HELLO FUTURE: The Beating Heart of the Internet

Speaker 1 (00:07):
So I talk a lot about data centers on this
program because it's what I believe going to be powering
next the start of the Second Industrial Revolution, which we
are all living through right now, artificial intelligence. I want
to keep talking about data centers, but I want to
explore today if there's a way that we can build
data centers without breaking the planet. Hello, Future, it's me Kevin.
This is a dispatch from the Digital Frontier. The planet
is Earth. The year is twenty twenty five, it's almost
twenty twenty six, and my guest today he's returning to
the program Virginia Tech University professor Valid Sad. He is
one of the experts on data centers. Of course, in
Northern Virginia has so many data centers. It's one of
the data center hubs in the United States. Okay, last
time we were on, we talked about understanding what a
data center is and how it's powered. How much electricity
is required for a data center. Now, explain to me
how do we build data centers, Like, what is the construction,
what is the manufacturing that's required to build these data centers, because,
as you told us last time, many of them are
about the size of a warehouse or a Walmart. What's needed.

Speaker 2 (01:16):
A data center needs several pieces of equipment. One is
the GPUs, right, So you need computing, the processors, the
so called server and I mean if you've heard of
a server rack, it's like in these movies they enter
into this room all black servers, right, and you know
they hack it and so on and so forth. So
that is a piece that is very important. That is
actually the heart and soul of the data center. And
then you need connection to the internet, So you need
networking equipment, connection to the communications system. You need power equipment,
you need to be able to do You need electricity access,
and of course you need as we said before, like
a warehouse or a land footprint that can host all
of these components.

Speaker 1 (01:54):
Don't you also need freezers or everybody's talking about freezers nowadays.

Speaker 2 (01:59):
Yes, So centers when they do compute, they consume energy
and they overheat. They potentially will generate heat, so you need.

Speaker 1 (02:06):
It's like when I've on my iPhone and the thing
starts getting hot or my laptop starts getting hot. But
if you told us last time that these data centers
are using the electricity of like fifty thousand homes. I
would imagine that thing could get pretty heated.

Speaker 2 (02:19):
You need cooling and terrible management, right, and that includes
that cooling, towers, chillers, you know, heat exchange, water treatment,
potentially equipment because without cooling you can addun these twenty
four to seven. Remember they have to be operating all
the time. Wow.

Speaker 1 (02:34):
Okay, So the reason I like that is because the
reason I like that question is because my genius, oversimplified
idea is put all these data centers in the desert.
How hard can it be? Just put the data centers
in the desert? But I guess it's really complicated because
the desert's hot, and so if you're trying to build
them in Arizona, for example, you need more water and
coolers and whatnot. And I guess here's another really basic
question is if you build the data center, how far
can it travel? I don't even know if that's the
right question, but what proximity do you have to be
to use the data center? Do you get what I'm
trying to ask?

Speaker 2 (03:11):
Yes, yes, absolutely. I mean I think of it this way.
The data center you use it through essentially the internet, right,
So no matter whether if you can reach it through
your Internet. Then you have potentially you're good to go.
But there is a caveat that the further away it
is the potential of latency will be high. How many
milli seconds does it take to use that data center?
So think if you're gonna do something like a remote surgery,
you cannot sustain very high delay because you need to
do it like remote surgery to a r VR.

Speaker 1 (03:42):
Okay, I just had to meet the future moment. Okay,
because maybe my original idea of putting all these data
centers in the desert wasn't so stupid. After all, there
is nothing worse than when you're on the phone with
someone and you hear an echo or delay. I mean,
I'm telling you, I love my mom. I'm a huge
mom's boy, but that thing drives me. I rate when
I'm talking to her back home in Delco every morning
and I'm thinking to myself, what is this. It's like
I got a walkie talkie and I'm somehow trying to
figure out. I mean, it's it's almost like you have
the two tin cans and the wires, you know from
when you were a kid and you would try to talk.
It's crazy. The delay thing trips you up because I
hear a lot about putting the data centers on the Moon,
and that's a brilliant idea. By the way, American data
center should absolutely be on the mood. But the delay
of that if you're a doctor putting on the VR
goggles and you're trying to do remote surgeries and then
the communication system on the Moon is a delay of
five seconds, no offense, doc.

Speaker 3 (04:38):
Doc, I don't want you operating on me on a
delay of five seconds, okay, So we got to make
sure that we're modernizing as we're building the communications infrastructure
in outer space, we got to make sure that it
doesn't operate on a delco delay.

Speaker 1 (04:55):
Am I right?

Speaker 2 (04:56):
Absolutely? So I mean and also think of a more
fun issue. When you're gaming, right, you want the AI
to be delayed. So I mean, I think the solution
is somewhat somewhere in between. So there are use cases
where it's okay, if it's very far, you don't need
that much of a small delay. But then you can
supplement that, which with what we call an edge server,
which is like closer to the users, maybe smaller data
centers or smaller telco infrastructure that you can use locally.
So kind of a combination of the two. One is
for the real time application, one for the more further away.

Speaker 1 (05:27):
It's the analogy like putting in a cell tower.

Speaker 2 (05:31):
It's not a cell tower per se. You need computing power.
So the cell tower gives you communication. Some of our
more advanced ones they can give you compute, but you
need what we call the technical term is a mobile
edge computer or mobile edge How big are they? It's
much smaller. I mean it's potentially the size of a server.
You can put in a room in a lab.

Speaker 1 (05:50):
It's a refrigerator, yes, more or less, more or less, okay.

Speaker 2 (05:55):
And it could be part of the telco. Like you know,
where you have the cell tower, there's usually a room
and control room with a memory and other devices. It
could be part of that. So it's co located with
with the cell tower, but it's not the cell tower.

Speaker 1 (06:06):
It's called a edge computer edge computer. So all of
this is modernizing our infrastructure. I mean, I always talk
about the analogy on this program. It was just one
hundred years ago when America said, you know what, we
don't want horses anymore, we want cars. And with the
advent of cars came a lot of pooper scoopers who
were put out of business. But what it created was
I don't know the automobile industry. I don't know the
international highway industry. And I think I will politely critique
the communications consultants in the nineteen nineties who said that
the Internet was the information highway.

Speaker 3 (06:40):
Eh.

Speaker 1 (06:41):
Really, data centers, I think are the highway to the future,
and these edge computers are the highway to the future.
And these computers that's what's really modernizing everything. Should we
put data centers on the moon? Please say yes?

Speaker 2 (06:53):
Perhaps if we want to move to Mars, maybe we
should put them on the moon.

Speaker 1 (06:56):
But for the demand for data centers and just there's
so much I mean, if we're trying not to break
planet Earth, if we were to put them on the moon,
wouldn't that make it better for us hypothetically.

Speaker 2 (07:07):
I mean better from an environment perspective, potentially, But again,
the latency is I mean, there's a laws of physics,
So the laws of physics, we cannot break them. And
there is a limit how much delay you can sustain
when you have something this far away.

Speaker 1 (07:21):
So what do you think we should do?

Speaker 2 (07:24):
I think there's a couple of solutions one is algorithms
that are using data centers, they can be designed in
a more efficient way, so we can design algorithms that
use less compute that are smarter. In fact, if you
think about it, the most efficient and intelligent device in
the world is the human brain, and in fact it's
the most energy efficient. When you're driving and you, you know,
encounter something new you haven't seen before, you don't need
to have to go back to driving school, right. So
for example, if an AI algorithm now, so see something
it has I mean, in AI, you have two phases
what we call training, what you can all inference when
you are using it, that's the space you're actually using
the AI to predict. When you talk to CHGPT, it's
usually doing inference, but it is limited what it can
do in inference depending on what it has learned in
the training. So the issue is that sometimes when it
encounters things it hasn't seen before, it has to do
new training and that is even more and more energy consumption.
You as a human, you don't do that, right, You
don't train yourself again. I mean, if you learn how
to open the door and I give you a window,
you will figure it out. You don't have to go
back to learning how to open doors. So AI today
doesn't do that, which is why the algorithms they are
not that energy or not that compute efficient, which eventually
ends up being energy efficient.

Speaker 1 (08:39):
Do you think America is well positioned for the demand
for data centers.

Speaker 2 (08:43):
I think we will figure it out. But I think
the solution is not building more data centers, but actually
doing more efficient algorithms and being more efficient in the
way we design those data centers. So there are people
working on the hardware research. I'm not a hardware person,
but they can build better devices. They can build smaller
footprint devices that can even do better computters. So there's
a lot of investment we have to do there, like
let's reduce the footprint and increase the power of those devices.
I can think computers. Who would have thought that you
can do what you could do on these big computers
twenty years ago, Now you can do it on a
small tablet.

Speaker 1 (09:19):
So I think I like the analogy that you gave
earlier about video gaming, and so if you think of
the gaming device as the data center, what you're suggesting
is we need better games. We need games that are
more energy efficient, we need video games that are using
and built in a way that doesn't require us to
break the system. Which is fascinating because I think, especially
as there are more LLM models that are created and
marketed to different types of sectors, that's just going to
be more and more important. Professor Valid Sad of the
Virginia Tech University, I learned so much talking to you.
I feel infinitely times Marner, thank you appreciate it.

Speaker 2 (10:09):
M