The Infra Pod - Ditch the cloud for bare metal in 2023? Chat with Zach (Equinix, ex-Packet)

Episode Date: June 20, 2023

2023 is when the cloud should have 100% adoption right? However, we're seeing a surge of embracing your own bare metal in 2023 from companies all over the world. Ian and Tim sat down with Zachery Smith (Head of Global edge at Equinix, ex-CEO of Packet) about his journey of starting Packet and the customer trends he's seeing from Equinix in 2023.

so welcome to our new podcast this is ian and tim just a quick intro between all of us i am our little vc funk essence vc and ian done so much awesome things oh maybe you should introduce yourself quickly thank you yeah i've uh started two companies cake privacy and manifold as a co-founder currently i'm helping sneak figure out their platform strategy and do a little angel investing on the side as well dabble around i around. I'm hyper excited. On the down low, right, Ian? On the down low. Yeah. And hyper excited to be joined by Zach Smith and learn all about you and get some perspective on bare metal. Tell us all about it. Yeah. Super excited, Zach. So maybe do a quick intro about your background and we'll quickly get into the

Packet story really quickly. Sure. I'm Zach Smith, based here in New York City. I guess I've lived more than half my life in the fine city, so now they'll officially allow me to call myself a New Yorker. I grew up in Southern California, dropped out of high school when I was 17 and came to New York City to play classical double bass. Thought that was going to be the career. And it turns out, no, Linux hosting was going to be the career. My mom was a little bit like, what are you doing over there in New York? But I can't say anything bad about being in the right industry at the right time. Super awesome entrance into the world of open source, which was in 2001. I had learned a little bit about Linux from a guy named Miles Cowan. He was at the Juilliard Computer Lab with a bunch of Mandrake CDs or something like that, just installing some

weirdo operating system on one of the five PCs we had. And in 2001, after I was done with music school, I kind of landed in the world of hosting. I was going to start a business offering web hosting for musicians. And I went online to webhostingtalk.com and posted that I was interested in a Linux server. And a guy, you know, by the name of Rajdha messaged me back. And that started a little bit of a relationship that continues to this day. We ended up joining him in a company he was running called Voxel, which is a really Linux-based, Debian-focused Linux hoster company. We wrote a bunch of open-source software and helped us to manage that and support that growing business.

And we sold it in 2011 to a public firm called Internapp. And the reason why we did that was because we really felt that public cloud was going to eat the whole thing. So right in, it's going to eat the whole thing. The reason why is they were doing data centers through databases, and we're just putting servers in a data center and trying to turn them on and off. A couple years later, a friend, a neighbor of mine, Alex Polvey over at CoreOS, and just basically saying like, no, no, no, no. Actually, we can move workload wherever you want. And I was like, wait, tell me more. Software would eat the world.

Well, why wouldn't software eat the cloud? And so we decided to start a company called Packet, which was a way to provide very fundamental infrastructure automation that would introduce the lowest levels of infrastructure to a generation of developers who a generation of developers who needed kind of repeatable APIs. And if any of you have been in a data center, it's anything but a repeatable process, at least at that level. And so, yeah, we started on that.

And my brother and I and another founder, a guy by the name of Aaron Welch, started that business in 2014. We didn't die. We raised some capital, ended up really focusing in on this niche called bare metal at the time, which is how to make any computer, no matter what it was, where it was, or what you were going to run on it, be repeatable and accessible to developers. Ended up selling that business to Equinix, where I am now in 2020. Equinix is the world's largest operator of data centers and interconnection. And so we're kind of taking our skill set of turning on and off computers inside the data center and hopefully making that more accessible to thousands of companies and users. Yeah. This is actually the main topic and why we

actually wanted to have you on. Very quickly, we want to talk about this whole, why would people care about bare metal in 2023, right? Back in the packet days, at that time you started packet, what was the things you are trying to say to the developers? Like, hey, come to me for once. What are the things are true back then? And what are the things you're hoping that will happen so that everybody will start to use packets? Yeah, we had a pretty focused mission. And the dream that we had was how could we put an API to hardware?

Hardware in the data center to us meant no matter what it was, I said that kind of very quickly in passing. But I guess what I felt is that for most people, compute and related was a total commodity, right? A virtual machine is a machine, you know, thing is whatever. Usually when I talk about this, I like then I hold up my iPhone and I was like, except for if what you're trying to do is really special. And then the hardware matters a ton as I speak on my Mac with an M2 with a super JavaScript code processor that makes my battery last forever, right? And so it was basically the idea that hardware was completely irrelevant

to like 99% of all developers, but to 1% of the developers who were trying to change the way that the world worked, trying to change the whole experience of how humans interacted with technology, it mattered a ton. And those abstraction layers were big in things like public cloud, where they chose most of that hardware opinion for you. It's super amazing for almost everybody. But then there was this like select group of workload or use cases or ideas that really needed access to technology in a more direct way. And there was really no way for you to do that without kind of building everything yourself, like literally going into the data center and getting co-located and racking and facking and learning how to build a backbone

and doing all this stuff just to create some sort of infrastructure automation, just so you could turn on the computer, just so you could put an operating system and start to work on it. And we said that that feels way too much. And I was inspired by an NPR podcast, Planet Money, if you've ever listened to it. And There was this show in 2015 or something about satellites in space and the origin of putting satellites up in the terrestrials. It was all about creating a standard that made it easy to ship this technology up into the skies. The reality was putting a satellite into space would be easier and faster for most companies

than putting a server in Hong Kong. And I was like, well, that's crazy. So we said, hey, how can we create this lowest level of automation for people who really wanted access to diverse and interesting hardware, but wanted to do so with repeatable constructs? So we focused on x86. AMD wasn't really back. So it was Intel and ARM. And we said, how can we make ARM a first class citizen? That was kind of our really first journey where vast majority of things just didn't even boot. There's no CI systems, there's no automation. So we just said, hey, why don't we provide hardware automation no matter what it was, an experience? And you know what? That turned out to be very interesting for a

subset. I use a banking analogy. We were always very focused on, hey, if AWS or Ben was building DigitalOcean is like the checking account of internet infrastructure. We wanted to be the wealth management relationship of internet infrastructure. How could we give that powerful access for changing the world to a generation that didn't want a bad mobile app? They wanted a good experience

for accessing their super powerful things. So yeah, that's what we did. Amazing. I love the analogy of wealth management and checking. I think that's great. The story around ARM, I'd love to learn more about how much work you had to do there. But it'd be great to understand this basic idea that you had,

which is at the time, these big clouds offered you a bunch of generic hardware, hyper-abstracted, you're paying huge performance costs. What was like from a buyer perspective, was it mostly like you enabled new hardware? Or was it also like there were costs a part of it? Was it like points of presence or availability, the automation? Who were your initial customers and users? And what was it that got them so excited about what you were doing? So I think it really came down to attracting a developer focused user who, you know, is looking for automation that was a non-negotiable. I must

be able to use cloud on it. I got to be able to like use an API that returns me a response. I got to have that. That's just part of how I build, run, innovate, et cetera. That was like a baseline. And then what we looked for is like, Hey, listen, there's going to be customers who want more control. And part of that control meant I want to innovate on the operating system. So our first really killer use case, the one that totally blew up in 2015, was actually Cisco. We started to get all these signups. And I was like, what's going on?

We were very early. We had a couple points of presence. We had these x86 and ARM nodes that we had gotten from Foxconn to turn on. And we had a few flavors. And then basically we started to get like hundreds of signups a week from Cisco. I'm like, what the hell is happening? Turns out what they were doing was there was this project that enabled network teams to create a virtualized Cisco device, a set of virtualized networks to try their network

virtually. And it was based on OpenStack. And they needed to do their own distribution and nested virtualization in order to turn up all these things. And like suddenly, they just wanted more control over tenancy. But somebody at Cisco had basically written some terraform to allow you to create your Cisco lab in a couple minutes. And that was like not possible beforehand. And so suddenly thousands of Cisco CCNA people and network people were signing up to access hardware by the hour, by the minute to do that.

Went from that into other people who really valued control. This is probably relevant to my role and job here at Equinix, but it was people who normally end up having to move into co-location because they had network requirements. They didn't want an overlay. They wanted to innovate down on BGP. They wanted to work on their own abstraction layers, et cetera, or hardware opinion that wouldn't be possible. And so machine learning

started to come out. ARM, obviously, with Android and other parts of the mobile ecosystem, places like operators. One of our big customers was Sprint, building an IoT network for LTE traffic, things like that, which were just not possible with the current abstractions. They just had more opinion. So I'd drive it down to that. You know, it's a very fundamental business being in the internet infrastructure world. Like, what can you do with a computer in a data center? Almost anything.

What we really ended up attracting was a subset of users who just had a high amount of opinion. One of the things is like, how far has this come? That's like a great beginning story. But where we are today in terms of like the bare metal racking and stacking automation story, or like obviously you're at Equinix, which is, you know, an at scale business. I think you have many points of presence, many data centers, huge amounts of bandwidth. Love to really understand how that's evolved over the time as your time is packet through the acquisition into where we are today.

First and foremost, Equinix is the world's largest operating data centers. I think we have over 250 data centers in 40 plus countries. But more importantly, we have 400,000 interconnections and 3000 networks that sit at our facilities. So that density of kind of network means that our facilities, our metro locations are kind of the center of how you reach everything from wireless networks to terrestrial networks like MPLS to all the clouds, right? So if you want to sit in between all the stuff that matters in London, you're probably going to end up at an Equinix data center. So that kind of network density and access is the unique aspect of Equinix's footprint, but it's really big. We have data centers in Mumbai, and then we just announced in Johannesburg and Lima, Peru. It is

hard to go and put your stuff in all those data centers. That was recognized several years ago by the Equinix leadership team, Charles Myers, our CEO, around, hey, we have this platform, which is so powerful and enables so much of the internet to work and so much of the cloud to work. And yet the friction point is too high. So how do we remove the friction while still being true to our ecosystems? Equinix stands for equality and neutrality in the internet exchange. And so our model is very neutral, right? So how do we build a easier, less frictionful way to access our global reach and access those networks and ecosystems without effectively competing with our ecosystems? We basically decided, hey,

let's remove the toil from putting the right thing into the right place and doing things like, how are we going to manage sustainability? Could we get it there efficiently? Could we recycle it? Could we operate it with the most energy efficiency? How could we do it regularly, like within a few weeks or within a few days? How could you get the Dell computer that you wanted to get into the right location and have an API to actually query it every few hours or days? How could we take that and make it more accessible? Still not for most people. Still for mainly people who are really looking for an opinion. It's not a generic platform for just anything you want. And so I would say that we've made

that a lot easier. Within a few minutes, you can access fundamental global reach in thousands of networks across the world. That's really cool. I think where it's going next, though, is fascinating because we're just starting to hit the tipping point of what I always love about Equinix's model and cloud in general is that our technology world is pretty addicted to a business model, I think, that is really not sustainable. And so when you look at it, I love Intel, but they don't make any money unless they sell you more chips. I love Dell. Michael was a great investor in my last business in Packet. But like, they don't make a lot of money unless they sell you more computers.

And even Equinix doesn't make a lot of money unless we sell you more space. And I think we're going to move to a world that's more outcome based. And we're gonna have to do it in a much more sustainable way. So I love the transition that we're seeing and supporting our business model at Equinix to bring partners like Dell and HP and Cisco to an as-a-service distribution model instead of sell you a server, like use one. Or the work we've done with NVIDIA with AI Launchpad, which is like, don't hawk you another GPU, but use it for what you need to. And then we will recycle it or send it to somebody else when we want to like license it. That's where I think we got to move to be a line. And that's where I think the really exciting stuff will happen. Because I believe that is the friction point that says,

why can't your new AI application be 1000 times more energy efficient? And the answer is because you don't have the right silicon with the right software. And it's too hard. You have to do like massive scale and get it like to be the bell curve use case to put enough things in one place because it's so hard and expensive. But what if that wasn't the case? What if it was like 10 bucks of shipping in three days to have the right thing in the right place times the right number of things. And then when we were done with it, it got used by somebody else or recycled and turned into something else. That would be cool. Then you would see software, which innovates pretty quickly, and hardware, which actually is kind of full of friction, move together at a much more interesting pace that could really solve

some big problems in our world. So that's kind of where I'm interested is the next phase of, hey, can we change the way and improve the way that hardware technology and software work together? So that's not just a few at-scale companies that can innovate on generic use cases, but how can thousands of companies innovate on use cases that are maybe more specific to them? Yeah, there's so many questions. We think we can go after this. So this is really fascinating.

I think fast forward to now, we're going into this direction now where this is what we see as engineers, not even knowing what's going on too much, even on the infra down beer metal world is Amazon selling the cloud, saying everybody is jumping on a cloud. Like every report is saying cloud is going to eat everything. Right. And that's kind of like the norm now.

Like it's not just single cloud, multi-cloud. There's this like this public cloud will eat all mentality in some level of belief. But we're also seeing startups and other providers and individual companies too. We're very curious about what the trend will continue down. Is, hey, cost is just way too outrageous. We got to go and control ourselves. So let's go buy hardware, right? Let's go down the old path.

And there's more and more software, like you said. Maybe the real question here is like, what do you think is actually happening now? Do you see your buyers no longer just become Cisco's and NVIDIA's of the world? Like we're seeing more and more enterprise and smaller companies all starting to try to build their own stack for cost or the cloud is just going to eat all?

What is the stance now from your lens? Is things are changing or not yeah well i mean you certainly have a great normalizer which is that software is becoming increasingly portable and even the really really hard stuff like databases are now i'm not going to call it trivial but even we i want to have a big info team i love my my delivery engineering team. I mean, we're operating 60 global Kubernetes clusters, and we're running Cockroach and a lot of them. And we have a pretty amazing distributed data store that like five years ago, it would have never happened.

Like it's now much more possible. I'm not saying that's like every pizza shop on the corner wants to do that or can do that, but it is way easier and way more possible today with distributed systems and I think open software than it was even just three or four years ago. And I think that is a great normalizer in giving choice to customers, which is where is the best place to run my workload versus it is so impossibly hard to run my workload anywhere else that I need to go only run it here. Or it's so hard to move, too expensive to move that I can only run it here. Or it's so hard to move, too expensive to move, that I can only run it here. That can go anywhere from a legacy data center,

fitting like on my AS400 to a public cloud where you put all your data and it's super hard because you tied yourself into three or four systems or whatever, right? And so I think that there's just been a ton of work through portable software. I mean, whether that's open software like Kubernetes, or commercial software like even VMware, that has just made that a lot more egalitarian. And I think that's awesome. So you're going to see a lot of movement and change, I think, as software starts to find the place where it lives best.

And that's not a static thing. That's because people are constantly reinventing and moving that along. That's one thing. The second would be that the infra model is changing. We are making accessible OEMs have as a service models. So do clouds. So do data center operators like Equinix, which we would all be like, really? Equinix does that, right? So I think that's great. Even silicon partners like NVIDIA and whatnot coming out with direct models. That's good, right? I think that provides more interesting

access and models for people to be able to consume and use what makes sense for their business or their application or what they're trying to optimize for. And then I think to like drive to the heart of your question, like, are people actually doing that? I have to go from two different angles. I'm 43. So like I skipped millennial by one year. But like, you know, I think there's a large portion of companies and developers and system operators who have just never done anything about public cloud. And they're like, what do you mean? What is this dark magic that I can reach all the places on, you know, crazy networks that don't charge me per gigabyte? Just tell me more. And so I think things like interconnection and network and storage that sits in between that and super crazy, interesting new hardware is starting to

become an accessible thing for people who've like never had that before. They've had layers of abstraction and a couple of choices from a catalog. They couldn't just like dream up what they wanted to have. So I think that's starting to become a reality for one side. And the other side is actually just still in a giant digital transformation movement. Most large enterprise customers we have are literally just trying to move their networks out of their own private built data centers and stop using MPLS and start using SD-WAN. There's two different ways to look at this. And we see both sides of that coin. The way I've kind of put it down on my little shorthand notebook is that companies whose business is about digital, and they're in the business of creating ecosystems,

not consuming them, are generally going to move toward the place where they have more and more opinion. They need it because their business is built on it. I think that's going to become more opinionated. And I think we'll see a lot of different models around how that occurs. But people will build more infra because A, the industry will make it easier and give them models where they don't have to like amortize hardware for nine years and hire an entire network team and build global backbone just so they can turn on some servers. And then those who are just like consuming the ecosystems will just go where the biggest ecosystem is.

The last completely rambling thought for you on this is that it's been a whole like 30, 40 minutes and we haven't said the word edge yet. I haven't got edge in my title. They didn't put in my title. But I do see a big movement towards highly distributed workload being the native instead of like runs in US East being like runs everywhere. And that's because front-end systems, architectures like Jamstack and related, you know, having a consistent backend, databases are like planet scale and like things are just like becoming more everywhere,

which I think is super interesting and totally changing the way workload distribution is happening. I'm not sure exactly what that means for like cloud or not or whatever, but I do think that's going to change the way that applications work this movement towards runs everywhere that was such a well-encapsulated answer in terms of like painting like a broad picture you have people who never racked and stacked only ever used the public cloud being able to come down because now they have a lot of the same automation layer that exists for like I don't have to deal

care about the hardware the nice folks at Econinix are going to like put the server in there and I get like, you know, it gets booted and there's an API. The next picture where you have all these enterprises that like literally are just trying to not own the CapEx of their data center, which is the original fail point of the public cloud, right? Which was like, okay, we'll stop doing CapEx, move it up to X. It looks like subscription revenue.

And I don't want to have data centers. I don't want a physical footprint. I don't want all these things. I don't want that to be my specialty. Those are all great arguments. But the third thing being the edge. And I'm really curious, from your perspective, it sounds like they're all operating on the same basic idea. There is a standardized layer. It's the operating system. There is the hardware. There is the API. You have the edge. Are there primitives you see missing that you think need to be at the hardware layer

and the hardware automation layer to enable edge deployment? And these new edge apps that we talked about, we start seeing like the very, very, very, very beginnings of. And I'm really curious also to see like, are there specific use cases you've seen emerging related to edge use cases that you think,

holy, this is like a whole new like amount of workload that we'd never seen before. Like you mentioned briefly, edge native being the first place, but really curious to like dig in on that as well. Yeah, well, I definitely think there's things that are missing. I still think we have a long way to go in the consumptive models of hardware. I don't think personally, that anybody is going to like wake up one day and be like, you know what I should do is put some super generic app and like 500 locations, just fun. That should be super fun. No, it's going to be like, I am trying to do this extremely specific thing. And it really either matters performance wise, it matters location wise, it matters, you know, kind of capability wise.

A friend of mine, Ivo Rook said when he was at Sprint, he's like, I'm trying to bring the intelligence to the data, like I'm going to bring something to it. And that's generally going to mean somewhere like opinionated. Nobody does it just because, I mean, maybe that will happen. Maybe there'll be a easier, cheaper, faster way too, but like usually a two out of three. I think that it's going to be somewhat opinionated. So I think we're going to have to improve still the way that we, especially right now in specialty silicon right now, which is changing so rapidly. It's all about offloads. Those offloads are in the ultimate Moore's law every 18 months.

It's like the A100 is killing the A200. The A200 is killing whatever it is. It's like, tear out all the things we just sold you and put in the new things because it's five times better. And so I think that is still a friction point that we need to make way better for various reasons. And then I think that we still have a long ways to go on the main form of connectivity that we all use really in our lives, which is wireless. Wireless is like the big surface area that developers have not been able to touch. It's still basically controlled by two or three software companies, Ericsson, Nokia, Cisco, whatever. It's basically controlled by a few operators who are not known for fast product cycles. that are meant to last for 15 years and amortize over that. You're not changing it. You're not doing some like agile process and add new features to the 4G thing, right?

Wireless to me isn't cracked open yet. Spectrum is becoming a little bit, right? You have CBRS spectrum, which is like open 3.5 gig and you have wireless through Wi-Fi 6 and whatever. Like the last little bit of the radio is becoming a little bit more connected. Like that's an access, but we haven't been able to touch anything from like there all the way back to like where it

gets dumped in the you know carriers lte core and i think that that's a super interesting area when you have distributed infrastructure everywhere with software like why wouldn't you have a cloud native super event driven awesomely done everywhere, IoT-driven, I don't know, wireless network or something? I don't know, like something where developers could give it intent. Most of our life we spend on wireless stuff, and yet we don't really write software innovative for that. That to me seems like an area of the edge that is ripe. A friend of mine says, just cloud out. Take cloud models and push them out.

And when you have infrastructure in more places and you have connectivity in more places and you have hardware opinion, like that is going to unleash. Even if you could just put the right kind of cloud style infrastructure in every country of the planet and run the software you want,

you could probably like dramatically change people's experiences of billions of people. That would be really cool. So I'm interested in that. What was the other thing? thing i don't know i always do things in threes so you only got two out of that one the first one was great and the second one was so interesting because i was going to ask you about like how does 5g roll out 5g towers how does this actually change a lot of this conundrum like the ability to push the point of presence at the tower right

you're so much closer but then to think about well, what if you could program the actual protocol? What would the actual wireless protocol itself, right? If that was up level to make it accessible and we could do certain types of prioritizations, if you could even pay for higher priority packets, right? And you have your app running inside the telco. So that point of presence, right? At the tower, I mean,

that opens up a huge new types of real-time applications. Things that we can't even think about. We spend a lot of time now talking about, Tim, we're talking about LLMs. It's a big topic, you know, du jour, right? Is LLMs everywhere, open AI in the world. But at the end of the day, it's super interesting to think, well, if the cell phone can't hold it, could the tower, right? And so all these things that open up and are completely inaccessible to the modern developer,

but if they could, that'd be pretty impressive. How about we could just try every city with over 200,000 people? I would start there with an accessible platform to be able to run code. When I see edge, I see like, wow, there's so much opportunity related to that. But just to put it in the most basic term, software has kind of unlocked and touched almost all aspects of our life. But the one that we spend the most amount of time on is not really open yet. So, Damon.

Assuming that this opened up, are there missing core software primitives or core pieces of technology you think that need to come to actually enable developers to access these things? You brought up an example earlier of how CockroachDB made disaster recovery and managing multiple points of database across multiple different regions relatively easy in an accessible way that wasn't before. I'm curious if there's pieces of software or problems that come of mind as you think about where the edge is taking us on a software layer. Probably one of the most interesting areas for me is actually just in networking. I see a very common thread. What's the thing that stops people from maybe they're really using a public cloud a lot and want to move their workload for one reason i would say almost always it's network vpcs are hard network works pretty good in the cloud i got lots of features and i

don't have to be a network person to like design all the stuff and you take that out and put that in way more places you know whether it's security groups or peering or overlays or just like address matching or nat or whatever like that to me seems like an area of awesome opportunity. I've been a huge fan and, you know, I just was enamored with Thomas Graf and his Starship Enterprise, like Death Star analogy for eBPF. And I can't remember, there's some demo he did back in 2016. I'm like a huge fan of higher level networking capabilities within Linux kernel. But that seems to be an area that there's still just like the primitives are hard.

There's still such a world of difference between like, I am working on the app or the Kubernetes or whatever. And like I do networking feels like super different. And I think that's an area where that, you know, there needs to be more. And I see a lot of that in like Linux foundation. There's like telco working groups and whatnot, but it's not really where the vast majority of Technorati play around. And so I think open source around networking primitives, especially given security and the related distribution of workload,

deserves a lot more. I see that as a missing component. It's too hard to network. It's too hard. It's a fascinating question. Like, should we eventually create a set of interfaces that application developers can all agree on? Because Kubernetes was one of that. But obviously, to get Kubernetes to be agreed to use everywhere will take a long, long time, no matter how many classes we run. But we're also seeing a rise of a bunch of easier-to-use developer experience startups,

Jamstacks of the world, workers from Cloudflare, or DeFlyos, or it makes it easier to just run stuff. And they were really honed on the experience. A lot of people, they don't even care where you really, truly want to care where it runs. But cost is usually the number one thing. And I guess control. We talk a lot about control, but actually, I think it does come down to cost

for a normal developer. Like how much does this actually cost? And is the experience oh so better because maybe I have a much more faster, more reliable experience overall. You know, do you see the future of the software stack coming from 2023 moving forward? Do you see a world where we actually

let developers just write code? Is it chipping away public cloud? Eventually our public cloud are just so dominant. They just move so much faster. There's really nowhere to play here. I'm just very curious because as developers, you get both of these signals and it's not always clear which one is actually more marketing or not. Let me go back in history. Let me just look at it in the rearview mirror. History hasn't shown that big companies are the most innovative.

They actually have a lot to lose in the current model. And, you know, it's kind of fascinating to me that we look at, and I'm like 100%, by the way, super public cloud. So don't take this in the wrong way. If you're listening to this and you're a public cloud. So don't take this in the wrong way. If you're listening to this in your public cloud, right? I just feel like this business model, it'll probably be very different 15 years from now. I think that technology is going to continue to change really rapidly. I think that we're going to have way different models around how this works. I just refuse to believe that we will wake up in like 2035 and have just one major big giant huge public cloud or two. Like I think we will have incredibly specialized sets of innovation. You know, you can already see the ball turning around access to more fundamental primitives. Maybe the fundamental primitive right now is like computers, or maybe the fundamental primitive is a little bit higher than that. But I feel pretty confident that that will become more accessible. Like I put a bet that

in five years from now, there will be a major silicon company who will be working very closely with series A businesses and enabling them with unique chip technology in all the markets of the whole world. And being like, you're a super cool fly.io, Kurt, awesome service is now enabled with super fast, amazing, this that we built for you. And then we made it happen everywhere. That is going to be both better and cheaper of an experience. It's like just the right tool for the job, right? And so you can see that there is a world where in a highly distributed way, you're going to end up with developer experiences that are really, really well crafted for their user base. And they're going to be an operator who gets in there and says, I can make that a

reality for you. And I can do it better and cheaper than anything else that was a generic tool. And that's back to my retail banking analogy. Public clouds are great if what you need is a checking account and you got 500 bucks in your bank account. But if you've got some really sophisticated problem sets, it's probably not the right product for you. And I get down to like, well, hey, if you're the largest car company in the world, and you want to have super incredibly good telemetry off of all your vehicles and maybe wireless capabilities on a EPC that you run yourself so that you can control roaming. Are you going to become a carrier?

Are you going to become a cloud company? No, you're probably going to use like the super specialty edge cloud for automotives that somebody makes, my guess. And there's a lot of those use cases. They don't have to be $100 billion markets. They could be like billion dollar markets. So special, not generic, is I think what we get out into. And then I would say, I'm not so sure that there's going to be one API, one layer of abstraction, but I think we'll normalize it enough that people create developer

experiences that those developers really love. That was really interesting. One of the questions I come back to when I think about sort of edge, the future, do you have a thesis around what you think the macroeconomic drivers are that are going to drive people to the edge and to the bare metal? Like, I'm curious, like, what your perspective is on what are some of the macroeconomic drivers

that are going to force companies to like really consider, you know, should I build that next data center myself? Should I go with Equinix and do bare metal through their awesome automation API? Should I go, youinix and do bare metal through their awesome automation API? Should I go with Azure? What are the things that are going to change the way people think about making some of these decisions?

I mean, macro, macro, if I'm going to zoom out, I'm going to say that probably the biggest one that we haven't yet really seen, but it's definitely bubbling. It's certainly part of my daily interaction, and I suspect it'll become part of the general conversation. I think sustainability is going to transform the way we consume this stuff. Sustainability is going to become not just a, it would be nice, it's going to become

highly regulated. This is going to be not a nice to have. So carbon impact, carbon consumption, real use, optimization of workload, how much carbon does this Zoom call cost is going to become a thing. And that is going to require companies and businesses all the way throughout the chain from like, I'm just a consumer and I need you to tell me vendor exactly what it does. And then I need you to reduce it by 30% per year because whoever government in wherever

told me I have to down to parts of the supply chain or other people are doing it for other reasons. The concept of unlimited, ultimate hyperscale of everything is not going to be a reality. We're going to be the most efficient or the right thing or the right controlled thing for your particular goals is going to be a bigger component. I think that will drive more specialization. I think that will drive different models. I think that will drive diversity, frankly, not generic. So that's one. I think the internet won't be flat. What you mentioned there about decoupling of supply chains and technology ecosystems, I think that's going to just happen in more places. Maybe not as big, but it will be privacy, it'll be data. It'll be, yes, all humans on planet Earth are allowed to say where their data gets stored. That seems to be another one, which is the concept of privacy and humans and their data and stuff is going to become a big driver. And then I think the last one is just that more companies are going to be digital businesses. Companies' entire value prop is going to be built on technology.

I think those companies will care deeply about what those are and will require it to be done their way. So I know this is the last question and I resist this question at 2023 March, right? Chat GPT, LLMs, foundation models. We're going to cram all this buzzwords in just before that. It's so apropos. We got to ask that.

But I'm very curious, Pinal, because I think in a hardware level, we're now seeing like SambaNovas and Cerebras, all this sort of ASIC chips really built around these. And we're seeing more. We're seeing more because of this whole craze. And those are not accessible to most people right now, right? We can only assess things

because cloud hasn't been available for most people, right? Have you been seeing more and more of these specialized processors for AI will want to become more available as it's through a story that Equinix or some vendors like you are able to actually help provide and build a stack around.

Like, I don't know if you're curious what's your stance right now around hardware and AI. The right thing in the right place at the right time, you know? Yeah, I think that that's going to be more and more common. I think we're just starting

and we haven't even gotten into like the real quantum stuff or like even any real shift. We're still doing Ethernet here, you know? Yes, is the short answer. We're going to move from like the vast majority of workload is pretty generic to the vast majority of workload is highly specialized. The silicon will optimize around it or you will work towards that. And that's going to require a different distribution model.

The last time this kind of really happened is we've gotten pretty far by just taking the same technology and putting it in big places with an API. That's been a great optimizer and unleasher of like it removed barriers called IT procurement processes and said, a developer can go and run open software with a credit card, which was not really accessible in 2001. You had to like call the IT person in order to rack of sun microgear and like put it on your, in your colo. And so, you know, now that's like really leveled the playing field, but it's still basically a very generic infrastructure or technology model. And I think that's kind of run its course. I don't think you're probably upgrading your laptop every year now because the, you know, Pentium DX2 is coming out. It It's good enough. Now what's not good enough is

this next level workload, which is going to demand something completely different. And that, in my opinion, will just cause a shift towards different distribution models and access models. It will also really require, and this is going to fight, seen it, super hard to be a dominant business. I always thought it would come from startups, but I can really appreciate now businesses that are at the top of their game. When they're in the zone of being the winning, it's very hard not to do that. In fact, it's almost impossible to change. And when you look at something as successful as some of the silicon models that have been dominating their industries for 20 or 30 or 40 years to then move to a different model where your customers, there's only like 10 of them, or there's only 50 that do this one thing.

How do you create a distribution and a licensing and a subscription model that works for that versus like, you know what you should do is you should buy a whole bunch of my stuff because everybody buys my stuff, right? I think that's going to have to change so dramatically. The current model, like who's going to take the risk? Who's going to buy those super expensive DGX2s and just put them around just to see if you want to use them for a couple months? Nobody. Nobody's going to take the risk and amortize that stuff where they're just going to be so expensive that you won't be able to do them.

And so I think that's going to have to change. And I'm really excited for that because I think the workload is here. It's now captured humanity's imagination and be like, oh, oh, that's really cool. Well, what else could I do with that? Says my mom at dinner, which is hilarious that my mom is asking me about AI, right? But like, I think that's just shown that it's in the pop culture and that will unleash a huge amount of innovation, which will be hard to meet with the current models. And that will cause friction.

And then entrepreneurs and VCs like you will make it happen. I'm excited about that. I'm excited too now. I'm jumping for joy. But I mean, it really brings home the point that the ability of an Equinix and these bare metal providers and what happens is you have this great financing model that makes it like all these new ASICs, these new chipsets, these new things accessible to people at the hardware level with no abstractions. And that's super exciting. Zach, this has been great. I've, one, learned a lot. Totally enjoyed all of your stories.

Where can folks find out more about you and what you're up to? I'm not on the discords. I'm not on the sub stacks. I got to up my game. I'm on Twitter, ZSmithNYC or LinkedIn. I think it's ZSmith, but yeah, you can look me up. Zach Smith, New York city, the guy with the, who went to Juilliard, who likes computers. And plays the bass. Plays the bass. Awesome. Thank you so much. I appreciate it. It's been so fun. Thanks Ian. Thanks Tim.