1
00:00:00,000 --> 00:00:05,720
Oh, the money in the data center. Yeah. Money. There's a lot of money. Open the

2
00:00:03,640 --> 00:00:09,480
doors of a data center and Data center? Oh, data centers could Data centers Data

3
00:00:07,720 --> 00:00:14,480
center, great. Yeah, data centers. Data centers. They've been dominating the

4
00:00:11,600 --> 00:00:19,840
headlines. New builds, big expansions, and trillions invested. But, what rarely

5
00:00:17,000 --> 00:00:24,640
gets mentioned is that a data center isn't just one thing. They're typically

6
00:00:21,920 --> 00:00:29,120
set up in nine different ways, and there are a few unique distinctions that give

7
00:00:26,720 --> 00:00:34,640
each one an archetype. Kind of like the members of your favorite boy band, but

8
00:00:31,440 --> 00:00:37,040
with less music and even more fan noise,

9
00:00:34,640 --> 00:00:40,600
which is surprisingly the annoying sound that started it all. In the beginning,

10
00:00:40,040 --> 00:00:45,880
uh it was really around enterprise data centers were the data centers, right?

11
00:00:44,120 --> 00:00:49,720
And a lot of the times you used to hear about were rooms called like computer

12
00:00:47,720 --> 00:00:53,480
rooms or server rooms, and there wasn't even a data center, it was a let's put

13
00:00:51,560 --> 00:00:56,880
all of our, you know, servers inside of one room so that all those fan noise and

14
00:00:55,280 --> 00:01:00,080
all that heat can be shoved into the one room, and we don't have to have that on

15
00:00:58,440 --> 00:01:02,800
our office floor. Well, that transitioned into, well, we want to make

16
00:01:01,800 --> 00:01:06,320
sure these servers are running efficiently, and this room's super hot,

17
00:01:04,879 --> 00:01:10,320
we need purpose-built air conditioning, etc. Let's make sure we've got power

18
00:01:08,200 --> 00:01:14,160
reliability, and all of a sudden you've got the like the the primitive makings

19
00:01:12,040 --> 00:01:17,440
of a data center, and every enterprise, every business in the world kind of ran

20
00:01:15,880 --> 00:01:21,960
their own set of servers inside of their own office building. Like John said,

21
00:01:19,680 --> 00:01:26,640
enterprise data centers are where it all began, the founding member of the band,

22
00:01:24,600 --> 00:01:30,800
so to speak. These are run by one company for its own use. Think banks or

23
00:01:29,240 --> 00:01:35,280
big retailers. Since they want to keep their data secure and meet all of their

24
00:01:33,040 --> 00:01:40,480
regulatory compliance rules. While enterprise data storage may have started

25
00:01:37,440 --> 00:01:42,520
small, today these operations tend to be

26
00:01:40,480 --> 00:01:48,000
bigger, and that's in part because there are more options now, like colocation

27
00:01:45,640 --> 00:01:52,080
data centers. These are places that rent out space for companies to put the

28
00:01:49,720 --> 00:01:56,000
servers that they own. A great choice for medium-sized businesses. For a lot

29
00:01:54,320 --> 00:01:59,080
of companies nowadays, you know, when you're a small like, you know, you're a

30
00:01:57,360 --> 00:02:02,440
startup or you're you're a small enterprise or like, you know, a

31
00:02:00,800 --> 00:02:07,040
home business or like that, a lot of folks just use the cloud, right? But once you get to the point where it's

32
00:02:04,720 --> 00:02:10,880
like, "Ah, I I've got probably, you know, a five five-digit kind of monthly

33
00:02:09,679 --> 00:02:15,720
cloud bill or something like that." That, you know, you start thinking buying a server isn't that expensive.

34
00:02:14,000 --> 00:02:19,200
Let me see Let me look at the economics of that and and maybe become a

35
00:02:17,040 --> 00:02:23,520
colocation customer. Colocation data centers help an organization save money

36
00:02:21,520 --> 00:02:27,520
while allowing them to keep control over their gear. That means, though, that the

37
00:02:25,959 --> 00:02:31,800
customer is still responsible for configuring, maintaining, and upgrading

38
00:02:29,560 --> 00:02:36,280
their own hardware and software. Now, if that sounds icky, you may prefer the

39
00:02:33,959 --> 00:02:41,760
next level, managed services data centers. In these, a service provider,

40
00:02:38,800 --> 00:02:45,480
like Equinix or OVH, not only houses your rack, but also handles the

41
00:02:43,520 --> 00:02:50,400
day-to-day running and upkeep of your infrastructure. But, of course,

42
00:02:47,880 --> 00:02:55,760
that'll cost you. Cloud data centers are the mysterious bad boy of the data

43
00:02:52,760 --> 00:02:58,240
center boy band, remote and occasionally

44
00:02:55,760 --> 00:03:03,800
unavailable for reasons no one fully explains. They're so misunderstood. Who

45
00:03:01,440 --> 00:03:08,160
is he? These data centers are all about giving you computing power over the

46
00:03:05,840 --> 00:03:12,519
internet, handled by enormous providers like Amazon Web Services or Microsoft

47
00:03:10,400 --> 00:03:17,160
Azure. Not only do you not own the computer, you might not even be using a

48
00:03:14,959 --> 00:03:20,160
whole computer. If you think of a cloud data center,

49
00:03:18,519 --> 00:03:25,440
uh there's again, it's a large centralized facility. Uh typically uh

50
00:03:23,440 --> 00:03:28,920
operates 24/7. And it's filled with what I would call

51
00:03:27,120 --> 00:03:34,000
computing infrastructure. Everything from servers to storage systems to sort

52
00:03:31,680 --> 00:03:38,360
of the network equipment. We literally create subsets of of virtual machines in

53
00:03:36,239 --> 00:03:42,480
a process called virtualization. And what that does is it allows, you know,

54
00:03:40,800 --> 00:03:46,840
one computer to act like maybe 10 computers as needed. At the other time,

55
00:03:44,680 --> 00:03:50,440
it allows 10 computers to act like one. And that's the notion of elasticity that

56
00:03:48,760 --> 00:03:55,280
that, you know, we we we we talk about where you know, whole unit can can

57
00:03:52,720 --> 00:03:59,960
operate as one or a subset of a unit can split up uh into a variety of different

58
00:03:57,520 --> 00:04:05,440
components and that allows uh you know, depending on on the ask or or the nature

59
00:04:02,080 --> 00:04:07,280
of um of of what's being processed,

60
00:04:05,440 --> 00:04:11,520
uh it it can it can operate with with with intent. Like Chris mentioned,

61
00:04:09,320 --> 00:04:16,359
elasticity is a cloud data center's secret weapon. It's the system's ability

62
00:04:13,959 --> 00:04:20,799
to adaptively allocate its resources as needed. Think of companies like Netflix,

63
00:04:18,600 --> 00:04:25,120
which mainly uses Amazon's cloud data centers, so they can easily grow or

64
00:04:22,760 --> 00:04:31,560
shrink streaming capacity depending on how many subscribers are watching Love

65
00:04:27,400 --> 00:04:33,240
is Blind, Secret Island, Getaway,

66
00:04:31,560 --> 00:04:37,680
Couples Forever. While cloud data centers can be any

67
00:04:35,680 --> 00:04:42,840
size, they're typically in huge compounds called hyperscale data

68
00:04:39,919 --> 00:04:47,680
centers. These expansive data facilities are used by giants like Google and Meta,

69
00:04:45,680 --> 00:04:52,640
and they're also taking off because of the AI boom. A standard hyperscale data

70
00:04:50,720 --> 00:04:58,800
center typically contains a minimum of 5,000 servers, 500 cabinets, and spans

71
00:04:56,080 --> 00:05:03,240
at least 10,000 square feet or about a quarter acre of floor space. But,

72
00:05:01,520 --> 00:05:07,760
they're still getting larger all the time with Microsoft's latest Fairwater

73
00:05:05,640 --> 00:05:12,160
data center in Wisconsin taking up an enormous 315

74
00:05:09,720 --> 00:05:16,000
acres and counting. Bear in mind though that hyperscale isn't a completely

75
00:05:14,480 --> 00:05:20,600
separate kind of data center, it's a size and design label. You'll hear about

76
00:05:18,680 --> 00:05:24,280
folks talk about hyperscale data centers and you can think about a hyperscale

77
00:05:21,760 --> 00:05:27,800
data center as a hyperscaler operating a data center that's entirely

78
00:05:25,760 --> 00:05:31,280
purpose-built generally for themselves, right? So, one, they'll either

79
00:05:29,320 --> 00:05:35,200
self-build that or they'll kind of uh lease that from a provider, and they

80
00:05:33,120 --> 00:05:39,280
want to have that full dedicated uh region and that could be for their cloud

81
00:05:37,080 --> 00:05:43,360
availability zones, it could be for AI training nowadays, it could be for, you

82
00:05:41,480 --> 00:05:47,080
know, their their SaaS applications that they're delivering to customers. Uh Uh

83
00:05:45,080 --> 00:05:50,840
it could be for ad analytics for instance. I mean, Meta's obviously got

84
00:05:48,920 --> 00:05:54,240
huge amounts of data center capacity, and that's just for them, right? They're

85
00:05:52,280 --> 00:05:59,880
not operating cloud services for anyone else. On the opposite side of the size

86
00:05:57,000 --> 00:06:03,920
spectrum, edge data centers usually have a smaller footprint that fits right in

87
00:06:01,760 --> 00:06:07,280
the thick of a business's work. But today, they're more often thought of as

88
00:06:05,760 --> 00:06:11,000
just data centers that are located closer to where people actually use

89
00:06:09,000 --> 00:06:16,280
their services, like on the edge of a city. This lets things happen faster

90
00:06:13,720 --> 00:06:21,560
with less latency. These kinds of data centers are thriving right now, too. Not

91
00:06:19,120 --> 00:06:26,880
exactly Harry Styles level fame, like hyperscale cloud centers, but definitely

92
00:06:23,840 --> 00:06:28,360
in the Zayn camp. Telecom companies use

93
00:06:26,880 --> 00:06:32,560
them for things like streaming or gaming, but they're also powering

94
00:06:29,919 --> 00:06:36,680
robotics, physical AI, and systems that need real-time processing. What we think

95
00:06:34,760 --> 00:06:40,600
about as as edge is really like kind of like almost a converged edge, which is

96
00:06:38,840 --> 00:06:44,520
close enough that all of the applications, like even things that feel

97
00:06:42,600 --> 00:06:48,880
pretty real-time, like video and and streaming analytics, they can happen

98
00:06:46,400 --> 00:06:52,919
within kind of a 10-ms radius. That's pretty close to as as

99
00:06:51,040 --> 00:06:57,360
as close to the end user as you need to be without actually being on the device

100
00:06:55,040 --> 00:07:01,800
or like on the factory floor or in the vehicle. You can probably sense by now

101
00:06:59,640 --> 00:07:06,080
that these different types can be used in combination with each other, like a

102
00:07:03,880 --> 00:07:11,040
tight three-part harmony. For example, OpenAI generally uses a giant hyperscale

103
00:07:09,040 --> 00:07:16,160
campus full of cloud data centers to power its AI, but they teamed with a

104
00:07:13,360 --> 00:07:20,680
company named Armada to deliver edge AI to remote industrial sites in form

105
00:07:18,280 --> 00:07:25,600
factors as small as a single shipping container. These smaller units are also

106
00:07:22,840 --> 00:07:29,520
known as modular data center. Even within a single data center, there can

107
00:07:27,360 --> 00:07:33,440
still be various types of server racks that allow a single building or campus

108
00:07:31,240 --> 00:07:37,600
to act in many different ways, which brings us to the more niche, perhaps

109
00:07:35,360 --> 00:07:41,160
even buzzwordy types of data centers, which we'll cover right after this

110
00:07:39,000 --> 00:07:44,440
message from our sponsor. War Thunder, it's the comprehensive vehicle combat

111
00:07:43,120 --> 00:07:48,320
game available on PC, mobile, PlayStation, and Xbox. There's thousands

112
00:07:46,760 --> 00:07:53,000
of combat vehicles for you to choose from and pilot on the battlefield. War

113
00:07:51,080 --> 00:07:56,920
Thunder is known for its authenticity and immersion. All of the vehicles you

114
00:07:54,840 --> 00:08:01,200
can take control of are real-life machines all throughout history from

115
00:07:58,920 --> 00:08:04,720
countries all over the world. So, join over 95 million other players for free

116
00:08:03,520 --> 00:08:10,920
today by using our link in the description and we'll even hook you up with a massive bonus pack if you sign up

117
00:08:07,960 --> 00:08:14,400
today. What? Yeah. Hey, here's a couple new kids on the block getting a lot of

118
00:08:12,480 --> 00:08:19,360
attention since AI started shaking things up. High performance, high

119
00:08:16,480 --> 00:08:23,400
density, and sovereign data centers. High performance/high density data

120
00:08:21,400 --> 00:08:27,280
centers draw a lot of energy since they're typically installed in giant

121
00:08:25,240 --> 00:08:31,040
hyperscale sites and pack as much compute as possible in each server

122
00:08:29,440 --> 00:08:35,320
stack. A recent report from the International Energy Agency states that

123
00:08:33,120 --> 00:08:40,440
in the past, a typical data center would consume between 5 and 10 MW of energy.

124
00:08:38,560 --> 00:08:45,960
While the new high density hyperscale projects can push upwards of 100 MW. And

125
00:08:43,880 --> 00:08:51,360
to put that into perspective, a single MW hour is enough energy to supply

126
00:08:48,240 --> 00:08:54,000
roughly 300 homes for a full day. High

127
00:08:51,360 --> 00:08:58,680
density centers are inherently also high power, but not all high power data

128
00:08:56,440 --> 00:09:01,880
centers are high density. If you're confused, it's okay, stick with me here.

129
00:09:00,280 --> 00:09:06,880
Like the boy band references in this video, they're fading fast. But one

130
00:09:04,040 --> 00:09:10,920
example of high power, low density would be something like vintage geographical

131
00:09:08,880 --> 00:09:14,920
weather gear which can draw a ton of power, but the hardware needs to be

132
00:09:12,880 --> 00:09:19,000
spread out since they generate so much heat. Don't expect to hear lots about

133
00:09:17,280 --> 00:09:23,760
these unless they're being retrofitted to pack in modern servers with more

134
00:09:20,800 --> 00:09:27,760
density. Finally, there's sovereign data centers. These are having a bit of a

135
00:09:25,920 --> 00:09:31,640
moment right now and they can take the form of pretty much every type we've

136
00:09:29,760 --> 00:09:35,440
mentioned already. They're popping up as governments and businesses realize that

137
00:09:33,560 --> 00:09:40,240
to control their data, it needs to be stored in a place where their local laws

138
00:09:37,360 --> 00:09:44,480
provide jurisdiction. That said, this can get murky. For instance, US

139
00:09:42,480 --> 00:09:49,440
companies operating in Canadian data centers are not safe from US government

140
00:09:47,040 --> 00:09:54,000
interference. For full sovereignty, you need the data center and the company

141
00:09:51,000 --> 00:09:56,320
operating under the same flag. So, those

142
00:09:54,000 --> 00:10:00,240
are the types, such as they are, but that's not the only way to categorize

143
00:09:57,960 --> 00:10:05,000
data centers. Another is the Uptime Institute's tier system. It ranks data

144
00:10:02,760 --> 00:10:09,240
centers from tier one at the bottom to tier four at the top, based on things

145
00:10:07,040 --> 00:10:12,640
like uptime, which means how long the system has been running without

146
00:10:10,280 --> 00:10:16,680
interruptions, redundancy, and how well they handle failures. And basically, it

147
00:10:15,080 --> 00:10:20,640
looks at four tiers. It looks at, you know, things like your availability,

148
00:10:19,320 --> 00:10:24,560
making sure that you've got available power that's consistent to process up to

149
00:10:22,920 --> 00:10:28,880
the maximum that you're you're allowed to process, cuz typically there's peaks

150
00:10:26,520 --> 00:10:32,320
and valleys in in consumption as well. It looks at redundancy, which means if

151
00:10:30,480 --> 00:10:35,360
one of these things fail, if your power fails, then what's your backup? If your

152
00:10:34,080 --> 00:10:39,160
cooling system fails, what's your backup? If, you know, your

153
00:10:37,120 --> 00:10:42,320
part of your IT infrastructure or your or your chips and servers go down,

154
00:10:40,760 --> 00:10:46,040
what's your backup? It also looks at fault fault tolerance.

155
00:10:44,320 --> 00:10:49,839
So, things like, you know, how often does do these things have

156
00:10:48,320 --> 00:10:53,440
errors, etc. You know, what's what's in the manufacturer's specs and and how are

157
00:10:51,800 --> 00:10:57,720
they performing? In some ways, I would call it a bit of an audit. What Chris

158
00:10:55,520 --> 00:11:02,080
told us is basically the checklist data centers live and die by. Power, backups,

159
00:11:00,640 --> 00:11:06,040
backups for the backups, and then backups for those, too. If you want to

160
00:11:04,320 --> 00:11:10,160
get a little more up close and personal with a data center, check out this video

161
00:11:07,680 --> 00:11:14,040
of Linus touring one, and drop a comment if you can guess what type it is.