We’re chatting with Jacob Smith, CMO at Packet, about running a bare-metal cloud & edge computing infrastructure business.
Jacob gives a great insight into how the future of software infrastructure might look like and talks about the various hats one has to wear when running a successful company.
Watch this Episode on Youtube
Full transcription below.
Semaphore Uncut is a show where we talk about developer tools and the
people behind those. With us today we have Jacob Smith from Packet. Hi
welcome – it’s great to have you on the show.
Jacob [00:00:14] Hi Darko. Thanks for inviting me. Great to be here.
[00:00:17] We’re doing it live. If you have any questions, if you have
trouble seeing or hearing us, please let us know in the comments. Jacob
feel free to go for it and introduce yourself.
[00:00:33] Thank you for doing the usual thing and change the schedule a
few times right. Yeah, it’s busy days for everyone but it’s great to be
here. A little bit about me so as you said I’m with Packet. I’m a
co-founder and we started the company about four and a half years ago.
As part of that journey, I’d got to wear all the hats right. I have a
Packet hat today but at the beginning you know it was me and my brother
we started the company together. He had a background in infrastructure. I
had a background in marketing. As part of that, I wore a lot of hats –
helped on the product side for a long time, in the beginning, even
running the product team for a while and then moved closer to my home
base of customer success and marketing. Right now I even direct sales
and alliances, so it’s one of those things where as a co-founder you get
to do all the things and then you learn to delegate to people who are
way better than you are which is a really exciting part of the journey
Darko [00:01:32] Yeah. I can confirm that. So it’s
like every year, year and a half maybe two years you kind of change
your hat as you said and you have to learn something new.
[00:01:43] Yeah it’s really fun to do that. It’s a little hard because
each year it’s a new company, sometimes every three months it’s a new
company. Right. It’s been a lot of learning and a lot of excitement.
[00:02:00] It’s something hard. But on the other hand that’s what makes
it exciting. So you blink and then it’ll be 10 years, but it’s really
fun because it will be a different company than it is now.
Jacob [00:02:13] That’s right. We earned all of our gray hair.
[00:02:16] So maybe if we can just go a bit deeper. That looks like a
great combination in the beginning to have someone who is very technical
and on the other side someone who could understand customers and
marketing. Can you share a bit about the journey through customer
success and transitioning through those roles that you had.
[00:02:48] Yeah well it ties right into our founding story. So for
anyone who does not know – Packet is generally known as a bare-metal
public cloud. We like to describe ourselves a little differently. We’re
in the business of automating fundamentally infrastructure. We happen
right now to have three products one of which is a public cloud. We’re
in 20 locations. We deliver a sort of you know easy to style experience
but with bare metal and physical networks. That’s kind of our place in
the market. We’re also offering on-premise software-only version and
then an Edge model.
Either way, it’s the same experience and
the same technology under the hood which is – we’re just really good at
turning computers on and off. Right. It’s not rocket science here. I
think that the backbone was really like “Why enter the cloud business in
2014/15?”. I mean it wasn’t like “I know we’ll go disrupt Amazon and
Google and Microsoft. They should be vulnerable”. Now it’s really about
this – thinking about the infrastructure you tend to think like 10 years
ahead because it’s like building railroad tracks – it’s really slow.
in the cloud is just like infrastructure and there real world – there’s
a lot of physical parts to it a lot of capital. This is really tying
into my experience as you listen to the users and you look at the cycles
of product and product/market fit. We saw that the public cloud really
represented mass adoption of automation for IT for pretty cheap and big
workloads, but pretty generic. What we saw coming next was the natural
sort of specialized workload and with that the change in who the buyer
The buyer of generic compute that kind of does most
things for you – it’s kind of like getting a free checking app at the
bank. It’s pretty good, works for almost everyone, it’s easy to get –
it’s there on every corner. And we thought that there would be sort of
that wealth management, you know like that totally different than
Understanding the customer and what they
wanted and who they were is really key to why we started the company and
how even farmed our product and our approach to everything. My partner
and my brother is my twin brother so we know each other really well. He
had about 15 years in the cloud infrastructure space so he knew all
about infrastructure and what we tried to do is to think who is coming
next and what do they need.
Darko [00:05:24] Yeah thanks for
sharing all that. There is one thing I mentioned that made me remember
Packet because just visiting the website and trying to use the service
and all that, it had a very different look and feel. When you enter,
when you need to get a bare metal instance.
I was mentioning
before to call we had that journey from being purely bare metal and then
we said: “OK maybe we should adopt something of the cloud goodness and
it feels that we need that”. We went all the way, we used SoftLayer
which is I think bought by IBM and kind of disappeared, at least in my
eyes. Then AWS, then Kubernetes brought us to Google and we ended up
using two public cloud providers for some workloads but having bare
metal as the backbone of the workloads that really need to be efficient
and cost-effective and all that. Just wanted to share that buying bare
metal hardware in comparison to dealing with hosting providers were was a
very different feeling.
Jacob [00:06:49] I think it’s about
understanding and this is all about how your product is positioned.
We’re all selling the same thing – we’re all selling computers. We’re
all actually selling Intel technology, we’re just packaging it in
different ways. So I think the key is if you think of IT as the old
“cattle versus pets” thing. That’s a server – it’s dedicated to me, it’s
mine, it’s particular, I like this I upgraded the RAM on it – it’s very
much the IT experience.
Like you said SoftLayer innovated
the data center enormously. They took IT and made it like super
efficient and actually almost fun for an IT guy. If you’re thinking
about a cloud buyer or developer – what matters more? Our best customers
deploy 100 physical machines which is a lot of cores and a lot of RAM. I
mean it’s great. One of my favorite users is Graham from NixOS and he
has posted on Twitter about how much… Graham in course he deployed to
get through it like a backlog of 30000 jobs for NixOS through the CLI.
He likes the power of the server and he’s excited by being able to get
the job done but he just destroyed them. There’s nothing special about
them. They’re just servers, just like VMs. Then, of course, you start to
design an experience that’s API-driven or that’s less about a
downgrade, change this changed that, and it’s much more about “I’ll take
another one, actually give me five”.
Darko [00:08:38] That’s
exactly the experience the developer wants. I want to make mistakes and
I want to destroy this and that’s the way of thinking. Generally in
software whenever you do something, tomorrow you’re going to throw it
all away or rewrite it. There is a cost to this time and with hardware,
it usually meant “Okay I need to ask some people to do something with
this physical server and then they really need to do something physical”
and removing that barrier of being scared to do something with that
physical server – an amazing thing.
Jacob [00:09:12] Yeah.
That’s a pretty simple concept, but it’s hard to do. Physical hardware –
it’s not smooth right. It has rough edges. Things are different between
physical machines, between each machine. They do not always behave the
way you want.
So our mission has always been to try to
provide just the lowest level of abstraction e.g. like tooling so that
you can deal programmatically with that. Instead of saying “Let’s
abstract you so far that you can deal with it” we’re saying “Let’s give
you tools to deal with that when it breaks itself”. When you change the
kernel and you can’t get back in because “Whoops”. Right. Well, how do I
help you deal with that? Because that’s a real thing. And the reason
why we’re most excited about this, why hardware and why does it matter.
It’s not religious for us it’s not because VMs are bad – VMs are great.
That’s all good. It’s really because we see a world in which there’s a
lot of hardware that makes a big difference in what we’re trying to do
over the next ten years.
For trying to fly rockets to Mars or
trying for taxis to drive themselves and we’re trying to talk to the
wall instead of at our screen. We just are going to end up with a lot of
hardware that kind of looks like our phone where you know you started
designing this hardware around the software experience and you combine
those two things. That means you’re just going to have… We as a
community as an ecosystem of developers just have to get good at dealing
with a lot of variety of hardware because it’ll make a difference to
our use cases.
Darko [00:10:58] Now that’s a new thing that’s
coming up in the last few years. There is one thing that keeps coming
up and that’s the ARM is something that’s powerful for that. I know from
different announcements that you guys have been making actually over
the years that you have been using ARM before ARM even existed. So maybe
you can shed some light on that. There was recently a discussion I
think Linus Torvalds wrote something that maybe that ARM is not going to
fly really before making it very close to developers – they need to
keep it in their laptops and their hands and all that. I kind of
disagree with some extent…
Jacob [00:12:04] So the question is why ARM and what’s it all about?
[00:12:06] Yeah I think that’s it. At least our customer base and
people that we are talking about are very close to the Internet and I
would say that it’s still far away from us. Let’s say to put us in the
group of you know web developers so maybe you can shed some light on
that. Where do you see it’s going and why it’s so exciting.
[00:12:35] Yeah well we got into the ARM business really not because we
thought there was some massive you know missing product in the data
center market. We’re huge consumers and fans of Intel technology and I
don’t see that it’s the reason to be excited about ARM. I do see it’s
because of diversity.
So the reason we got involved in ARM
was that we looked around for ecosystems that we thought would support a
diverse hardware model in which you know what was it Tesla yesterday
talking about their specialty chips that help them drive cars better
than the video chips. Yeah. It’s their own chip but it’s probably (and I
didn’t check) but I’m almost 100 percent sure that it’s probably an ARM
licensee. And the reason why the ecosystem has some very interesting
benefits – does it make sense for…
I can’t speak for any
company but does it make sense for an Intel to go build a chip for
Tesla’s cars. Maybe? But maybe not. The reason why we got into ARM is
that we saw diversity, we saw the ecosystem of a thousand licensees and
partners who will essentially build you anything from a set top box chip
has no fan you know to a ninety-six core dual socket data center chip.
Now not all ARM architecture ecosystem problems are real, but we thought
that hardware getting more diverse meant that we need to enable
developers to have that in the cloud. I think it’s absolutely important
that it’s in your laptop like what Microsoft you know notebooks and
stuff are doing with ARM is really critical.
experience is good but it moves pretty quickly to the cloud. When it
gets real it moves to the cloud and so that’s why we released an ARM
server two and a half years ago. Since then more use cases have come out
where it makes sense, there are certain ones around Android or around
particular workloads where it’s like a really good. To be honest, we’re
not religious – we don’t care. I mean I really don’t care if we’re gonna
be working on RISC-V, I’d have Ariane in my data center right now if I
could. Because mainly that to me is a developer or a customer choice and
what we need to be good at is kind of automating anything no matter
what it looks like. It could be an X86 Dell server and a data center or
it could be a small ARM device on a telephone pole. It’s still a
computer that you need to `terraform apply` and deal with and that’s
sort of the that’s what we’re working on.
Darko [00:15:21] Now
that you mentioned cell phone tower and all that in one chat that we
had you mentioned 5G so ARM is something that I can somehow relate to. I
do have it on my phone and so on. 5G and how that will influence how
data centers are built and how all those devices are talking together is
pretty much out of my depth. I know that you are very excited about so
maybe you can teach us a bit.
Jacob [00:15:56] I’d be happy
to. Yeah well I mean it’s a big buzzword right and we love buzzwords
like edge and 5G or all that’s going to change the world. So let’s not
start there. I think the most exciting thing about the wireless world
right now is actually that there’s a chance for developers to change the
shape of it over the next five years.
So 5G is obviously a
major standard that’s going to need some technology it’s a long time and
coming. It will begin to be in lots of places but that’s sort of
aligning with a bunch of other trends which are… The main one being that
software developers just keep eating down the stack. Right. So a couple
of years ago it was crazy when Alex boldly told me he was going to
start CoreOS and begin building a new operating system. I was like “What
are you talking about?”. People are like “Oh I can probably do overlay
networks better than they are now and I can probably start working on
unikernels”. So just down, down, down and as you get down we’ve
stretched you know, we’ve kind of automated through the stack and
there’s one thing we haven’t touched at all.
owned by about 10 companies in the world is the wireless stack we all
use all the time. Developers don’t get to deploy/touch/automate/control
and use wireless. They get to do something up until the point where it’s
wired. In the US we have things like CBRS which is called Citizens
Broadband Radio Spectrum where it’s basically rentable spectrum. You can
take 4G/5G spectrum allocation and use it for just this one mile. Think
about a football stadium. We have a data center a small data center
next to a football stadium outside of Boston. Well, it’s not very
important except for when there are 80000 people there who want
connected experiences and video and all these other things. So suddenly
it’s like “Huh?”.
So what we’re seeing is that wireless is
starting to get democratized for developers and I think that will be the
most exciting thing that we haven’t talked about, over the next five or
10 years, is that wireless will just totally be a part of the developer
stack just like any other part of networking. And that’s really cool.
Now, 5G is causing some interesting trends. I think the biggest one is
what I would lump into disaggregation. So basically 5G and any other
advancement are creating that much more data. It’s just more data we’re
all using more of it our cars are generating petabytes of it.
makes it very expensive in the current model to backhaul it and deal
with it – bill it, route it. And so the first thing it’s doing it’s
taking what I would say is the first edge use case is Telcos and Telcos
are having to push their network out to where the people are. Instead of
centralizing it in a few places – it’s just too expensive to move all
the data. So they’re pushing it outwards and putting computers in every
city and having it terminate and do it there.
I think that’s
exactly what all other kinds of use cases around 5G do say “I want to be
close to that data. What can I do with that data? Can I do store
automation differently?”. Now that doesn’t necessarily mean a data
center could be a few Raspberry Pis in the back of a Chick-fil-As. And
so it’s just a matter of moving the compute and things closer, mainly
because of cost but also because that’s where the experiences are.
That’s what the people are.
Darko [00:19:34] So if I’m
starting to understand this. If for instance, AWS has 20-30 data
centers. So you’re essentially talking here about hundreds/thousands of
data centers being very close to people like in every city?
[00:19:59] I don’t know the future, but the one that we’re planning for
architecturally is millions of things and thousands of places. So
that’s what we’re thinking. That’s because if you look far enough – if
you can get a model together.
Think about IT closet ten years
ago, we all had an IT closet in a building where we worked and that’s
where you get the emails from right. That’s also where you ran a
database and then we moved it all to the cloud. Now I’m just going to
pretend that you could put parts of Zoom and parts of Slack in the IT
closet again. Would that change what Zoom and Slack could do for you?
Would it be better? Different? I don’t know. That’s actually not my job –
that’s developers jobs to figure out but I think it would be different.
I think if we could do that then we would do what
Chick-fil-A is doing in St… I’d got this localized data I want to do
something to get better intelligence about my business and align my
supply chain. I mean that’s IoT basically. Or I want to think about new
experiences I can power with local GPUs – I don’t need a lot of them but
just a few of them so I can use a camera to watch everyone who comes in
and out of a building and understand is that person an Airbnb person or
So this is I think the opportunity, whether it
will be a data center that looks like what we know today or if it will
be something much different, I think that’s yet to be determined. But I
do think we’ll see a lot of computers and a lot more places and the
current model is not suitable. The current model in the cloud is that
we’ve put a lot of the same thing in a few places. You know 15 or 20
places around the world. And I think we’ll probably see the other
challenge – how do I put a lot of different things in a lot of different
Darko [00:22:02] It’s a different scale and different
approach with different industry but we saw a lot of challenges of
that. Customers in Australia, talking to customers in Europe and data
centers in the US and all that…
Jacob [00:22:18] The speed of light issues.
[00:22:26] Yeah and what you mentioned at some point – at some scale
you don’t see that. And then when you start having that other scale –
just a different class of problems hits you. And then you discover that
the internet and the network is a very unreliable and very slow thing.
Although you can watch 4K YouTube it’s not streamed from the other part
of the world.
Jacob [00:22:50] It’s caching, that’s caching,
right? I think it’s funny we named our company packet. Because it’s
networking terminology and it’s the one thing you have to buy from a
cloud providers network. You can not buy it. And yet it’s kind of like
one of those things that we assume just works you know the architecture
of the Internet – it’s actually really fragile in many ways. It’s like
“Whoops I kind of announced your BGP section and routed everything
through China. Sorry about that”. It’s kind of weird right?
think the internet is going to go through an overall re-architecture.
We push the architecture of 20 years ago pretty far, 25 years ago and
we’ve changed enormous amounts in software, enormous amounts. I think
that’s the cool thing is that hardware and networks are going to start
to move at software speed. They’re going to start to be influenced by
that velocity and they will have to change. That doesn’t mean it’s all
going to be my way or someone else’s way. I think it’s going to be a lot
Just like Kubernetes is all the rage now. And
if you ask Dan Cohen it’s like a lifecycle. Five years from now it’ll be
something else. That’s how it goes. We go through these waves and I
think it’s just a little more complex of a challenge when you talk about
infrastructure because it’s so expensive to deploy and to put it and
market it and to put glass in the ground or fiber under the ocean. So
there’s a little bit of a nuance there that we will have to go through
together between you know the ecosystems to figure out how to get the
Darko [00:24:32] Yeah it’s true – we as
software developers, the physical world is so far from us and as the new
generations are coming up maybe IoT will put them closer to hardware.
[00:24:43] Are you seeing that through your company? So people are
asking you? I’m sure that they’re building four more kinds of devices,
Darko [00:24:53] We know that we have iPhones for
more than 10 years and all that but it was kind of a single thing. And
as you were mentioning throughout this conversation that it’s many of
those little and very diverse things. That seems to be picking up.
People want to talk with different things and you know to test them and
all those things.
Jacob [00:25:17] They’ve got to do it on CI.
You know this is the thing. That’s I think a huge opportunity. We are
working with our friends at Grafana recently on how can we bring cloud
native metrics and monitoring to all the other stuff down below. Because
at some point yes I need to know about my PD use and my switches and my
BMCs and all these things. Right now it may be in a data center but
what if it’s at a phone poll? What if that
really-special-expensive-cool-does-the-thing-I-needed-to-do widget is
not in your building?
There are all kinds of issues there and I
think it’s supercool, but it all requires software and testing. So then
how do you automate testing against a growing variety of hardware? I
even just think about GPUs right. GPUs and accelerators and suddenly you
have a whole another layer. And we’re starting to see… They’re
releasing hardware every 12-18 months. I mean look at Google – they
released an entire TPU hardware thing with software in like 18 months.
So I think we’re going to see the challenges of that coming to software
development where especially as we’re seeing it enterprises who say “Yes
cool I’ll buy that but support it for 10 years”. Then we start to have
this legacy in this broad variety – we’re going to need to think about
how we do that.
Darko [00:26:57] Maybe one final topic which
is closer to the developers which we were mentioning, but this is not
the territory that they are thinking about and seeing so I think that’s
one of the reasons that this could be exciting for a lot of people. You
mentioned Kubernetes and the way that it is changing the landscape and
there is the multicloud thing or hypercloud. You mentioned AWS and
they’re covering a territory which is you know IT-oriented and you being
a very different company in that area. How do you see Kubernetes
influencing that? Maybe changing how the light shines on the Packet and
AWS and other providers.
Jacob [00:27:55] Yes. It’s certainly
making a huge difference. I mean I was working on a blog post. Speaking
of marketing right, I’m sure we’re all working on that next blog post
right. But for me I was working on a blog post about “Whoops we forgot
to build a managed Kubernetes service” and it’s true that we don’t and
we won’t. And that’s not helping us. You know I’ve got customers who
would say “Well I’d move everything over to you if you had a Manage
Kubernetes Easy button”. And you know my answer and I think it’s
presenting a challenge to us as a company, it’s testing values like
“What are we trying to do here”.
Because for me – people
build businesses whether their name is Red Hat or working at Google for
GKE On-Prem, or their name is something that just started out yesterday.
We’re not in the software business. We’re not in the “as a service”
business. That’s what the ecosystem of really excellent software
companies like Elastic and Grafana are in – that’s what they do. And so
it’s presenting a challenge for us, but I think it’s okay we’re kind of
working through it because in the end it makes everything more portable
and provides I’d say that single kind of control plane that people are
looking for whether they’re you know at Google or at Packet or On-Prem.
don’t think we’re there yet. I think that there is still basically a
reflex of Kubernetes ecosystem to assume primitives that you can’t take
home with you: “Everything’s fine. You just need an object store.
Everything’s fine you just need a really good load balancer”. And then
suddenly I think as we’re pushing things into enterprises and I think
that’s where it’s going. So in Telcos and Banks, a lot of people are
saying “Yeah but I’ve got this problem in that I know I need to have
staff in Geneva and there’s no cloud there and I need to bring all that
myself”. So that the ecosystem I think is reacting to that and started
to say “Just bring it as software” instead of assuming that it will
always be there. Right. And that’s I think a big shift by Kubernetes and
I think we’ll be on the other side of that in a very good way.
[00:30:09] Kubernetes I would say in very early stages. As a data
center operating system it’ll be around for a very long time because I
think the world wanted something like that for the last 20 years but it
was not around. Well, there’s still a lot of things to be built on top
of that. I love the title of your blog post.
I think that’s one of the big topics of the day obviously Open Source
is going through our moment and as a cloud provider, you have an
enormous advantage. I don’t judge the business models of any other
cloud, I think it will all work out but I do think that you make a
decision about are you in the software business and you just happen to
own the real estate and the data centers and the hardware and the
connectivity and everything. Or are you in the infrastructure business?
For us, it’s really clear we’re in the infrastructure business.
invite partners to be in the software business to build on top of us.
Almost all of our customers are delivering a service, kind of like you
are, right like to their users. That’s why we’ve decided “Hey you get to
choose – do you want multiple tenancies? Do you want virtualization? Do
you not want that?” and you’ll decide because you have opinions about
that may matter to you. I think that’s an important part of who we are
and hopefully, of course, we can build a great business with that in
[00:31:45] It was a pleasure talking to you. If I maybe
missed to ask something which you think is very important and you would
like to share it?
[00:31:56] I think I know how to talk
about a lot of things but mainly I just love that I’ve been… I didn’t
tell my background but my training is not as a marketer or as a
technologist. I was trained as a musician and my brother was as well.
Funny enough he was a junior art grad and I just find that the
technology ecosystem right now is a really exciting place to be in and
that everyone can come in with creativity. You can come in and say “I
have a new way of doing things”.
It’s one of the most
exciting things that we can do and so technology aside, I think it’s Tim
Hockin said: “It’s an exciting time for boring infrastructure”. For me
like a musician and my colleagues and everyone comes from a different
background but they bring ideas and say “You know what, I think we could
do that better. That would be cool.” That’s what’s so exciting to me
about software as well: “You know what. I’ve got an idea about how that
could work better”. That’s going to be really important for how this
world works. If we want to do all the things we want to do and keep the
planet healthy and figure it out, we’re gonna have to come up with new
ideas. So that’s just an exciting part of being in the business – I feel
Darko [00:33:10] Yeah that’s where very
exciting. Being able to you know welcome all the makers, people who want
to make things and ship products that’s an exciting thing to be in. You
know I love that part of this industry, that people from very diverse
backgrounds can come in and contribute in very creative ways.
[00:33:35] I think it’s generational too – we’re a generation of
builders and makers and that’s why abstracting us away from some of the
things is really intriguing to me. So what does it say “Build” right
“build a better Internet”.
[00:33:47] OK great. It was a pleasure talking to. Thank you so much.
[00:33:53] We’ll see you online. See you. Bye!