Mike Volpi on Why AI Infrastructure Will Outlast the Hype Cycle

Mike Volpi 0:00 We live in a world where demand for AI capacity tokens is near infinite, and supply is incredibly constrained. So another way to put it is we all just live in Jensen's world. You were doing back then M&A for the most valuable company on the planet. The retro on the M&A process that I think worked well for Cisco was we were not afraid to fail, and we did it with high enough frequency that we got a few outliers. You bought this thing for $500 million, and like it's doing a billion in earnings two years later.

Raghu Raghuram 0:29 When you see the AI buildout, how do you look back on your previous experiences?

Mike Volpi 0:34 If you look at the revenue ramp of Anthropic, OpenAI are definitely experiencing growth rates that were far superior to what we saw at the time. Do I think the buildout that's happening now is useful and will be fully utilized, if not more? 100%. Will it be owned by the same people that own it today?

Raghu Raghuram 0:53 Great to have you here on the podcast, Mike, and it's a very, very opportune time to have you because you were one of the architects of the great internet buildout when you were at Cisco. And, uh, now of course, we've got the great AI buildout happening. So, uh, Guido and I and the audience would love to get your thoughts on then and now, uh, from all different dimensions, from the actual buildout, the capitalization of it, the technology differences, et cetera, et cetera. So, so let's start for those that don't, uh, know or remember, uh, Mike was a senior exec at Cisco, overseeing a lot of the M&A and strategic moves that Cisco made in the late '90s. Um, and of course, that was the time the internet got built out, so Mike was in the middle of all of it. Um, so clearly let, let's start from there. So yous were there when the internet started taking off, and then you guys rushed in to, to a lot of great things, and the industry rushed to build a lot of bandwidth, et cetera, et cetera, which, uh, for some time was not useful.

Mike Volpi 2:06 Mm-hmm, mm-hmm.

Raghu Raghuram 2:06 And then it turned out it was actually too little, right? And there are some loose similarities to what's going on now. So when you see the AI buildout, how do you look back on your previous, uh, experiences?

Mike Volpi 2:21 Well, first of all, it's an absolute privilege to be here with two legends here. Both you, Raghu, and, and, uh, in the case of, uh, Guido and I, we got the chance to work together at least once. So, so, um, uh, it's fantastic to be here and, uh, as you said, an interesting time because, you know, history doesn't repeat itself, but it definitely rhymes.

Raghu Raghuram 2:41 Yes. [laughs]

Mike Volpi 2:42 So, so, um, so yeah, there are some similarities. Um, maybe useful to give some context on essen- uh, some of the observations of the late '90s buildout and what some of those things might apply in today's world, uh, versus not. Some of the kind of similar themes was that we, we were building out this technology called the internet and the web and all the things that rode on the internet. We had expectations that this would be massively transformative to the world, and we imagined a wide range of applications that would ultimately massively benefit from this buildout. Applications were different media types, right?

Raghu Raghuram 3:23 Yep.

Mike Volpi 3:23 So first we just had pictures and text, and then we had audio, and we had video, uh, and then we had like high-quality video, applications like commerce, applications like social networking. All of these things were imagined in the '90s, uh, but didn't exist. Uh, in parallel with that, there was a very big surge in demand. At a minimum, there was a surge in demand for connectivity.

Raghu Raghuram 3:47 Yeah.

Mike Volpi 3:48 And on top of that, a reasonable surge in demand for, uh, for bandwidth. Um, and, uh, in parallel with that, there was a great influx of capital of two shapes and forms, equity and debt. Um, so, uh, i- if you look back at sort of what happened, first of all, in b- in the bandwidth world, uh, and in particular in fiber optic bandwidth, the elements, the increments of increase of the amount of bandwidth are quite different than compute.

Raghu Raghuram 4:19 Mm-hmm.

Mike Volpi 4:19 And that's because, you know, we invented this technology called WDM that essentially takes a piece of fiber and said, if you used to put, you know, one gigabit on there or 10 gigabits on there, by putting in more colors of light in, in that same piece of fiber, you can now have 40 times or 80 times that capacity. So essentially, capacity increased quite dramatically with the advent of technological change. That's actually different-

Raghu Raghuram 4:43 Yeah

Mike Volpi 4:43 ... 'cause compute ultimately is bound by something that resembles Moore's law. So the pace of improvement, uh, uh, is, is more moder- more moderated. Hard to believe, but it is actually more moderated in compute than it was at the time. Uh, and the other component that is, I think, different is that the practical applicability of AI, I believe today is higher than the practical applicability at the time of network bandwidth.

Raghu Raghuram 5:14 Yep.

Mike Volpi 5:14 Uh, and that represents itself by the revenue ramp of the providers. So if you look at the revenue ramp of, uh, Anthropic, OpenAI, I don't know what the numbers look like within, within Google. Uh, e- even the providers of the open source models like Fireworks, Together, Modal, et cetera, et cetera, are, uh, definitely experiencing growth rates, maybe not quite at the same margins, but certainly growth rates that were far superior to what we saw at the time.

Raghu Raghuram 5:42 Yeah.

Mike Volpi 5:43 Now, uh, then what happens, uh, uh, at, at some point, there's a shock that some of these businesses that are built on the internet are not good enough, uh, and then that ripples through the industry, and ultimately, the players that were most saddled with debt- ... went bankrupt. Um, now it's interesting to note that that debt was buying assets, was fiber optic cables, and those assets actually turned out to be quite useful in the future.

Raghu Raghuram 6:17 Exactly.

Mike Volpi 6:18 Except they were operated by somebody else because the guys that went bankrupt didn't have them anymore.

Raghu Raghuram 6:22 Yeah.

Mike Volpi 6:22 So, uh, the first thing to recognize that's the underlying assets that were built at the time as bandwidth, and I suspect this, you know, compute this time-

Raghu Raghuram 6:30 The cycles, yeah

Mike Volpi 6:31 ... it'll be useful. Th- there's no question that'll be useful. Whether it will be owned by the same companies or not is an interesting economic question. So, you know, that's a long-winded answer to say, do I think the build-out that's happening now is useful and will be fully utilized, if not more? 100%. Will it be owned by the same people that own it today? Lot of interesting dependencies, and a lot of things that in fact are outside of the control, 'cause they're, they're macroeconomic and geopolitical and so forth. It's really outside of the control of the builders themselves.

Raghu Raghuram 7:06 Mm-hmm.

Mike Volpi 7:06 And in particular, when I talk about the builders, uh, you know, some of the balance sheets are pretty healthy, right? At last count, when I was talking to folks at Nvidia, there are 200 Neoclouds, and those balance sheets are not good.

Raghu Raghuram 7:20 Yes.

Mike Volpi 7:21 So, so I'm gonna guess that there's, uh, I, I'll predict that there'll be some-

Raghu Raghuram 7:26 Not all of them, of course. Yeah

Mike Volpi 7:26 ... y- yeah, some of them better than others. But broadly, I expect that there will be an economic correction, uh, in the market. Uh, but the underlying assets will be quite useful for a long period of time.

Guido Appenzeller 7:38 I- isn't that one other big difference from how it was back then, which is that everything is now much, much larger. I mean, back then we were investing a small fraction of a percent in, in infrastructure for, for the internet. Today, we're investing 2% or so of our GDP, right, in, in, uh, AI infrastructure.

Mike Volpi 7:56 Yeah. Uh, well, it is much bigger, but i- it's an interesting concept, right? 'Cause in the 100,000 feet abstraction, what is it that is being built, right? Back then, it was communication that was being built. Today, we're building intelligence, and the economic value of intelligence is probably higher than the economic value of communication.

Guido Appenzeller 8:19 Probably, yeah.

Raghu Raghuram 8:20 Yep.

Mike Volpi 8:20 Right? Um, so, uh, so, so, uh, I think it's, it's unsurprising. I also think that, that, believe it or not, I mean, it's been nearly 30 years, the markets are more efficient now, right? Which is why so much more risk capital is willing to make its way, uh, i- into AI, which, you know, uh, e- e- even back then, uh, was not able to. So, uh, net is I, I'm actually more optimistic about this cycle, but I also believe that there will be some correction and some cleaning up of the industry structure.

Guido Appenzeller 8:52 The, the role of IT has also massively increased, right? Back then, I mean, in the late '90s, PCs were, were somewhat common, but, but not ubiquitous. Today-

Raghu Raghuram 9:02 Yeah

Guido Appenzeller 9:02 ... so to speak, software has eaten the world, and it's sort of, for any company out there, it's, it's a very central part of what they do when-

Raghu Raghuram 9:09 Cost of distribution of intelligence and it's-

Mike Volpi 9:11 Oh, for sure. And not just any company-

Raghu Raghuram 9:13 Sort of low

Mike Volpi 9:13 ... but like, you know, seven billion people have a supercomputer in their pocket.

Raghu Raghuram 9:16 Yeah, exactly. Yeah.

Mike Volpi 9:17 So, so, you know, there's, there's, you know, there's m- much more evenly spread, uh, consumption of, of the, uh, the underlying, uh, benefit that's being produced.

Raghu Raghuram 9:28 Yeah.

Raghu Raghuram 9:29 Uh, don't fully recall all the puts and takes in the '90s, uh, and early 2000s on the financial side. Here, obviously, one of the recent shocks to the system was the rising interest rates.

Mike Volpi 9:42 Hmm.

Raghu Raghuram 9:43 And the scale of capital causing people to say, "Hey, this is gonna crowd out everything else, and we may not even have enough capital." During the bandwidth build-out and the internet buildout, that was not the case, correct? There was no sudden increase in interest rates, was it?

Mike Volpi 10:00 No, there wasn't a massive increase i- in interest rate. There, there's a couple of big differences, right? One, one is that more... I would say on a proportional basis, I believe more of the capital came from the balance sheet of the equipment vendors-

Raghu Raghuram 10:13 Mm-hmm

Mike Volpi 10:13 ... than today. Today, it's a more, uh, capital markets driven, uh, build out. So it, you know, it's private debt, it's public equities, et cetera, et cetera. There is Nvidia vendor fin- I don't think you're supposed to call it vendor financing, but it looks a lot like it. [laughing]

Raghu Raghuram 10:29 [laughing]

Mike Volpi 10:30 Uh, so there, there's a, there's a little bit of that, but, uh, uh, and there's some corporate financing, but it's largely equity.

Raghu Raghuram 10:36 Yeah.

Mike Volpi 10:36 Uh, uh, and, uh, it's not tied to specific equipment purchases and so forth. So I, I'd say the capital structures are a little cleaner now. Um, the, the one thing that I think interestingly, though, favors the '90s is a lot of the money went into the ground in the form of fiber optic, and when you put a fiber optic in the ground, it's good for a long, long time. It's like it's a usable asset for 20, 30 years. The-- I, I think one of the deterministic factors is the compute you put in the ground, what is its actual life? And I, I-- this is a fascinating phenomenon, right? Because the dominant part of that compute is, is, is chips. It's, it's, it's GPUs. That's the-

Guido Appenzeller 11:19 80-plus percent, yeah.

Mike Volpi 11:20 Yeah. It's the big chunk of it, all right? So I remember back when people were very excited about Hopper H100s, right? It-- you were, you're-- y- y-- they were like, "Okay, well, maybe it has a five-year useful life," and you're renting it out at a, you know, buck 80, $2 per GPU hour, okay? And that was four years ago. So it-- the chip should be defunct now, like near defunct. Not at all defunct.

Raghu Raghuram 11:42 No.

Guido Appenzeller 11:42 Nope.

Mike Volpi 11:43 Not only that, it rents at a higher rate.

Guido Appenzeller 11:44 Yes.

Raghu Raghuram 11:44 Yeah. It's a, it's a-

Mike Volpi 11:45 It's like, it's like it's 250 or 3. So the, the, the kind of the interesting part is we thought this asset that have a five-year life, partly because of AI itself, is a useful asset for quite a bit longer. And in that conte-- I mean, if you're in the business of financing an asset and that asset ends up having a much more, a longer useful life, then you're the beneficiary of that because your business models were constructed on a short-- So in some sense, I, I actually think- This, the, the fact, it, it-- an unintended consequence of AI being good at software is that the underlying asset life is getting longer, thereby helping it to be more financeable. And that's an interesting phenomenon that I didn't quite expect-

Raghu Raghuram 12:32 Yeah

Mike Volpi 12:33 ... when this thing got underway.

Raghu Raghuram 12:34 Although the, although the investors are not, the debt investors are not broadly accepted the extended life of a GPU, but the data clouds clearly believe that.

Guido Appenzeller 12:42 I, I, I'm not sure, I'm not sure I have either. I mean, like, it, right now it seems what's mostly happening is that we have a massive scarcity of an asset, so even previous generations are still worth a lot, right? If we ever get to a point where we have, uh, you know, an abundance of, of GPUs, that might completely change. You might go, you might go back to those numbers, uh, in the chat.

Mike Volpi 12:58 Yeah. So I think the key metric is, is, is, uh, uh, how much token revenue can you extract per watt?

Guido Appenzeller 13:05 Yep.

Raghu Raghuram 13:06 Yep.

Mike Volpi 13:06 Right? And-

Guido Appenzeller 13:10 Maybe, yes.

Mike Volpi 13:11 So the, the Hopper is gonna be a lot worse than Blackwell's gonna be a lot-

Guido Appenzeller 13:14 That's correct

Mike Volpi 13:15 ... worse than Vera Rubin, right? They're all just more efficient at getting dollars out of a watt. So in theory, at some point, when you have sufficient Vera Rubins, it doesn't really make sense to run H100s anymore-

Guido Appenzeller 13:26 Correct

Mike Volpi 13:26 ... because, like, it's just a worse way of converting electricity into dollars.

Guido Appenzeller 13:30 Correct.

Mike Volpi 13:30 Right? But the problem is right now that the demand is so high that even if the watt to dollar conversion is low, you're going to still use the asset. So, uh, a- and right now the scarcity is so significant. And ultimately, you know, at some point in the conversation maybe we get there, but ultimately the scarcity is in the manufacturing of the semiconductor itself.

Raghu Raghuram 13:53 Yeah.

Mike Volpi 13:53 And that doesn't look like it's changing anytime soon.

Guido Appenzeller 13:56 Correct. Yeah.

Raghu Raghuram 13:56 Yeah.

Mike Volpi 13:56 And so then scarcity persists, then the life of the asset continues to be long.

Guido Appenzeller 14:00 Yeah.

Raghu Raghuram 14:01 Yeah.

Guido Appenzeller 14:01 Yeah.

Raghu Raghuram 14:01 Yeah.

Guido Appenzeller 14:01 I think it's correct. Um-

Raghu Raghuram 14:02 Yeah. And the other big difference, there was a period of time where there was unused bandwidth. Today there is no such thing as an unused-

Mike Volpi 14:10 It's true. [laughs]

Raghu Raghuram 14:12 ... GPU, right? I mean, and-

Mike Volpi 14:13 There's a, there's a bit of unused energy-

Raghu Raghuram 14:15 A little bit

Mike Volpi 14:15 ... and, like, you could probably, you know, I don't know if we're gonna talk about networking, but with better networking-

Raghu Raghuram 14:19 Yeah, of course

Mike Volpi 14:19 ... you could get more flops out of a given GPU.

Raghu Raghuram 14:22 Yeah, I was just gonna say that. A fuse might have factor in certain use cases and, uh, how the people have built it, but, uh, but certainly it's generally not.

Mike Volpi 14:31 Yeah.

Guido Appenzeller 14:31 Back when you were at Cisco, w-were you still there when Cisco was the most valuable company on the planet?

Mike Volpi 14:36 I, yes indeed, I was there.

Guido Appenzeller 14:37 So you were doing, back then, M&A for the most valuable company on the planet, and it worked pretty well, I want to say, looking back, right? That's-

Mike Volpi 14:44 I was very happy with it, yeah

Guido Appenzeller 14:45 ... um, what would be your words of advice to Nvidia in, uh, in 2026? Like, the, whoever runs M&A there, uh, is in exactly the same spot as you are, correct? As you were.

Mike Volpi 14:56 Yeah, Vishal's, [laughs] Vishal's in the same spot I am.

Guido Appenzeller 14:58 [laughs]

Mike Volpi 14:59 Um, look, I, I, it, the, the retro on the M&A process that I, I think worked well for Cisco was that f-fundamentally, uh, we were f- okay with making enough acquisitions that we got a statistically significant outcome on those acquisitions. Meaning some worked really well, some worked so-so, some didn't work at all. And, you know, if you kind of drew it out, one in, one in each of those three sectors. Th- a third of them worked really well, a third of them were whatever, and a third of them failed. Um, the benefit that we did is we were not afraid to fail, and we did it with high enough frequency that we got a few outliers, uh, Andy Bechtolsheim's company, uh, uh-

Guido Appenzeller 15:50 Yep. Sun, right

Mike Volpi 15:50 ... Mari Mazzola's company, uh, a, a bunch of these things that were sort of, you know, even for us, were venture-like returns. So you bought this thing for 500 million, and, like, it-it's doing a, a billion in earnings two years later.

Raghu Raghuram 16:08 Yeah. [laughs]

Mike Volpi 16:09 So- [laughs]

Guido Appenzeller 16:09 That's amazing. [laughs]

Mike Volpi 16:10 So, a-a-and so it ended up that, uh, during the period that I was there, uh, by... I, I joined in '94, and I was, like, the one biz dev hire. Uh, in 2000 and, uh, in 2000 roughly, uh, 50% of Cisco's revenue came from acquired sources. So the, that was incredibly useful, and we actually never really spent much more on a given acquisition than 4% or 5% of our market cap.

Raghu Raghuram 16:35 Mm-hmm.

Mike Volpi 16:36 Most of the acquisitions were actually, quote, "cheap" relative to our market cap standing.

Guido Appenzeller 16:41 How, how much would that mean for Nvidia? I don't even want to think about that. [laughs]

Raghu Raghuram 16:44 Yeah.

Mike Volpi 16:45 Yeah. Well, my, my rationale at the time was, like, if your stock trades l- more than the pa- price you paid for the acquisition in one day, then you can do the acquisitions as many as you want.

Guido Appenzeller 16:55 [laughs]

Mike Volpi 16:55 No-nobody, nobody will care. [laughs]

Raghu Raghuram 16:58 And, uh, all of them done with equity or was-

Mike Volpi 17:01 All, all equity.

Raghu Raghuram 17:02 All equity.

Mike Volpi 17:02 Yeah, almost e-everything was equity.

Raghu Raghuram 17:03 And, and-

Mike Volpi 17:04 A, a handful of the small ones were cash, but it was all equity

Guido Appenzeller 17:05 ... and at Cisco, I mean, they became major product lines. So it was not like this was a little tuck-in-

Raghu Raghuram 17:10 Yeah, exactly

Guido Appenzeller 17:10 ... that added a feature somewhere, but this was-

Mike Volpi 17:12 Yeah

Guido Appenzeller 17:12 ... of a new product line-

Mike Volpi 17:13 Yeah

Guido Appenzeller 17:13 ... that in some cases became the leading product line in an industry.

Mike Volpi 17:16 Yeah. Now, you-you're a networking guy, so you'll appreciate this. The one thing that's a little different about a networking acquisition versus other ones is that networking operates on open protocols, which means that I could buy a switch company made by somebody else, and im- it immediately talks to my switches or routers, whatever, freely-

Guido Appenzeller 17:36 Yeah

Mike Volpi 17:36 ... because all the protocols are open. So, you know, at, at Cisco, we had a couple of proprietary protocols that we enjoyed, but by and large, all the-

Guido Appenzeller 17:43 [laughs] A few. A few. [laughs]

Mike Volpi 17:44 ... OSPF, B-BGP, ISIS, Ethernet switching, 802 dot whatever, they all worked. So you could immediately integrate the product into your sales team and have your sales team, like, literally you just change the badge on the thing. And so that allowed us to be, like, super high velocity. Okay, how does that contrast to, to, uh, Jensen's world today? F- one, they've made some really clever w- acquisitions. You know, like Mellanox is the world's largest-

Guido Appenzeller 18:13 Amazing

Mike Volpi 18:13 ... networking company.

Guido Appenzeller 18:13 That was amazing, yes. Yeah. It was-

Mike Volpi 18:14 Right? It's like-

Guido Appenzeller 18:15 Incredible

Mike Volpi 18:15 ... genius. Now, they're making other acquisitions that sometimes I scratch my head. I'm not entirely sure what Hugging Face is all about, for example. [laughs] Like, maybe there's a rationale Hard to me to see it. I don't think it matters. I think it's fine for them to do some of these and take risk, and some of them will work, and if they work as well as Mellanox, holy cow, and some of them are not gonna work. And I think the market has granted them a moment of incredible optionality, right? Ba-basically, they get to do whatever they want, which is the optimal moment for them to take these risks. So I think they're being very clever, and you can always critique any given acquisition. Trust me, I got critiqued endlessly at Cisco 'cause I bought Stratacom and now Cascade, and I, you know, bought an optical networking company from Vinod that went nowhere and wh-whatever. But in the end, and you're like, all right, well, I, I did, I did 70-- I think the, our team did 70 acquisitions when I was there, and those turned into half the company's revenue, and we were doing $30 billion in revenue, so it's $15 billion in revenue. Uh, the strategy worked.

Raghu Raghuram 19:25 Yeah.

Mike Volpi 19:25 And I, I think that, I think that, um, uh, the team at Nvidia is, is, i- is, is doing that. Um, I think that's interesting and wise. The degree to which they are financing the industry, we'll see.

Raghu Raghuram 19:41 Yeah.

Mike Volpi 19:41 Yeah.

Raghu Raghuram 19:42 Yeah.

Mike Volpi 19:42 'Cause that's different, right? That, that-- in that sense, Nvidia is basically, I think they're basically trying to shape what an industry looks like.

Raghu Raghuram 19:49 Yes.

Mike Volpi 19:50 And I think industries are powerful, and they find their own shape. It's hard to reshape them.

Raghu Raghuram 19:55 Look, let, let me, let me, let me defend-

Mike Volpi 19:57 But, but, uh-

Raghu Raghuram 19:57 ... Nvidia a little bit. H-having been at Intel and, and having seen how the big hyperscalers essentially disintermediated Intel from the end customers, right? And at the end, they could build better chips than Intel just because Intel had no idea anymore what the customer was doing with the chips, right? I, I, I, I have some sympathy for wanting to have more than three customers to sell to, or four maybe, huh?

Mike Volpi 20:18 As do I.

Raghu Raghuram 20:18 Yeah.

Mike Volpi 20:19 Yeah.

Raghu Raghuram 20:19 So, and you could argue, what is an efficient capital allocation strategy? Is it giving the ball back to investors, or could it be at this time, which is probably once in a time in history, put it all to work on future innovation, right? So.

Mike Volpi 20:35 Well, I think both of you make ve-very valid points. Um, the thing I would say, Guido, is that an industry wants to look a certain way-

Raghu Raghuram 20:44 Mm-hmm

Mike Volpi 20:45 ... because when it does, it creates maximum utility, maximum va- economic value. Why are there three hyperscalers? 'Cause it's a business where scale matters.

Raghu Raghuram 20:55 Software, yeah.

Mike Volpi 20:56 Right? It, it's a business where, like, s-scale matters, and therefore the industry wants to concentrate. The, the, no matter how much you finance them, the little ones are always weaklings. So, and obviously, if you're Nvidia, you don't like that. You don't want three customers. You want 50 customers. So you fund them, but at some point they're gonna die. It's just that because the industry, whenever it's a scale business, wants to consolidate. You know, we could fund 100 airlines in America, but in the end you're gonna have three. So I, I, I-- my point is not that the, um, not that it's a bad idea. It's just that it's futile.

Raghu Raghuram 21:34 Yeah. [laughing]

Mike Volpi 21:36 [laughing]

Raghu Raghuram 21:36 So, so, so noble intentions, but, uh, [chuckles] but, uh-

Mike Volpi 21:40 N-now to your point-

Raghu Raghuram 21:41 ... just a question

Mike Volpi 21:41 ... like, uh, it could be a very good return on... I don't know that they, they, I don't know if they actually care about the return on capital. They, they, they, they, they might. Um, and if they get out at the right time.

Raghu Raghuram 21:52 Yeah, it's not even about the return on capital. It's-

Mike Volpi 21:55 Efficient use of capital

Raghu Raghuram 21:55 ... I agree. It is, it is to some degree shaping industries, uh, I mean, shaping the s- deciding the shape of the industry. But on the other hand, it is, uh, enabling a lot of innovation as well because our startups and lots of other startups do get, get access to that capacity now.

Mike Volpi 22:15 Yeah.

Raghu Raghuram 22:15 Right?

Mike Volpi 22:15 Yeah, yeah.

Raghu Raghuram 22:16 Um-

Mike Volpi 22:17 No, no. Look, we're, uh, overall, I get why they're doing it-

Raghu Raghuram 22:20 New startups, yeah

Mike Volpi 22:20 ... and, and, and, you know, m-many of our companies are beneficiaries of it, so thank you, Jensen. [laughing]

Raghu Raghuram 22:25 Exactly.

Raghu Raghuram 22:27 So, um, I mean, at VMware we did probably not as many as you did, but, uh, maybe 50 p-plus acquisitions over the full lifetime of the company. And, uh, the observation was whether it didn't matter whether you were acquiring a 20-person company or a 2,000-person company, the amount of post-acquisition work in terms of integration, culture, business model, sales team, et cetera, et cetera, was al-always a lot of work. How did you guys pull it off at that time at Cisco?

Mike Volpi 23:04 Um, okay, structurally the way we did it was that we had a, a very small, when I say small, less than 10 people, uh, deal team that was job was to acquire, and we had a somewhat larger, uh, integration project management team that had, uh, sort of rules and guidelines about how we did integrations. And, uh, uh, th-they, they had colleagues, y- those people that project managed it, had colleagues in the various business units and divisions and functional units within Cisco that collaborated with them. So the important thing to recognize there is that, uh, com- businesses, companies are good at processes. Uh, they're very bad at one-time events. And so essentially if we were going to do a large number of acquisitions, we needed to have repeatable processes that were, uh, performed over and over again, and we had a centralized group that was the administrator of those processes. Um, more often than not, and in the beginning at Cisco we did this a lot, there was nobody that was f-formally assigned to this, so it'd be like, "Raghu, it's, it's, it's your job. You, you go have fun with that integration."

Raghu Raghuram 24:22 Yeah.

Mike Volpi 24:22 "Let me know when it's done." Right? And you're like, "Where, where do I even start?" [chuckles] You know? That was like the modality and we're like, "Okay, this is bad." So w-we, we created this process and it, and it worked pretty well. Um, where it got challenging is we liked to buy small things a-and small things became too small to become significant to the overall business, 'cause the business got so big-

Raghu Raghuram 24:46 Yeah

Mike Volpi 24:46 ... that like, you know, it just didn't make a difference anymore. And then when they tried to-- This was a little bit after my time 'cause I, I stopped doing the M&A stuff in 2000 and switched over to running a business unit. But, uh, after my time, they started buying bigger things, and then you get into the problem that you're talking about. Big things are very, very hard to integrate.

Raghu Raghuram 25:04 Yeah. Yep.

Mike Volpi 25:05 They have their-- They've been around a long time. They have a long culture. They like-- You know, we bought like Scientific Atlanta, which was a disaster. We bought like a bunch of things that didn't work very well. We also got into the mode of like, "Oh, we're not fully integrating them. We're like partially integrating them."

Raghu Raghuram 25:19 [laughs]

Mike Volpi 25:19 Yeah. Um-

Raghu Raghuram 25:20 It's terrible.

Mike Volpi 25:21 Yeah. It do- it doesn't work very well. So, um, so, so s- in other words, the mitigation is to have a process, but then when you stop respecting the process, you do have big problems

Mike Volpi 25:32 like you talked about.

Raghu Raghuram 25:32 Yeah. And what about from the portfolio company side? I mean, you got portfolio companies. We've got portfolio companies that, uh, at least in recent memory, we have had, uh, some exits into being other large companies buying them. Two questions there. One is, if you're a portfolio company CEO, what would you watch out for, right? And then sitting on the boards of these companies, how do you advise your portfolio companies these days about exits-

Mike Volpi 26:03 Yeah

Raghu Raghuram 26:03 ... about exits in general?

Mike Volpi 26:04 Um, look, uh, uh, you, y-y-y-when you're, when you're in our business of being on boards and you're facing a situation where you need to sell a b- uh, you know, there's an opportunity to sell a business, it's one of three outcomes. One, your company's doing great, and it's a really bad idea to try to sell it. So then you try to convince the CEO, "No, don't do it." Two, the company's struggling. We don't have a lot of finances. It's a nice way to get out. Everybody's happy. Then you just help the deal happen, and you sort of manage a process and, uh. The-- Then you have the tweeners. Uh, we are not totally sure, and, uh, I think that there are a couple of important indicators that help you determine what to do in that moment. The first is, do, does, does the CEO, the founder, the founding team have the fire? Like, do they wanna do this? Do, do they b- do they believe they can get through this sort of intermediary period and, uh, uh, bust open into the next phase? Do they, do they want it? Because if you have a CEO or a leader that's kinda like, "Man, you know, maybe I should consider this. I'm a little tired. It's been a long haul-"

Raghu Raghuram 27:20 Yeah

Mike Volpi 27:20 ... "then ch- let's just sell this thing and, and kind of move on." Um, and then the second factor is you have to be pretty cold and objective about like, is this gonna be a interesting enough business? Not like, will it double from here? Because at the end of the day, if it's double, I'll take the money today. But will it 10X from here? And that, I would say, gets to be a really hard decision because it's a little bit emotional also for you. Like you're, you know, you're not a founder, but y-you know, you've been working with this company for four, five, six years. Y-your heart's in it. You've, you've told your good friends to go work there. And, and yet you have to step, step away from it and go like objectively, uh, this is not 10X from here. And then you help steer the ship. But I think the most important parameter is where's the founder's head at?

Raghu Raghuram 28:13 Yeah. Yeah. And most-- So there is a certain element of mystery or unknown for a lot of, uh, the portfolio company founders on how do these things actually happen, right? People think one day somebody calls another company and say, "Hey, I wanna buy you," versus and then there's other thing where people have built relationships for a long time. How did you approach it at Cisco, and how do you advise your portfolio companies, even if they are expecting a 10X?

Mike Volpi 28:48 Yeah.

Raghu Raghuram 28:48 Right?

Guido Appenzeller 28:48 Do you just call them up, say, "Hi, I wanna buy you?" Is that it? [laughs]

Mike Volpi 28:51 No. Well, obviously you do as, as a good a job as you can building a network and knowing what's going on. I, I mean, it's basically cataloging the inventory is the way we used to think about it. But e-e-every company in the market is a potential acquisition, so you catalog them and, you know, where, where are they in their growth rate, what kind of founders they have, who are the backers, what are their aspirations, who are their customers. Just like be informed. So, and the, a-and the, you know, uh, the business, the corporate development group had to do that, but I kind of considered it everybody's job at the company to be constantly informed about what's happening around them. So, uh, and then e-essentially, uh, we, uh, at the company, we would try to weave in-- at Cisco, we would try to weave in the inventory against a strategy statement that was there. For example, we're a router company, but there are gonna be more switches than routers, therefore, we need to have a switch. What are the switching companies that are available, right? Or optical networking is becoming important because of bandwidth, blah, blah, blah. What are the companies that are available? Uh, you often have the o- the other way where companies come to you and says, "Hi, I'd like to get bought." I, I think broadly a bad idea, right? Because there's either an auction going on, uh, or you should have thought about it earlier if you or were interested in it. But basically, you almost never pay attention-- We never tried to pay attention to inbound. Uh, flip side, you're, you're now a small company or a startup-

Raghu Raghuram 30:23 Yeah

Mike Volpi 30:24 ... and how, how do you think about that? I always encourage always the founders of our companies to talk to big companies, to talk to competitors, and to always be in dialogue. Uh, some people will say like, "Well, you're shopping yourself." Uh, no, it's an inverse intelligence gathering mechanism. I even encourage my CEOs to talk to their direct competitors. Go to coffee. All right. Like, yeah, you th-- if, if you're clueless, you might leak some information towards them. [laughs]

Guido Appenzeller 30:51 [laughs]

Mike Volpi 30:51 Hopefully you're not clueless.

Guido Appenzeller 30:53 No.

Mike Volpi 30:53 And they might tell something to you that's interesting, or they might shape your view on an industry or something. So I encourage people to be out there always talking and seeing what's happening in the market. And, um, usually you kind of have a pretty good idea of like, okay, my company's struggling a little bit. I think I'd like to sell. I kind of already know then because I've been in conversations with who's where. So, uh, I would say, uh, uh, you know, don't s- a CEO of a startup shouldn't spend their entire day talking, but going on one or two coffees a week with, you know, the powers that be at Nvidia, the powers that be at Amazon and OpenAI, et cetera, et cetera, when the right people are talking, do it. It's fine. You, the information exchange, you don't lose that much in that. Th-they're already thinking about it. Uh, but you gain a lot by having that relationship.

Guido Appenzeller 31:46 I think this is really good advice. I, I, I keep telling my portfolio CEOs the same thing. If, if you don't have the cell phone number of the CEOs of each of your competitors, that's a problem. There will be a day when you need it.

Mike Volpi 31:55 Yeah.

Guido Appenzeller 31:55 Right, uh-

Mike Volpi 31:56 Absolutely

Guido Appenzeller 31:57 ... so you, it's, it's

Guido Appenzeller 31:58 a network you need to build.

Raghu Raghuram 31:59 Yeah, yeah. Now enough of this M&A stuff. Let's get back to actual technology. You are-- spent 30 plus years in infrastructure.

Guido Appenzeller 32:07 [laughs]

Raghu Raghuram 32:08 Very few people know more about infrastructure than what y-you do, what, uh, you've done in the past. Summarize the landscape for us right now. And when you think about infrastructure opportunities versus where is the technology gaps that you see and, um, what founders should be really looking at.

Mike Volpi 32:31 Well, super interesting topic. You know, I would say y-you wanna think about, I think, the current in, the shape of this industry in two dimensions, right? One is the vertical stack that starts with, you know, EUV equipment and goes all the way up to AI, applied AI and applications and everything that sits in between-

Raghu Raghuram 32:50 Yeah

Mike Volpi 32:50 ... that sandwich. And then you wanna think about particularly the infrastructure side. W-what's that, what, what's sort of the, the va- the value chain of it? Um, the first thing you would say is right now we live in a world where demand for AI capacity tokens, whatever, is near infinite and supply is incredibly constrained. So another way to put it is we all just live in Jensen's world. [laughs]

Guido Appenzeller 33:21 He's, uh, he's king of the hill. There's no question.

Mike Volpi 33:23 Um, um, uh, I-I think, uh, the shape of the industry will begin to change when either the de- the characteristics of demand change or the characteristics of supply change. And I think the, my, my view is right now the, the supply side, because so much of the limitation-- Well, there's, there's, there are some power limitations for sure, but the dominant limitation is silicon. And, um, I would say that unless there's more fab capacity, that doesn't change, which is, means it's not likely to really change in the next two or three years, maybe five years out. You could imagine-

Raghu Raghuram 34:04 Yeah

Mike Volpi 34:04 ... that there would be more fab capacity. You could put more wafers through the same machines, or you can make more ASML machines and so forth. But that's probably gov- a, a governing factor of our industry, which means we live in a reasonably scarce world, and those who have fab capacity are, are actually kind of kingmakers of, of the infrastructure-

Raghu Raghuram 34:24 Yep

Mike Volpi 34:24 ... AI universe.

Raghu Raghuram 34:25 Absolutely.

Mike Volpi 34:26 Uh, on the demand side, I, I do think that the world is gonna get smarter about the type of tokens they consume. I think we've been in a kind of an era over the last couple years that everybody just wanted the super high-end frontier token. And I think the topic of conversation today is model routers and how do I move stuff to, you know, the economic difference between a top-end Anthropic token and a top-end Kimi token is like five to one or seven to one or whatever. It's a massive economic difference, so people are gonna get, start smarter about consuming what. Uh, and, and I, I do think that that will begin to at least pr- put some price dependency on demand, and that's l- more likely to happen in the, in the, in the near term. So, uh, those are first of all, I think the dynamics that are gonna govern our industry and everybody that sits in between. You know, if you're a Neo cloud or you're an inference cloud or you're a model builder, you live between those two poles of supply and demand. Now for the infra itself, um, I, I think right now we still live in a world where you need to think of the entire piece of infra as a single system, right? So, uh, it's quite different, I think, than the networking days, for example, of old, where you would assume a very variable amount of traffic going from one to B, types of traffic, times of day, lots and lots of variability. Much easier to statistically multiplex that workload. Um, today basically you have, uh, lots of GPUs that are all working all the time, always connected with much more constant flows. And you, you, uh, because the interrelationship between the various components are largely still proprietary, you know, you're on NVLink or whatever, it's very hard to like swap one thing out for the other. And, and the, uh, uh, in fact, uh, today I would argue that probably the best architecture is the Google TPU architecture because it is literally thought of as a single system.

Raghu Raghuram 36:42 Mm-hmm.

Mike Volpi 36:43 And I, I think an Nvidia style architecture is migrating ... in the direction of where the TP- TPU architecture is, but it still has a, a, a few pieces that you can kinda swap out.

Guido Appenzeller 36:54 Yeah.

Mike Volpi 36:55 Um, and then you have everybody else who's like, "You know, I'm gonna take a Cerebras chip and throw an Arista switch on top of it," and, like, all that good stuff. And at least in today's world, those systems have inefficiencies, uh, because they're not constructed as a system. They're more of a patchwork. Um, uh, I, I think that industry will evolve, but today people that can think of the whole thing as a single system and optimize the architecture of it through that have a competitive advantage.

Guido Appenzeller 37:24 Yeah. I mean, yeah, it seems we are moving from, from a, from a card as the, the minimum unit to a rack as the minimum unit, or even, even a few racks in some cases now.

Raghu Raghuram 37:32 A few racks.

Mike Volpi 37:33 Yeah. I would-- That, that's true.

Raghu Raghuram 37:34 The NVIDIA part that they announced recently-

Mike Volpi 37:36 Yep

Raghu Raghuram 37:36 ... that is the minimum unit possibly.

Mike Volpi 37:39 Yeah. Well, but if you look at how Google does its thing with these, like, toroidal pass-through optical MEMS switches and everything, like, it's literally, like, the whole fucking data center-

Raghu Raghuram 37:48 Yeah

Mike Volpi 37:48 ... is, is the minimum unit.

Guido Appenzeller 37:49 So the row is the minimum unit [laughs] it seems like.

Mike Volpi 37:51 Yeah, exactly.

Raghu Raghuram 37:52 Yeah.

Mike Volpi 37:53 Yeah. But I, I-- my sense is that's where NVIDIA wants to go also.

Guido Appenzeller 37:56 Yeah, clearly.

Mike Volpi 37:57 And so the two dom- two dominant providers are all going to that systems approach

Mike Volpi 38:02 versus-

Guido Appenzeller 38:02 And I mean, even, even the smaller startups are, are heading that direction. But jump- jumping on that, how, how do you see, how do you see AI chips evolving? And, and right now, Jensen, king of the hill, incredibly dominant, right? Google TPU, very successful, um, currently mostly sold by Google. It's expected to change.

Raghu Raghuram 38:20 Not enough TPUs.

Guido Appenzeller 38:21 Uh, but not enough TPUs, exactly. Uh, how, how will the-- how's the future looking like? Will, will we see more, more competitive silicon?

Mike Volpi 38:28 I think for sure. Um-

Guido Appenzeller 38:30 We all want that.

Mike Volpi 38:31 Yeah. We all, we all want it. Um, uh, and we all wanna invest in it. But-

Guido Appenzeller 38:36 There we go

Mike Volpi 38:36 ... I, I would say, I would say, first of all, what I find interesting is that model architecture is becoming reasonably diverse.

Guido Appenzeller 38:44 Mm-hmm.

Mike Volpi 38:45 Right? So you have, uh, pe- some people using FP4, FP8. Uh, some people are using mixture of experts with lots of, lots of experts. Uh, some have fewer experts. Some are jamming up the parameter count, some not so much. And I actually think that there is a interrelationship between your chip design and the model that's optimized-

Raghu Raghuram 39:07 Yes

Mike Volpi 39:07 ... around it, right?

Guido Appenzeller 39:08 Yeah.

Mike Volpi 39:09 So-

Guido Appenzeller 39:09 Hardware-software co-design. Yeah, yeah

Mike Volpi 39:10 ... in, in some sense, to your point, like, is there going to be more diversity in chips? And I think yes, because there is gonna be more diversity in model designs. And depending on what kind of model you wanna do, uh, uh, what kind of things you want that model to do, you can certainly select a different chip architecture for that. Now, uh, reasoning intensity has dominated to date-

Guido Appenzeller 39:36 Yeah

Mike Volpi 39:36 ... uh, what chip architecture should be because, uh, reasoning intensity equals big parameter counts and big KV cache.

Guido Appenzeller 39:43 Yep.

Raghu Raghuram 39:43 Yep.

Mike Volpi 39:44 Uh, and that means that you need to talk a lot between memory and chip.

Guido Appenzeller 39:47 Yep.

Mike Volpi 39:48 Uh, and, you know, uh, when the only way you can talk between chips are the edges of it, it's just less connections-

Raghu Raghuram 39:56 Sure

Mike Volpi 39:56 ... than if you can use the top and bottom of the chip. So, uh, I'm a big fan of, uh, 3D DRAM architectures. Um, and, uh, some of the new stuff that's coming up with hybrid copper bonding is even more interesting 'cause you can have even higher density. Uh, so, uh, I think that architecture has huge promise, so long as we continue to see, uh, these KV cache, in particular KV cache intensive use cases. And it seems like the most value-added labor that occurs, labor or AI that occurs, is reasoning-driven labor, right?

Raghu Raghuram 40:34 Yeah.

Mike Volpi 40:34 So, uh, uh, and, and therefore, that should favor architectures that have very good memory bandwidth.

Guido Appenzeller 40:41 Yep.

Mike Volpi 40:41 Um, I, we're investors in Cerebras. Uh, Cerebras did very well to go fast, um, but if the model doesn't fit in the SRAM, y- you know-

Raghu Raghuram 40:53 You need a lot of them.

Mike Volpi 40:54 Yeah.

Raghu Raghuram 40:55 Yeah.

Mike Volpi 40:55 Well, yeah, you need a lot-- Well, y- because once you get off, off board of Cerebras-

Raghu Raghuram 40:59 Yeah

Mike Volpi 40:59 ... it is not gonna go super fast.

Raghu Raghuram 41:00 Yeah.

Mike Volpi 41:01 Right? And the, and also because the chip, the wafer is so large, the actual-- you're actually at a significant disadvantage in terms of the memory bandwidth once you go off chip.

Raghu Raghuram 41:09 Yeah.

Mike Volpi 41:10 So, um, now they, they run Sol on, on the Cerebras chip, uh, so that's a big model. So they've figured out how to fit it in, but my guess is OpenAI had to do some hard work to get that model to, to fit in the chip.

Raghu Raghuram 41:25 Yeah. No, we are fans of 3D architectures as well, but the manufacturability of it is a lot more complex with thousand-

Mike Volpi 41:33 For sure

Raghu Raghuram 41:33 ... thermal effects and-

Mike Volpi 41:34 Yeah, yeah

Raghu Raghuram 41:34 ... so on and so forth.

Mike Volpi 41:35 Yeah. No, I'm a big fan. I'm a m- mechanical engineer that studied thermodynamics, so-

Raghu Raghuram 41:39 There you go

Mike Volpi 41:40 ... um, heat is my friend. [laughing]

Raghu Raghuram 41:44 Yes. [laughing]

Guido Appenzeller 41:44 That makes you one of the very few people who would say that. [laughing]

Raghu Raghuram 41:47 Yeah. So I'll-

Mike Volpi 41:48 So we'll see how that package-

Raghu Raghuram 41:49 All the packaging divisions of TSMC and everybody else are calling you right now. [laughing]

Mike Volpi 41:54 Uh, yeah. Yeah, no, I agree. The packaging of those chips are gonna be, uh, complex, uh, and especially obviously the thermal attributes of it. And then when we get to a world where you have both a logic chip and a memory chip in the same package, the different, um, uh, coefficients of... Well, they're, they have the same coefficient of thermal expansion, but logic chips get a lot hotter. So, you know, the ability to put that all into the same package will be non-trivial, shall we say.

Guido Appenzeller 42:22 Chips are hot again, and they're getting hotter. [laughing]

Raghu Raghuram 42:24 Yes. The other possibility, of course, is to keep them still separate but have optical interconnectivity between them. Are you a believer in that?

Mike Volpi 42:32 Yeah. Um, uh, I think yes. Uh, optics, though, are very f- finicky things, um, that need to be, uh, experimented with a lot. So, you know- He- he-- w- when you heat up optics, they, they don't, they don't like it. They don't-

Guido Appenzeller 42:55 Sure

Mike Volpi 42:55 ... you know, photons don't like b- getting hot. And, um, I mean, in theory, you would, like, have a substrate that had waveguides in it or something, but the minute the logic chip starts getting hotter and you get sort of outside of the comfortable temperature range, you're gonna have a variety of connectivity issues and so forth. So I, I think-

Guido Appenzeller 43:11 [coughs]

Mike Volpi 43:11 ... optics are, uh, interesting. Um, w- if you can do it with copper, I'd do it with copper, is m- is my view. But if you, if you ha- if, if y- y- you-

Guido Appenzeller 43:24 Really?

Mike Volpi 43:25 Yeah.

Guido Appenzeller 43:27 Even for scale-up racks, not just interconnect now, but even for-

Mike Volpi 43:30 No, no. For, I mean, obviously for scale-up racks, that's a different story. That just simplifies the cabling and so forth. I'm talking about at the-

Guido Appenzeller 43:35 Inside the packet

Mike Volpi 43:35 ... at the modu-

Guido Appenzeller 43:36 Inside the server.

Mike Volpi 43:36 Yeah, yeah, yeah.

Guido Appenzeller 43:37 Yeah.

Mike Volpi 43:38 Yeah. No, for, for-

Guido Appenzeller 43:38 Inside the tray

Mike Volpi 43:39 ... when you start going top of rack and all that-

Guido Appenzeller 43:41 Yeah, you need it

Mike Volpi 43:41 ... de- definitely op... Yeah, no, no question. You just can't, you don't have enough bandwidth. Um...

Guido Appenzeller 43:45 So, so Cisco back then, most valuable company on the planet, early-

Mike Volpi 43:50 [laughs]

Guido Appenzeller 43:50 ... early hardware leader in networking, which sort of powered, powered all of the internet, at some point wasn't most valuable company on the planet anymore. What, what, what triggered that? What can we learn from that?

Mike Volpi 44:01 W- what tends to happen when you have a bubble in demand, in perceived demand, is that, uh, customers over-order because they know they're gonna get cut back.

Guido Appenzeller 44:12 Yep.

Mike Volpi 44:13 And so w- w- uh, every part of our product line wa- was out of capacity. We were, we couldn't, we didn't have, we couldn't make enough switches. We couldn't make, we had orders on top of orders on top of orders-

Guido Appenzeller 44:24 Yep

Mike Volpi 44:24 ... on top of orders. And then what, uh, uh, we were, of course, putting orders on all of our supply chain, and because the demand felt so robust, we were taking the financial risk on those orders. In other words, we were like, "Here's the money. Keep the money-

Guido Appenzeller 44:39 Yeah

Mike Volpi 44:39 ... and just get, get us this stuff as soon as possible." So-

Guido Appenzeller 44:41 What could possibly go wrong with that? Yeah.

Mike Volpi 44:43 Yeah. So then-

Guido Appenzeller 44:44 [coughs] Hmm

Mike Volpi 44:44 ... when demand dries up, it's not like people stop buying. They cancel all their orders.

Guido Appenzeller 44:50 Yeah.

Mike Volpi 44:50 They're like, "Yeah, I never, I ordered twice as much as I wanted to anyway, so I, I, I need half an, half of half of that." So the orders went to, like, one quarter of what they, what, what they were. And then you, you, A, you've taken the risk on the supply chain, and B, your demand falls through the roof. And so I think we ended up in the December quarter, and in particular, this is kind of memorable because it w- happened in December quarter, so we are one month off, so, uh, November, December, January. And usually what tends to happen with businesses that are a month off, you, you know this, right?

Guido Appenzeller 45:22 Yeah.

Mike Volpi 45:22 Like December's soft, and then the first week of January-

Guido Appenzeller 45:25 All the things get sorted

Mike Volpi 45:26 ... you start seeing it coming back, and then by the end of January, you've made it up. So we go through December, it's soft, and it's like no big deal. December is always soft. And then week one January, nothing happens. Week two January, nothing happens.

Guido Appenzeller 45:35 [laughs]

Mike Volpi 45:35 Week three January, nothing happens. You're like-

Guido Appenzeller 45:37 Oh, shit [laughs]

Mike Volpi 45:37 ... "Oh, shit." So the quarter came in, I don't know, maybe we missed by 35, 40% or something.

Guido Appenzeller 45:45 Wow.

Mike Volpi 45:46 Yeah.

Guido Appenzeller 45:46 Mm-hmm.

Mike Volpi 45:47 A massive mi- For a company, by the way, that pu- publicly traded for a decade, never missed a quarter.

Guido Appenzeller 45:52 Yeah.

Mike Volpi 45:53 Uh, and, uh, yeah, and then that's how it all corrected. And it was, uh, it was difficult. Stock went from 80-some dollars a share to $9 a share, I wanna say. So it was, like, literally one tenth the value over a period of six months.

Mike Volpi 46:08 Very tough.

Guido Appenzeller 46:08 Yeah. So, so look, e- every time when the workload changes enough at the top layer, the, the underlying infrastructure, software, and hardware needs to change with it, right? With AI, with AI, we've seen this. We talked about compute, right? What, what's gonna happen in networking? I mean, we've, we've seen big changes in networking. We no longer have one network. We now have suddenly the second backup network. Where did this come from?

Mike Volpi 46:32 Yeah.

Guido Appenzeller 46:32 Did you just say internet is dead?

Mike Volpi 46:34 Uh, there we go. Internet is dead.

Guido Appenzeller 46:36 No, no, no. Perfect.

Mike Volpi 46:36 No, no, no. I mean, ethernet is a framing mechanism fundamentally, so-

Guido Appenzeller 46:39 Well, is it, it's coming back. It was dead, right? [laughs]

Mike Volpi 46:42 Here, here I think, I think the, the core tenets, the world of telecom started with circuit switching, right?

Guido Appenzeller 46:49 Mm-hmm.

Mike Volpi 46:50 Like we made the phone, and the idea with the phone is that you had starting point A, starting point B, constant bandwidth between the two. You connect the line, and off it goes. And then we found out, well, this is, like, uh, reasonably unreliable 'cause one line cut and you lose it. Uh, and obviously it doesn't scale very well in a world where you have uneven traffic patterns. So then we decided let's go to packets, and packets are very convenient because they allow stat muxing-

Guido Appenzeller 47:16 Yep

Mike Volpi 47:16 ... uh, which is the big, big-

Guido Appenzeller 47:18 Yep

Mike Volpi 47:18 ... benefit of it.

Guido Appenzeller 47:19 David and Andy back then, right?

Mike Volpi 47:21 Yeah.

Guido Appenzeller 47:21 They were cheerleading it, yeah.

Mike Volpi 47:22 Yep. So, so, uh, that and, uh, you know, variable length packets and all of these, you know, IP packets essentially. So essentially the world grew up building networking equipment, assuming that you had uneven arrivals and departures into a device that spread out the workload, and you would underprovision the device to connect more bandwidth because not everybody all communicates at the same time through the thing, and the outputs of that inbound are highly variable. Okay, that w- that was, that's the internet. That's the workloads that we live with, blah, blah, blah.

Guido Appenzeller 47:57 That was almost dogma, right-

Mike Volpi 47:58 Yeah, yeah

Guido Appenzeller 47:58 ... for, for a long time.

Mike Volpi 47:59 Okay. So fast-forward today. What does an AI data center look like? It does not look like that. You don't have random arrival. Um, you, it, links are consistently fully utilized. GPUs wanna be fully utilized. Traffic patterns are actually much more predictable than they ever were before, particularly in training, they're incredibly predictable. They're less predictable in inference, but you can see the traffic pat- especially if you do, like, big batching.

Guido Appenzeller 48:23 Good, good chunk of it is still very predictable, right?

Mike Volpi 48:25 Yeah, especially if you do, like, big batching.

Guido Appenzeller 48:27 Yeah, exactly. Yeah.

Mike Volpi 48:27 Like you, you can definitely make it more predictable. So the fundamental tenet of the, the, the workload has changed. Yet, if you look at the dominant networking architecture today, which is the Broadcom, uh, uh, what's it? Not Trident.

Guido Appenzeller 48:43 The Tomahawk.

Mike Volpi 48:43 Tomahawk.

Guido Appenzeller 48:44 Yeah.

Mike Volpi 48:44 Uh, which is the Dune acquisition that they did, is a stat muxed packet switching architecture.

Guido Appenzeller 48:49 Yes.

Mike Volpi 48:50 So I think that at some point For data center networking, we're going to have a, a, a new kind of architecture that assumes a high degree of predictability and a full utilization as opposed to an unpredictable, a stochastic statmuxed approach. Um-

Guido Appenzeller 49:09 So essentially back to circuits is what you're saying.

Mike Volpi 49:11 Yeah.

Raghu Raghuram 49:11 And, and optical circuits with this.

Mike Volpi 49:14 Y- yeah. So and then I-

Guido Appenzeller 49:15 Or, or, I mean, time division-

Raghu Raghuram 49:16 Or, yeah

Guido Appenzeller 49:16 ... multiplexing works as well, right?

Mike Volpi 49:18 Or combinations of those.

Guido Appenzeller 49:20 Yeah.

Mike Volpi 49:20 You, you know, you can have-

Guido Appenzeller 49:21 Most likely

Mike Volpi 49:21 ... you can have many modalities of multiplexing, but it, they're, they're not stochastic, right? They're, they're, uh, uh, constantly timed-

Guido Appenzeller 49:29 Yeah

Mike Volpi 49:29 ... somehow. Like, you can do TDM, you can do, uh, whatever else you want, but basically... So, and I think these architectures, the, the, the issue is the switches you buy today have the correct amount of bandwidth and the incorrect architecture, not incorrect, but, uh, less adapted architecture. Uh, and I think the data center switches of the future at some point will look more like that, will look more, uh, like, you know, potentially you could do it with MEMS.

Guido Appenzeller 50:00 So we bring ATM back into the data center. [laughs]

Mike Volpi 50:02 But even ATM had statmuxed stochastic cells, right?

Raghu Raghuram 50:05 Yeah, true.

Mike Volpi 50:06 It was straight up circuits.

Guido Appenzeller 50:07 Yeah.

Mike Volpi 50:08 Um, so now networking vendors, including my former employer, this is bad news, because when things don't change, you can't charge a lot of money for them, right?

Raghu Raghuram 50:17 [laughs]

Mike Volpi 50:17 Like, the whole point of a switch is that it's a complicated animal to have a, a stochastically, statistically multiplexed traffic flow. It's a hard thing to have, ah, one packet came in here, I have to throw it into memory, I have to do a lookup, do the lookup-

Guido Appenzeller 50:30 Yeah

Mike Volpi 50:30 ... I gotta send it out of this gate. That's, like, hard to do, so you make the thing expensive. But in a world where it's relatively static, you don't need that much intelligence.

Guido Appenzeller 50:37 Don't need buffers anymore.

Mike Volpi 50:38 Bam, send it right through.

Raghu Raghuram 50:39 Yeah.

Mike Volpi 50:40 Right? So I think that there's an opportunity for somebody to, to sort of rethink network architecture given this workload and make it way cheaper. It's like classic innovator's dilemma, because no matter what vendor you are today, this is not a good trend. You're not gonna wanna do that. But if a newcomer, and you say, "You know what? Forget all that statmuxing shit. Like, the world's different. I'm gonna go with my thing," actually you might be able to build something interesting. Um, now, it could be optical, it could be electrical, it could be a combination of the two. Um, but I think we're gonna, what will emerge... And I think you can see clues of that from how Google does its-

Raghu Raghuram 51:23 Yeah

Mike Volpi 51:23 ... network.

Guido Appenzeller 51:23 Yeah.

Raghu Raghuram 51:24 Yep. Yep. That's, that's-

Guido Appenzeller 51:25 I mean-

Raghu Raghuram 51:25 ... led them to that architecture

Guido Appenzeller 51:26 ... I think Amin has, has been the pioneer of specifically opt- optical circuit switching, right?

Mike Volpi 51:31 Yeah.

Guido Appenzeller 51:31 As, um, they were the first ones. And-

Mike Volpi 51:33 That's for, like, the toroidal thingamajig.

Guido Appenzeller 51:35 Yeah, yeah. And we're starting to see that gain traction.

Mike Volpi 51:36 I don't understand why it's so slow, by the way. It takes them, like, a minute to switch. So that one I don't totally understand.

Guido Appenzeller 51:41 Uh, I think to some degree it's because of the use case that right now your users are set up fairly static circuits.

Raghu Raghuram 51:46 I mean, when it is static, why you care?

Guido Appenzeller 51:48 Fair. I, I would-

Raghu Raghuram 51:49 I mean, for availability reasons you would want to

Mike Volpi 51:51 ... but still, I know, but give me a few milliseconds, not a second.

Raghu Raghuram 51:53 [laughs]

Guido Appenzeller 51:54 Yeah, which you probably could, honestly, right? I mean, the technology, I think, is there.

Raghu Raghuram 51:57 Yes.

Guido Appenzeller 51:57 It's just the use case is not there.

Raghu Raghuram 51:59 Yeah.

Mike Volpi 52:00 Yeah. Yeah, so yeah, no, I've been looking at a bunch of these, whether it's MEMS or, uh, these, uh, Max Zander thingies, the MZIs and whatnot, to see if there's something in there. But, you know, somebody has to emerge with an architecture. And then-

Guido Appenzeller 52:12 Yeah

Mike Volpi 52:12 ... the, the issue is gonna be who embraces it, because-

Guido Appenzeller 52:17 Yeah

Mike Volpi 52:18 ... you can't just build a switch without the rest of the system coming together.

Raghu Raghuram 52:22 Yeah. The-

Guido Appenzeller 52:23 I mean, that's the other thing that-

Raghu Raghuram 52:23 ... the innovator's dilemma will extends to the user's inertia as well.

Mike Volpi 52:29 Exactly.

Raghu Raghuram 52:30 Right.

Guido Appenzeller 52:30 Well, and this what, the network to some degree, coming back to the, you know, ship me a row concept of infrastructure today. The network-

Raghu Raghuram 52:37 Yeah, I think it's a-

Guido Appenzeller 52:37 ... has become an integral component of, of your architecture. So this innovation to some degree almost has to come out of the-

Raghu Raghuram 52:43 It's a component

Guido Appenzeller 52:44 ... the GPU vendors.

Raghu Raghuram 52:45 Yeah.

Mike Volpi 52:46 Yeah. Yeah, but it's interesting, right? Because it, there's on- only Nvidia, maybe AMD will get there at some point, but only Nvidia and Google right now have the ability to ship you, uh, a, a, an, a rack system.

Guido Appenzeller 52:59 That's right, yeah.

Mike Volpi 53:00 So today they would have to adopt it. Uh, it'd be interesting to see, like, you know, as more chip vendors evolve, i- is there, uh, you know, is there an alternative architecture that they can all agree on? Um, sort of the shit that, you know, we did.

Guido Appenzeller 53:19 Yeah. [laughs] Back in the days.

Mike Volpi 53:20 Back in the day, right? So, like, if, the thing is, like, you have two dominant vendors.

Guido Appenzeller 53:24 The ITF 2.0.

Mike Volpi 53:25 Yeah.

Raghu Raghuram 53:25 Yes.

Mike Volpi 53:25 Well, they, they, they, you have two vendors. They're dominant. Uh, they, they have their own approach of doing things. They're not likely to shift off of that.

Guido Appenzeller 53:32 Yeah.

Mike Volpi 53:32 And can you have a, a, a collection of everybody else that says, "No, actually, this is a standardized network architecture that we're all gonna agree upon, and that's what we're gonna build against"?

Guido Appenzeller 53:42 The Rebel Alliance. I like it.

Raghu Raghuram 53:43 Yeah, the buyers have to agree. That's the key.

Mike Volpi 53:46 Yeah.

Guido Appenzeller 53:46 Yeah. So, so Mike, y- y- you've seen a lot of these cycles now, some bigger, some larger. You've seen the last really big one. H- how, how does this compare?

Mike Volpi 53:55 As you probably know, I've, I've started a new venture fund called Hanabi Capital. Technically speaking, I retired in 2023, and I was supposed to get a yacht and, and play golf.

Guido Appenzeller 54:04 That didn't work out very well, did it? [laughs]

Raghu Raghuram 54:06 No.

Mike Volpi 54:06 Why? I, I actually think this is the most fascinating time to be doing our job and to be in our industry, not just because the change is big, and we all know that, you know, how transformative the change is, but because it requires more, uh, intellectual exertion to understand how, w- what next. It's like you don't have a crystal ball. It's not obvious what the next move is.

Guido Appenzeller 54:34 Yeah, it's hard.

Mike Volpi 54:35 And so you have to keep thinking, and you have to think at a systems level. Like, I don't think I've ever been in an industry before that I had to think about how light gets concentrated on a Zeiss mirror to figure out whether my chatbot works well. And, like-

Raghu Raghuram 54:50 [laughs]

Mike Volpi 54:50 ... but it's all, it's this amazing-

Raghu Raghuram 54:53 Yeah

Mike Volpi 54:53 ... interconnection of things, where there's bottlenecks and this changes and this change, that change. So I find it just the most intellectually fascinating time. And because the end prize is so significant, by figuring out how these different parts unlock, you can create a company here and a company there, and this one can become worth a trillion dollars, and this one can become 500 billion and so on and so forth. And I honestly, I did not expect to encounter another time like this in my career, and by far I fi- I find this the most dynamic and interesting time to be doing what we're doing, and I think we should all feel quite fortunate about it.

Guido Appenzeller 55:31 Amazing.

Raghu Raghuram 55:32 Yeah.

Guido Appenzeller 55:32 This, this tops '99.

Mike Volpi 55:33 Oh, yeah.

Raghu Raghuram 55:34 [laughs]

Guido Appenzeller 55:35 Easily.

Mike Volpi 55:35 Yeah.

Raghu Raghuram 55:35 Totally, totally agree. And we are all in the same business for very much the same reason. This has been a fascinating hour. Thank you so much, and we could be continuing this conversation endlessly, but perhaps we'll do it another time.

Mike Volpi 55:48 This was super fun. Thank you, guys.

Guido Appenzeller 55:49 Thanks for coming, Mike.

Raghu Raghuram 55:50 Thanks.

Guido Appenzeller 55:50 We appreciate it.