn of computing. I'm old enough to remember when ATM was a thing, when Sonnet was a thing, et cetera, all of those token ring was a thing. All of those things died behind the force of Ethernet, Open Networking, TCPIP, etc. And I predict that is going to continue to be the case. In fact, the very word network effect, the phrase network effect comes from networking. Because networking is where the benefit accrues exponentially when you are open.

So as much as there are closed solutions today, I appreciate that close solutions exist. Close solutions are valuable and they are sometimes at the leading edge, but they are going to be left in the dust. What will win? Guaranteed is open networking. There's no question in my mind about that. Great. So as organizations are rushing to adopt AI, how do you see that enterprise journey? How do they start today?

Just get started with AI. How do they, in four or five years from now, how do they become that AI native company that you were describing earlier? And what are the risks along the way on that enterprise journey? Yeah, there are lots of risks also, unfortunately, along this journey. So I already mentioned one, which is that, you know, the earlier proprietary networking stacks have a lead right now in some areas because that's where Nvidia started, they bought Melanox, etc. I predict it's going to be all open.

That is by definition of transition that's going to occur. Because the buildout is so rapid right now, the greenfield, new buildouts will take on open first and then eventually everyone will have to move there. That's just something that is going to happen. Overall, on the AI journey, the other risk which exists for enterprises, is that the rate of innovation in AI is extremely high right now. It's not like we've solved AI. And people are innovating at the chip layer, as we talked about.

They're innovating at the architectural layer. So transformers is not the be-all end-all solution for AI. Someone tomorrow is going to come out with a new academic paper, perhaps going to win a Nobel Prize down the line, and come up with a new architecture. And everything we've learned about LLMs is going to go out of the window, except they're still going to need data. they're still going to need to talk to each other,

and they're still going to be able to be more intelligent than they are today. Yeah, they're still going to need a network, too. And they're going to need networking. That's how data talks to each other. That's how computers talk to each other. And the risks are tremendous, but a greater risk is not embracing it. If you're an enterprise, you have zero choice today. There is no doubt in my mind, you have to embrace it.

You will get killed otherwise, because we don't know when that next generation architecture will come. The architecture today is so good. AI today is so good that it leaves everything else at the dust and you have no choice but to embrace it and move forward as fast as you can. And so that's what's going to happen in the enterprise over the next three, four, five years. They are going to adopt it. They will start with point applications.

We're already seeing for software development. We're already seeing customer support. All of those things are being overtaken. That is then going to diffuse through the entire enterprise. and there will be no such thing as an AI application or a non-AI application. It will be all AI. Yeah. And if there is really an end of transformers and something better that comes along,

if you've already worked out the use cases where you're going to apply artificial intelligence, and there's a new method of computing artificial intelligence, if anything, it's just going to be more cost efficient, right? More cost efficient or more performant, right? I mean, we don't know. There's no limit to intelligence, right? we don't know what more intelligence could come along. But we do know for a fact that whatever this form of intelligence is,

it will have data which has to talk to each other. In other words, we network with each other and communicate with each other. And so that is just going to be a core part of whatever comes in the future. This is really about information theory. There are philosophers and physicists who posit that the world is information, that the entire root of physics is information. I'm not going to get that philosophical here. But for sure, it's an intelligence world going forward.

Okay, so we've been talking about networking. We've been talking about infrastructure, Nvidia being the early leader in AI infrastructure. And you mentioned Melanox. Can you talk about the Melanox acquisition, why you think Jensen did it, why did he do it when he did it, and how does it rate on the all-time list of M&A success?

It was a brilliant acquisition. It was definitely the right, decision for Nvidia at that time. And it was the right answer. So Jensen is a wonderful CEO, and I commend him for picking up what was sort of an undervalued asset, not totally undervalued, and understanding the potential for applying that asset from a high-performance compute, which is kind of the market that they had gone into earlier, and applying it to this nascent market AI that Jensen, to his credit, saw before anyone else. So he doubled down on it. He bet on it, he's built it. And this is what always happens. When there's a brand new market with a

brand new technology that comes in, there are early adopters, they go in, they build whatever it takes to go solve that solution. Remember, some of the earliest cars that were sold in America 100 years ago were electric cars. We forget that. You may have the more, the default, blew everyone away because gasoline was such a better solution. It took 100 years for Elon Musk to come back and bring that back. So maybe 100 years from now, Melanox's. will come back. But what happens after this early innovation is that people start figuring out what this new stack is, what is optimal for that stack, and then a lot of innovation occurs to build that. That is the stage we are at with AI. That is why you are seeing open compute, open storage,

open data center architectures, et cetera, rushing in. And that is where development is now going to happen. That will be the main state. That's the middle of the curve. That's where the bulk of the spend, build, and usefulness and productivity comes from. The earlier people, you know, sometimes they say you can be the pioneer and you can end up with an arrow in your back. That's the risk of being a pioneer. I'm not saying there's arrows in the back of Nvidia. Again, Jensen has executed flawlessly as a CEO in terms of establishing the GPU architecture.

But again, I'll reference the Cerebrus IPO as an alternative architecture. I'll reference TPUs coming in from Google, et cetera. And the Nvidia acquisition of Kroc, right? And in the Nvidia acquisition of Brock. That is again, Jensen trying to, you know, race ahead of everyone else, recognizing that a chip that was designed for PC gaming was used for Bitcoin mining, then applied to AI is not the be-all, end-all, and the right architecture to build your massive AI infrastructure.

Yeah. And, you know, Jensen said in his keynote, hey, look, we're vertically integrated. We're horizontally open. and have really taken some fundamental steps toward... No, he's a great CEO. He understands that if he doesn't stay open, he's going to get killed. He does not want to go the way IBM went or the way H.P. went or the way Deck went.

We are old enough to remember those were the three biggest companies in the world. And nobody talks about them today. They are irrelevant. They are not that they've disappeared or they don't have revenues. Yeah, there are legacy customers. And every time I go to the ATM, yeah, I'm using an IBM mainframe in the back end. But it's irrelevant to the AI store. because they didn't innovate, they didn't stay open,

they didn't understand what you have to do. Jensen doesn't want to make that mistake. He's smart. Again, I'll make a second prediction over here. Jensen will be forced to adopt open networking. He will embrace it. He will adopt it. And at some point, Melanox will become irrelevant.

Yeah, well, I think he already is, really. I mean, with Spectrum X, right? It's Ethernet. We support it. It's part of our open network fabric. And Nvidia has been a great partner to us in helping us do that. So, okay, well, now I got to ask. I got to ask the closing question.

So you invested in Hedgehog, and you're a board member, you're really a trusted advisor for me. It's been a great journey so far. Why did you invest in Hedgehog almost three and a half years ago now? Yeah, so Hedgehog had all the ingredients of an engineering capital company. When I met you, it was a company which had a great team, obviously yourself, originally Mike, now Mnish, Sergei, etc., on the team. So the team is a critical starting point. For me, there are no great companies without great teams. They cannot exist.

And so that's the starting point. You were going after a market space that was clearly underserved. Everybody talks about LLMs and models and everybody talks about AI. Nobody talks about the networking, not realizing that there is no model without networking. There is no model without interconnecting all of these things. There are no applications without networking. You and I are talking right now over a network. And so it was clear that the market area was, I wouldn't call it completely greenfield,

but it was seriously underserved. And most Silicon Valley VCs, like any financial market, they're in the herd mentality. They run away. They run towards where everyone else is. And so when I saw that opportunity, the two elements were already there. And then, of course, when you layer on top of it, the open approach, which I am a huge believer in, especially for networking, it was an obvious bet to make over there. Mark, I'm very grateful that you chose me as your investor.

It's been a pleasure to watch you execute. You are a fabulous entrepreneur, CEO, and you're executing beautifully. You have all the hallmarks of companies that I have had the privilege of investing in companies that went public, companies like Azure and Rubrik or sold for billions of dollars like Signal FX, now robust intelligence, Cisco, etc. I've been investing in AI for the last seven or eight years. And I think you're on exactly the right trajectory. Great.

Thank you for the kind words, Ashmi. Any advice for hedgehog going forward? What should we be thinking about? Do more of the same. Stay focused on the customer, solve real problems, build revenues, everything else will take care of itself. I like to say revenue solves all problems. And ultimately, no problem can be solved without revenues.

In other words, you can patch over things temporarily by getting a financing or hiring the right person or doing a preference. release or signing a partnership or doing business development, ultimately it all catches up with you. I'm thrilled to see the customer traction you have, the deployments that are ongoing right now with Hedgehog. That is the leading indicator. Eventually, everybody will figure it out and they will be coming running to us. So it's just a pleasure for me to watch it. That's why I love being an early investor. It is a lot of fun. All right. Hey, Ashmeet, thank you so much just for your wisdom, your historical perspective and your support. And really,

enjoying this journey with you and it's going to get even more fun. Thanks a lot. Thank you, Mark. It's a pleasure working with you.