1
00:00:00,594 --> 00:00:02,005
Welcome to Let's Get Digital.

2
00:00:02,005 --> 00:00:05,976
I'm Carrie Charles and I am so glad that you are here.

3
00:00:05,976 --> 00:00:08,938
Today's guest is Chris Sharp.

4
00:00:08,938 --> 00:00:12,839
He is the Chief Technology Officer of Digital Realty.

5
00:00:13,119 --> 00:00:23,884
Digital Realty operates the world's largest data center platform with over 300 data
centers spanning 55 metros across six continents.

6
00:00:23,884 --> 00:00:29,066
Now, Chris has been on this show before and he has more than two decades of

7
00:00:29,066 --> 00:00:33,048
At the center of cloud interconnection data center technology.

8
00:00:33,048 --> 00:00:34,869
This guy's brilliant.

9
00:00:34,869 --> 00:00:37,991
I follow him every single day on LinkedIn.

10
00:00:37,991 --> 00:00:40,272
You are gonna love this episode.

11
00:00:40,612 --> 00:00:47,596
And he was on Let's Get Digital, I think it was maybe a year ago, maybe a little bit more,
but he's back again.

12
00:00:47,596 --> 00:00:50,026
Chris, thank you for joining me today.

13
00:00:50,026 --> 00:00:52,191
I love our conversations.

14
00:00:52,191 --> 00:00:53,131
No, absolutely appreciate it.

15
00:00:53,131 --> 00:00:54,412
Appreciate the kind words.

16
00:00:54,412 --> 00:00:58,574
Um always good to chat, you know, and just kinda transfer knowledge.

17
00:00:58,574 --> 00:01:02,545
And, you know, I think one of the things when you say two decades like that, I'm like, my
gosh.

18
00:01:02,545 --> 00:01:09,188
It's not only like I feel old, but like what's changed in those two decades and how it
just gets quicker and quicker in the changes.

19
00:01:09,188 --> 00:01:11,008
But absolutely happy to be here.

20
00:01:11,008 --> 00:01:12,541
So true, so true.

21
00:01:12,541 --> 00:01:14,484
All right, let's dive right into it.

22
00:01:14,484 --> 00:01:24,561
So let's cut all the hype about AI and tell me what is the digital infrastructure story
right now.

23
00:01:24,725 --> 00:01:26,676
Yeah, so there is a lot of hype.

24
00:01:26,676 --> 00:01:27,957
Um, unfortunately.

25
00:01:27,957 --> 00:01:29,857
Um I don't know how it it depends on how you look at it.

26
00:01:29,857 --> 00:01:35,260
Some people enjoy the hype, but I always try to break it down to the basic principles or
first principles.

27
00:01:35,260 --> 00:01:38,742
Um, what you're seeing today, it's a supply chain story.

28
00:01:38,742 --> 00:01:43,254
And I think one of the things that we've always watched is like it's electrons, equipment,
and then expertise.

29
00:01:43,254 --> 00:01:54,259
And that last one is often overlooked, but you know, we went from kilowatts to gigawatts
in the last decade and nobody really has a true appreciation for what that means and what

30
00:01:54,259 --> 00:01:54,671
it takes.

31
00:01:54,671 --> 00:01:59,555
To kind of overcome some of the challenges we're seeing today that are in the headlines
everywhere now.

32
00:01:59,555 --> 00:02:04,469
Like I think it's it's funny that you know I've been doing this, as you said, for for a
good bit of time.

33
00:02:04,469 --> 00:02:10,403
But now when people ask me what I do, I I tend more towards the technology side and not
the data center side.

34
00:02:10,403 --> 00:02:21,051
But I I get the sentiment because, you know, the grid, the talent, you know, potential
disruption around like jobs and and how that's gonna come together, but you know, the

35
00:02:21,051 --> 00:02:24,985
pipeline of talent didn't come with us as we evolved.

36
00:02:24,985 --> 00:02:28,684
In that supply chain story of the kilowatts, the gigawatts, and where we're at.

37
00:02:28,684 --> 00:02:31,520
But I would tell you, stripping that hype down.

38
00:02:31,848 --> 00:02:36,520
Time to market and time to power, more importantly, is the product, right?

39
00:02:36,520 --> 00:02:43,861
And so working within the communities and making sure you have the right designs and just
really executing in a frame.

40
00:02:43,861 --> 00:02:51,167
And I'll I'll kind of just start with the data center piece and then work in the the tech
piece, but making sure you're doing the right things with the right resources, right?

41
00:02:51,167 --> 00:02:54,148
Like I think it's awesome that water is at the forefront.

42
00:02:54,148 --> 00:03:01,842
I think it's awesome that efficiency of power and you know how do you work with utilities
and making sure that you know everything is balanced out.

43
00:03:01,842 --> 00:03:04,864
In the longer run um is extremely important to us.

44
00:03:04,864 --> 00:03:07,776
And you can read that in our ESG reports.

45
00:03:07,776 --> 00:03:15,380
I don't think that's the the purpose of this talk, but more around what's actually
happening in the AI realm.

46
00:03:15,400 --> 00:03:16,251
It's amazing, right?

47
00:03:16,251 --> 00:03:24,005
As a technologist, um watching some of these open weights models, like even in the last
couple of days, it's changed tremendously, right?

48
00:03:24,005 --> 00:03:24,846
And

49
00:03:25,076 --> 00:03:35,311
What that means to me and what I think that means to most of us is that having the ability
to run private AI and these capabilities and and being able to save a material amount of

50
00:03:35,311 --> 00:03:43,935
money and also have a lot of control and security associated with your AI kind of
evolution, not only for you as personally and also as a as a company.

51
00:03:43,935 --> 00:03:53,549
Like I actually run an open weights model on my Spark right here from NVIDIA, that you
know, the these savings and ROI to tokens are phenomenal.

52
00:03:53,549 --> 00:03:55,473
And so having that ability, I think.

53
00:03:55,473 --> 00:03:59,236
is something that I think more hype needs to be pulled away from it, right?

54
00:03:59,236 --> 00:04:02,589
Where you need to enable it like the cloud, right?

55
00:04:02,589 --> 00:04:12,056
Like people I think overcomplicate a lot of things, and that cloud came about where
everything was public, and like everybody went to these public clouds, and you know the

56
00:04:12,056 --> 00:04:14,948
data centers were over and the industry was over and the clouds owned it.

57
00:04:14,948 --> 00:04:16,580
That didn't stay for very long, right?

58
00:04:16,580 --> 00:04:20,853
Like the CFO woke up and saw the bill and was like, whoa, whoa, whoa, let's let's rethink
this.

59
00:04:20,853 --> 00:04:22,394
And so hybrid came into play.

60
00:04:22,394 --> 00:04:23,495
That was very hard.

61
00:04:23,495 --> 00:04:24,856
Like people like moved.

62
00:04:24,856 --> 00:04:28,308
Miss and like reminisce around there's a lot of work to do hybrid.

63
00:04:28,308 --> 00:04:36,013
I built my career around that and doing private interconnectivity, building the first
private kind of on-ramp into a major public cloud and building the first exchange to

64
00:04:36,013 --> 00:04:37,945
remove the interconnectivity complexity.

65
00:04:37,945 --> 00:04:40,987
But the same things happen with AI, right?

66
00:04:40,987 --> 00:04:49,462
You can go into these frontier models, you want to build a bit of private, and then the
next frontier I think is beyond that is it's gonna be just like with cloud, hybrid

67
00:04:49,462 --> 00:04:50,323
multi-cloud, right?

68
00:04:50,323 --> 00:04:55,196
You're gonna have hybrid multi-AI, you're gonna have a mixture of experts of models, and
you're gonna pick and

69
00:04:55,396 --> 00:04:57,327
how you're gonna get your tokens back out of that.

70
00:04:57,327 --> 00:04:59,246
But I would say that's a critical one.

71
00:04:59,246 --> 00:05:03,378
And then the last one is we're not gonna be talking about tokens.

72
00:05:03,399 --> 00:05:03,659
Right.

73
00:05:03,659 --> 00:05:08,580
I think people are really watching this hype token maxing and the expense to it.

74
00:05:08,580 --> 00:05:14,612
What you're gonna see happen probably before the end of this year pretty materially is the
value outcome.

75
00:05:14,612 --> 00:05:22,590
So what is the value you're guaranteed are getting out of your AI implementation and how
you're utilizing the tools today and the value you get out of it?

76
00:05:22,590 --> 00:05:23,561
That's everything to me.

77
00:05:23,561 --> 00:05:31,398
And I spend more of my time looking at certain companies and enabling them to guarantee
that value and that return associated with it.

78
00:05:31,763 --> 00:05:34,683
So uh the equipment behind you, right?

79
00:05:34,683 --> 00:05:37,631
Is that what we're all gonna have someday?

80
00:05:37,771 --> 00:05:39,202
Yeah, I mean absolutely.

81
00:05:39,202 --> 00:05:47,651
And and it's gonna be maybe even a smaller form factor than this, because like, I don't
know, you saw the Apple news with the new uh M6 coming out.

82
00:05:47,651 --> 00:05:57,289
I I'm a technologist at heart, so I watch the glow, but Xeomi just came out with another
one that I I I think it's phenomenal with their AI kind of framework workstation that sits

83
00:05:57,289 --> 00:06:01,401
on your desktop that can, you know, run these models rather efficiently.

84
00:06:01,401 --> 00:06:03,422
It will absolutely happen because

85
00:06:03,584 --> 00:06:08,136
Tokens, if you're just using a frontier model for everything, get very expensive.

86
00:06:08,136 --> 00:06:12,039
And yeah, yeah, and the amount of tokens we're using.

87
00:06:12,039 --> 00:06:20,823
Like I don't know how many you're using, Carrie, but like why this thing pays for itself
is we're doing, you know, hundreds of millions of tokens in a very short time frame.

88
00:06:20,823 --> 00:06:25,601
When you look at the ROI of renting those tokens, it pays for itself rather quickly.

89
00:06:25,601 --> 00:06:26,923
So true, so true.

90
00:06:26,923 --> 00:06:32,849
So I I've heard you I've heard you say so many things and I call them criticisms because I
uh I love them so much.

91
00:06:32,849 --> 00:06:39,876
But one thing I heard you say is uh that AI automates tasks, not people.

92
00:06:39,956 --> 00:06:48,084
And you know, there's again more hype out there about AI o you know, taking jobs
automating people, but tell me a little bit about what you mean by that.

93
00:06:48,202 --> 00:06:54,615
Yeah, so it's one of the things that I I I really watch the humanity element of this,
right?

94
00:06:54,615 --> 00:06:59,522
And the impact humanity and in in what it does, because there it's definitely disruptive.

95
00:06:59,522 --> 00:07:01,928
But like never, never step back from that.

96
00:07:01,928 --> 00:07:03,919
But you want to be on the right side of disruption.

97
00:07:03,919 --> 00:07:08,992
And so I always try to break it down into the most simplistic components or first
principles, right?

98
00:07:08,992 --> 00:07:18,216
Where like the future really belongs to organizations that combine that human expertise to
that overall technological tool.

99
00:07:18,216 --> 00:07:31,419
tools are becoming increasingly powerful, but how you demystify and how you let people
apply it matters even more than the actual technology itself because what we're seeing is

100
00:07:31,419 --> 00:07:40,271
a lot of companies and a lot of individuals, because it's not just about your professional
life, it's your personal life as well, really understanding how these tools run.

101
00:07:40,271 --> 00:07:42,562
And I think this is an important factor.

102
00:07:42,999 --> 00:07:44,360
Really go slow.

103
00:07:44,360 --> 00:07:44,600
Right?

104
00:07:44,600 --> 00:07:52,593
Like if you didn't learn anything from OpenClaw where people were just throwing everything
in it and whatever it just ran rampant with their credit card numbers and it was just

105
00:07:52,593 --> 00:07:53,334
doing crazy things.

106
00:07:53,334 --> 00:07:56,305
Like be mindful of what's happening with your data.

107
00:07:56,305 --> 00:08:06,739
Be mindful of the amount of control you give to these capabilities and just watch these
tools and watch why some of these tools are afforded to you because they're mining your

108
00:08:06,739 --> 00:08:06,999
data.

109
00:08:06,999 --> 00:08:10,531
They want specific uh understanding of your capabilities.

110
00:08:10,531 --> 00:08:12,802
But I'll tell you, like, that's the

111
00:08:12,802 --> 00:08:16,294
future of what I think a successful individual is going to be.

112
00:08:16,294 --> 00:08:23,198
I mean being the father of two small children, that we we we we do a home education w with
a private educator, but a

113
00:08:23,198 --> 00:08:30,401
core component is like what happens with your data and like is like how does your data get
transacted and the how that's gonna evolve over time.

114
00:08:30,401 --> 00:08:42,067
But yeah, these tools today and and even with agentic agents, like some jobs, some things,
kinda, but there's always a human in the loop on the backside to manage the agents.

115
00:08:42,067 --> 00:08:42,307
Right.

116
00:08:42,307 --> 00:08:51,781
And I think that's the future of work is how you manage these agents or manage these
tools, these agentic capabilities, and you harness them for that value.

117
00:08:51,781 --> 00:08:53,272
Like I was saying earlier,

118
00:08:53,272 --> 00:08:57,668
Value being the returning factor is everything, right?

119
00:08:57,668 --> 00:09:03,335
And so how that comes to market I think is is something that excites me, but also makes me
nervous as well.

120
00:09:03,335 --> 00:09:07,089
And I believe me, I'm pretty steep and deep and and get the

121
00:09:07,314 --> 00:09:18,212
Opportunity to talk with some of the brilliant, like some of the most like forefront, like
leading thinkers of AI, both from these frontier models, also from the private, from the

122
00:09:18,212 --> 00:09:19,422
hardware.

123
00:09:19,794 --> 00:09:27,818
It is absolutely amazing what you're able to harness for not only your own benefit,
because I don't want to say benefit, but for your own betterment.

124
00:09:27,818 --> 00:09:30,190
Like it really makes you better if you do it correctly.

125
00:09:30,190 --> 00:09:32,972
But that's something Kara I've always kind of watched in the market.

126
00:09:32,972 --> 00:09:36,254
But yeah, people are a core component of this going forward.

127
00:09:36,597 --> 00:09:37,557
I couldn't agree more.

128
00:09:37,557 --> 00:09:44,180
I am now using a team of agents for my productivity and my team's productivity.

129
00:09:44,180 --> 00:09:47,952
And Chris, it's it's not replacing anyone.

130
00:09:47,952 --> 00:09:49,183
It's just making us better.

131
00:09:49,183 --> 00:09:53,024
And I'm telling you something, those agents need me.

132
00:09:53,024 --> 00:09:54,665
They need us in the loop.

133
00:09:54,665 --> 00:09:57,176
So I agree with everything you're saying.

134
00:09:57,176 --> 00:10:06,010
Um, speaking about people, I want to go a little bit deeper into this subject because I
also heard you say something that I absolutely loved, and it was talent is the new

135
00:10:06,010 --> 00:10:06,960
megawatt.

136
00:10:07,412 --> 00:10:18,228
And how do you feel about what's going out you know going on right now in that's in our
space with talent, with workforce shortages, with the desperate need for uh skilled labor?

137
00:10:18,228 --> 00:10:20,104
Um talk to me.

138
00:10:20,104 --> 00:10:20,664
No, I love it.

139
00:10:20,664 --> 00:10:21,044
I love it.

140
00:10:21,044 --> 00:10:25,647
And and you know, I don't know, sometimes I get in trouble with that statement that like
talent is the new megawatt.

141
00:10:25,647 --> 00:10:31,639
But like I'll go one step further because I think I I I love to be able to speak a bit
freely on on your show here.

142
00:10:31,639 --> 00:10:37,622
It's like one of the things I've always looked at is we industrialized electrons faster
than we industrialized expertise.

143
00:10:37,622 --> 00:10:39,003
It just happened, right?

144
00:10:39,003 --> 00:10:46,716
Because of the demand and how that came about, but like really what we're watching is like
talent is the new megawatt, but we really need a gigawatt, right?

145
00:10:46,716 --> 00:10:50,078
Like that's where I was really headed in a lot of that position, is that there's

146
00:10:50,078 --> 00:10:51,589
Scaling mismatch, right?

147
00:10:51,589 --> 00:11:00,555
Just like we saw in the industry for kilowatts to mega to megawatts to gigawatts, roughly
in a decade, like the human pipeline stayed pretty flat.

148
00:11:00,555 --> 00:11:04,858
And a lot of the training that's afforded to these, like, I don't know, you have to be
self-taught.

149
00:11:04,858 --> 00:11:09,171
And like one of the things I always try to impress upon everybody is be a critical
thinker.

150
00:11:09,171 --> 00:11:14,084
Like ask the right, like spend more time on the problem to then be asking the right
questions.

151
00:11:14,084 --> 00:11:17,927
And so what that means to me in this landscape is there's no one class.

152
00:11:17,927 --> 00:11:20,008
Like I love some of the exposure.

153
00:11:20,008 --> 00:11:24,881
I get from like Stanford, and I've had the opportunity to speak with a lot of professors
there.

154
00:11:24,881 --> 00:11:34,368
And what's amazing to me is even their curriculum is so open to kind of interpretation
because of the rapid innovation that's happening.

155
00:11:34,368 --> 00:11:42,843
And like what it really means is read a lot, make sure you like really take the time to
understand what the benefits are going to be.

156
00:11:42,843 --> 00:11:49,841
And it's not just about workforce constraint as in thought jobs or thought leadership kind
of jobs, but also.

157
00:11:49,841 --> 00:11:58,538
just the trades as well, which I think are important to a lot of individuals and the
community enablement as well, where a lot of these facilities that come online, and we've

158
00:11:58,538 --> 00:11:59,719
been doing this for a long time.

159
00:11:59,719 --> 00:12:03,071
I've been at digital 12 years, the company's been over 20 years.

160
00:12:03,071 --> 00:12:12,218
We spend a lot of time with master planning, but also working with the community so that
you know they understand the benefits associated with a lot of these capabilities coming

161
00:12:12,218 --> 00:12:12,769
to market.

162
00:12:12,769 --> 00:12:20,124
But you know, I I'll leave you with this is that, you know, a single resume or a couple of
resumes

163
00:12:20,124 --> 00:12:23,196
Can gate a gigawatt of capability, right?

164
00:12:23,196 --> 00:12:32,172
Because the human expertise and heritage of expertise to prosecute this and make it come
to market is absolutely imperative.

165
00:12:32,172 --> 00:12:34,044
And that's the constraint we see today, right?

166
00:12:34,044 --> 00:12:39,077
Like I think we all read the headlines with the capital allocations that are coming from
around the world.

167
00:12:39,117 --> 00:12:46,862
That's rather well maintained, but then it's the individuals that sit there and watch the
complexities of these things coming to market.

168
00:12:46,942 --> 00:12:49,714
And I'll give you this last piece: the individuals.

169
00:12:49,714 --> 00:12:52,036
That can simplify the complexity.

170
00:12:52,036 --> 00:12:55,128
Like I often tell people, tech for tech's sake is useless, right?

171
00:12:55,128 --> 00:12:58,310
And anybody buying a science project, they better be a scientist.

172
00:12:58,310 --> 00:13:05,154
And so one of the things I spent a lot of my life is just simplifying and making sure you
have repeatable value-based outcomes.

173
00:13:05,154 --> 00:13:08,577
So when you ask Carrier, like, are we all gonna have these on our desks?

174
00:13:08,577 --> 00:13:13,980
Not today, because it takes me a lot to actually build it, maintain it, it breaks all the
time, and it's amazing.

175
00:13:13,980 --> 00:13:19,944
But we you will see the formation of capabilities and product.

176
00:13:19,944 --> 00:13:23,388
that it will be embedded with us around us.

177
00:13:23,388 --> 00:13:26,851
And I'll say it again, like I just be mindful of it.

178
00:13:26,851 --> 00:13:28,973
You want to have visibility of control.

179
00:13:28,973 --> 00:13:34,067
You don't want to give it too much control and what it's going to be able to do for you
because you can wake up and have some bad surprises.

180
00:13:34,067 --> 00:13:34,817
So

181
00:13:35,329 --> 00:13:41,482
Did I read correctly and digital realty hires people without data center experience?

182
00:13:41,631 --> 00:13:43,222
You know, I I I love that you asked that.

183
00:13:43,222 --> 00:13:55,978
And I know we had a a pre-call on this one, but like I'll I'll go one step further that I
I'm just so proud of our our product team and our HR team here is that we absolutely try

184
00:13:55,978 --> 00:13:58,249
to find individuals and this is how we talk about internally.

185
00:13:58,249 --> 00:13:59,599
Can they do the math?

186
00:13:59,615 --> 00:13:59,805
Right?

187
00:13:59,805 --> 00:14:06,458
Like, can you like because you can't fake the technical expertise, but that that math is
not just directly applied to data centers.

188
00:14:06,458 --> 00:14:15,602
So we absolutely, you know, scour schools to where we actually have now provided housing
in critical markets to where we'll educate them.

189
00:14:15,602 --> 00:14:20,294
And so I I joke that it's like a like all of these reality shows, but it's not.

190
00:14:20,294 --> 00:14:29,528
And I just hope in my team that like all of these young college kids are living together
for a one-month-ish kind of program to get in.

191
00:14:29,528 --> 00:14:40,796
Immersed into taking their traditional mathematic type skills and applying that to a more
data center operational understanding of what's required inside of the work.

192
00:14:40,796 --> 00:14:45,478
And then they can kind of pick and choose across our portfolio of where they want to
evolve their career.

193
00:14:45,478 --> 00:14:48,761
And that's been something that's been very rewarding, very successful to us.

194
00:14:48,761 --> 00:14:56,526
But yeah, you know, skilled operational expertise is as much as important as the silicon
that goes inside of it.

195
00:14:56,526 --> 00:14:59,448
But like what I'm really watching is we need to really.

196
00:14:59,448 --> 00:15:10,997
really start defining like the challenges that exist for human capital and the pace of the
build-out that the human kind of talent wars and skill sets have to also match that.

197
00:15:11,488 --> 00:15:20,409
When you were on the show last time, we talked a lot about training and I wanna talk more
about inference now, because I guess we're entering the inference era.

198
00:15:20,409 --> 00:15:22,291
Um, what has changed?

199
00:15:22,291 --> 00:15:24,823
Why is inference so important right now?

200
00:15:24,938 --> 00:15:26,958
Yeah, so like training ends.

201
00:15:26,979 --> 00:15:28,439
Inference never does.

202
00:15:28,439 --> 00:15:33,360
Like so, like I think people understand it probably in a general sense that training is
kind of episodic.

203
00:15:33,360 --> 00:15:39,002
You make a run, you train the model, and then you take the ex like the outcome of that,
and then that becomes inference.

204
00:15:39,002 --> 00:15:47,924
But I'm also a bit frustrated in the market, which is why I always love that you afford me
these opportunities, is that training's wildly over over kind of simplified.

205
00:15:47,924 --> 00:15:55,275
And that really what we're looking at with with with with I'm sorry, inference is overly
utilized as well, and that

206
00:15:55,275 --> 00:15:58,546
You just want to watch how inference is applied.

207
00:15:58,546 --> 00:16:05,540
And that formation of what inference is looking at is the monetization of the consumption
of that capability.

208
00:16:05,540 --> 00:16:15,864
And that's why it's an ongoing scenario that, quite frankly, we've surpassed the amount of
tokens produced for training and the inference because so many more services, so many more

209
00:16:15,864 --> 00:16:17,645
capabilities have come to market.

210
00:16:17,645 --> 00:16:25,348
So we're entering this kind of utility phase of AI where these inference capabilities are
everything, but

211
00:16:25,348 --> 00:16:29,591
Really breaking it apart, and I'll push on this a little bit, and this is where I always
get the negative feedback.

212
00:16:29,591 --> 00:16:32,523
But like inference, there's a pre-filled and a decode.

213
00:16:32,523 --> 00:16:39,157
So there's two kind of scenarios, like when you kind of talk to it, and then how it
actually answers your questions and gives you that response back.

214
00:16:39,157 --> 00:16:40,938
There's a lot of dynamic to that.

215
00:16:40,938 --> 00:16:44,831
And the dynamics come in infrastructure required, right?

216
00:16:44,831 --> 00:16:51,686
Where you're starting to see a lot of individuals like it just showed up in some of the
frontier models where you can click to go faster, right?

217
00:16:51,686 --> 00:16:55,188
Now, you know, because I've always believed

218
00:16:55,188 --> 00:16:59,411
In the distribution of infrastructure globally, so that we can all benefit from it.

219
00:16:59,411 --> 00:17:03,293
But you're gonna see more distribution of inference capabilities.

220
00:17:03,554 --> 00:17:06,776
Last I would say is like I did this, right?

221
00:17:06,776 --> 00:17:10,578
Like we lived this in the industry was cloud.

222
00:17:10,578 --> 00:17:15,108
People forget rather quickly that we consumed a lot of that cloud over the internet.

223
00:17:15,108 --> 00:17:16,602
It was just straight transit.

224
00:17:16,602 --> 00:17:21,486
So no class of service, quality service, private connectivity has to come into play for
inference.

225
00:17:21,486 --> 00:17:24,788
So you have business class capabilities, high-end security.

226
00:17:24,788 --> 00:17:25,088
That's

227
00:17:25,088 --> 00:17:26,779
Where we've brought our service fabric to market.

228
00:17:26,779 --> 00:17:36,245
I won't bore everybody about that, but you can read about that on your own and the
programmability of service fabric because the complexity of AI is you want to orchestrate

229
00:17:36,245 --> 00:17:45,011
multiple either frontier models, you want to multiple Neo clouds, compute, agentic
capabilities, all of that's done through software.

230
00:17:45,011 --> 00:17:49,845
And I think that's where this, what's changed is the consumption of it.

231
00:17:49,845 --> 00:17:54,961
Um, and we're entering an era of, again, it's it's not the cost per.

232
00:17:54,961 --> 00:18:00,474
It's it's it's not the cost per unit, it's it's the actual uptime and the value that's
actually coming out the other side.

233
00:18:00,474 --> 00:18:08,867
And that's why, you know, I don't I don't know, I I maybe overindex on this, but like we
all talked about IaaS, infrastructure as a service for a long time.

234
00:18:08,867 --> 00:18:13,570
Then SAS came along, which just broadened the target addressable market tremendously.

235
00:18:13,570 --> 00:18:21,763
We're in that same era here with a gentic and some of these other capabilities, which at
the end of the day, the most simplistic way to call it is like, it's inference, but that

236
00:18:21,763 --> 00:18:23,714
inference has a lot of

237
00:18:24,138 --> 00:18:34,104
complexity associated with which will be simplified for the broader masses, but I think a
lot of your listeners want to understand, dig into how inference is going to mature and

238
00:18:34,104 --> 00:18:38,267
how customers are going to consume that value and stay aligned to it.

239
00:18:38,267 --> 00:18:42,389
Because as with most tools, like this is what I used to always love in my job.

240
00:18:42,389 --> 00:18:43,770
Um

241
00:18:44,170 --> 00:18:46,512
When I would go in and we would talk about SaaS.

242
00:18:46,512 --> 00:18:49,404
And like, you know, I'm running my business on these two SaaS applications.

243
00:18:49,404 --> 00:18:52,736
And then it went, I'm running my business on these 30 SaaS applications.

244
00:18:52,736 --> 00:18:59,321
I think, Carrie, you were saying you have an agent, it may be a GENTIC, it may have that
bi-directional piece and a bit of logic in itself.

245
00:18:59,321 --> 00:19:04,184
You're gonna have hundreds, if not thousands, of those agents, and you'll have agents
managing those agents.

246
00:19:04,184 --> 00:19:12,429
So that's where the world is headed, and you have to be able to harness all of those tools
and those capabilities together to the best outcome for your business or yourself.

247
00:19:13,243 --> 00:19:22,007
So fast forward let's say three years, how different is the is the data center going to
look from what we're building today?

248
00:19:22,007 --> 00:19:26,083
Like the like if I walked into a data center, what would be different?

249
00:19:26,346 --> 00:19:32,671
I think the data center becomes it's i the data center stops being real estate.

250
00:19:32,671 --> 00:19:36,603
Right, and and it really becomes an industrial machine for manufacturing.

251
00:19:36,603 --> 00:19:45,567
Like when you see, and this is where I think it depends on your exposure to the world and
how items actually get manufactured and distributed, right?

252
00:19:45,567 --> 00:19:51,219
Because there's different types of shipping capabilities, and there's local distribution
for consumption.

253
00:19:51,219 --> 00:20:00,073
And like what we're really watching is and what will change inside of the data center,
it's its most basic principle is that it's a manufacturing of intelligence, and that

254
00:20:00,073 --> 00:20:02,644
intelligence is getting cheaper and how you utilize.

255
00:20:02,644 --> 00:20:09,533
It is going to be everything, but if you go inside of the data center, I mean there's
there's a lot to change, right?

256
00:20:09,533 --> 00:20:19,424
Where looking forward, density rewrites the entire building, and I'll go one step even
higher in the hierarchy is

257
00:20:19,486 --> 00:20:21,247
The chipset sets the design.

258
00:20:21,247 --> 00:20:30,191
So with whatever chipset you're going to bring to market are quite frankly, because of the
supply chain statement I made earlier, that you're afforded to be able to get access to,

259
00:20:30,191 --> 00:20:38,775
because there's always a bit of a constraint of how many chips can you get and at what
what vintage, because a lot of people are using older vintages, but that chipset really

260
00:20:38,775 --> 00:20:42,637
dictates what you have to do inside of the actual facility.

261
00:20:42,637 --> 00:20:49,430
So this is where you get to rack densities, 130 kilowatts today, that's headed to 600.

262
00:20:49,430 --> 00:20:53,322
To a megawatt, that changes a lot within the facility.

263
00:20:53,322 --> 00:21:01,336
Quite frankly, power distribution is a in the power architecture is a is a very, I would
say, rapidly evolving component of that.

264
00:21:01,336 --> 00:21:05,528
If you if you want to go down a rat hole, you can talk about 800 volt DC distribution.

265
00:21:05,528 --> 00:21:10,911
That's gonna happen because the amount of electrical equipment to densify those individual
racks.

266
00:21:10,911 --> 00:21:12,892
And so we work a lot with our partners.

267
00:21:12,892 --> 00:21:15,333
I spend, I spend more time

268
00:21:16,500 --> 00:21:18,581
I try to spend as much time as I can with NVIDIA.

269
00:21:18,581 --> 00:21:19,832
I always say more time than I should.

270
00:21:19,832 --> 00:21:21,342
It depends on who you who you're asking.

271
00:21:21,342 --> 00:21:24,062
If you're asking my boss, he's like, you're in San Jose again.

272
00:21:24,062 --> 00:21:25,494
I'm like, yes, I am.

273
00:21:25,494 --> 00:21:33,248
Because if you solve for some of their higher end kind of Vera Rubin type of complexities,
everything else falls within that.

274
00:21:33,248 --> 00:21:35,109
And so they're setting a standard for us.

275
00:21:35,109 --> 00:21:45,984
And so, you know, I'll I'll also emphasize again, we design for, and I'm gonna use my own
words now just because it's the state of play today, we design for tokens.

276
00:21:45,984 --> 00:21:46,407
And you

277
00:21:46,407 --> 00:21:50,110
You you'll see will design for value, not just space and power.

278
00:21:50,110 --> 00:21:55,264
And so that's where it's the collective of the capability that's extremely important,
right?

279
00:21:55,264 --> 00:21:56,605
To a lot of our customers as well.

280
00:21:56,605 --> 00:22:05,722
That it's not just about one compute block, it's about the conglomerate of multiple
compute blocks that may be in our facility, maybe in other facilities, because we're very

281
00:22:05,722 --> 00:22:06,312
open.

282
00:22:06,312 --> 00:22:07,765
And that's very unique in this industry.

283
00:22:07,765 --> 00:22:11,256
There's other individuals that talk about being a platform, they are not.

284
00:22:11,256 --> 00:22:16,310
And so we are very much open where we allow the interconnectivity beyond.

285
00:22:16,310 --> 00:22:22,362
the four walls of our data centers and enable that globally so that you get the full
benefit of it, right?

286
00:22:22,362 --> 00:22:31,425
And that's where you're seeing these true platform capabilities come to market where it's
not pockets of infrastructure, it's just transparent to that in consumption of that

287
00:22:31,425 --> 00:22:32,606
capability.

288
00:22:33,884 --> 00:22:36,597
So I wanna talk about energy and power.

289
00:22:37,078 --> 00:22:47,109
And I don't know, this may be a a crazy something crazy to say, but are data centers
essentially becoming energy companies, whether they want to or not?

290
00:22:47,158 --> 00:22:51,121
Yeah, I I would say we're we're definitely not becoming an energy company.

291
00:22:51,121 --> 00:22:53,162
We're becoming energy fluent.

292
00:22:53,162 --> 00:23:04,710
Like I always really try to land that with a lot of individuals that you know we treat
kind of power as the product that we we we really engineer.

293
00:23:04,710 --> 00:23:05,550
And

294
00:23:05,844 --> 00:23:14,818
You know, we might, you know, get that through utility, you know, we might wait on the
utilities and do some bridge, but you have to become energy fluent because the demand of

295
00:23:14,818 --> 00:23:18,690
our customers, and quite frankly, customers can mean multiple things.

296
00:23:18,690 --> 00:23:28,365
I don't I don't think people think about the true pr like spectrum that it's not just the
frontier models, the neo clouds, the hyperscalers that we service, but it's also the

297
00:23:28,365 --> 00:23:28,805
enterprise.

298
00:23:28,805 --> 00:23:31,806
And we're very unique in the market that we have a comprehensive portfolio.

299
00:23:31,806 --> 00:23:36,121
I'll sell you a rack, I'll sell you a cage, I'll sell you a data hall, I'll sell you a
gig.

300
00:23:36,121 --> 00:23:42,927
So, like, that's very unique because I always watch both sides of that because there's
people who are selling capability and there's people consuming it.

301
00:23:42,927 --> 00:23:48,131
So, the customers are also on the enterprise and even consumer side.

302
00:23:48,131 --> 00:23:50,613
How are they going to be serviced?

303
00:23:50,613 --> 00:23:54,746
And so you have to be very fluent in power and do the right things, right?

304
00:23:54,746 --> 00:24:03,002
And I think this is where the news is filled with, quite frankly, some artificially like
false information and misinformation.

305
00:24:03,082 --> 00:24:05,644
And there's more towards like, how do you balance

306
00:24:05,644 --> 00:24:14,540
That how do you make sure you're not overreacting to something and really aligning the
power to the demand that you have and looking in that longer-term horizon to be as

307
00:24:14,540 --> 00:24:16,231
efficient as possible, right?

308
00:24:16,231 --> 00:24:26,338
Because the efficiencies around the like a lot of our ESG goals with our carbon neutral
and leveraging water, because a lot of water is associated with a lot of the electrical

309
00:24:26,338 --> 00:24:27,889
electrification of the grids and stuff.

310
00:24:27,889 --> 00:24:29,560
So we're always balancing that out.

311
00:24:29,560 --> 00:24:35,584
But you know, one of the things I like that I just watched for a long time inside of us,
and it's just one simple

312
00:24:35,584 --> 00:24:43,646
Metric is that we've grown our portfolio 30, 35% in actual capacity, but only have
utilized three percent more water.

313
00:24:43,646 --> 00:24:48,128
And so that's one of the elements that we're always watching that balance and how that
comes together.

314
00:24:48,128 --> 00:24:56,702
But you know, one of the things that I'm very proud of, and I just posted about it on
LinkedIn, and you guys can read about it, is our work with Emerald AI and NVIDIA, right?

315
00:24:56,702 --> 00:24:59,754
So in Yeah, so in Massa's brickyard.

316
00:24:59,754 --> 00:25:00,124
Yeah.

317
00:25:00,124 --> 00:25:06,407
So like one of the things we're flipping it is it's not gonna be a grid burden, it's gonna
be a grid asset, right?

318
00:25:06,407 --> 00:25:07,817
Like I'll I'll be very clear.

319
00:25:07,817 --> 00:25:09,938
Like I live right here in Boulder, Colorado.

320
00:25:09,938 --> 00:25:13,610
I wish we would have a data center capability that's enabled.

321
00:25:13,610 --> 00:25:18,203
To become that asset in the grid that at peak times you can do peak shaving.

322
00:25:18,203 --> 00:25:21,715
You can really be responsible with how you're bringing that in.

323
00:25:21,715 --> 00:25:29,730
And quite frankly, we also do a tremendous amount of investment in the grid and not only
in the generation and the distribution, but finding those right elements.

324
00:25:29,730 --> 00:25:32,363
And that's where we're always watching how that comes to market.

325
00:25:32,363 --> 00:25:35,495
But again, we don't want to become energy companies.

326
00:25:35,495 --> 00:25:40,228
We just really want to become energy fluent and finding the right appropriate path to
support it.

327
00:25:40,228 --> 00:25:43,613
And my last piece is, and this always gets me in trouble, is

328
00:25:43,613 --> 00:25:48,517
I'm big on solar, I'm big on wind, I'm big on small module reactors.

329
00:25:48,517 --> 00:25:57,864
You have to use every toolkit that is afforded to you to really solve for this demand and
balance because there is a tremendous amount of d demand on power.

330
00:25:57,864 --> 00:26:09,113
Finding the right tool set, I think hopefully we can start to see in the next you know
four or five years a true small module reactor come online, which will be elegant enough

331
00:26:09,113 --> 00:26:13,336
to tuck into the grids, over or tuck into the grid, overcome some of that distribution.

332
00:26:13,336 --> 00:26:21,705
distribution challenge, but I really see the world evolving towards leveraging more types
of power just for the benefit of not only data centers, but just the community as well,

333
00:26:21,705 --> 00:26:29,173
because once those things are fully executed across all of those tool sets, the average
price for everybody goes down.

334
00:26:30,505 --> 00:26:39,507
So if megawatts are no longer a measure of success in the industry, so five years from
now, what will that measurement be?

335
00:26:40,372 --> 00:26:46,213
You know, I think I think, you know, the the theme that we've been talking about today is
like

336
00:26:46,858 --> 00:26:50,040
The industry will quote like the intelligence per watt.

337
00:26:50,040 --> 00:26:51,810
Like what is the outcome?

338
00:26:51,810 --> 00:26:58,443
And I think outcomes are the deliver like outcomes delivered are like actually the yield
out of the facility, right?

339
00:26:58,443 --> 00:27:07,167
And that's where we just generally throw these statements around that like intelligence
has become like democratized and it's free, but it's that what you do with it and how you

340
00:27:07,167 --> 00:27:09,178
how you you know execute against that.

341
00:27:09,178 --> 00:27:17,173
And that's where, you know, I've talked about it for many years and you know, we did a
special project in Copenhagen with the Novo Nordisk Foundation.

342
00:27:17,173 --> 00:27:20,836
It's a very large DGX pod that we worked with NVIDIA to stand up rather quickly.

343
00:27:20,836 --> 00:27:23,239
There's some use cases and case studies out there about it.

344
00:27:23,239 --> 00:27:31,768
But what has always excited me about that is like the cancer research and the drug, like
it's the yield or the intelligence outcome out of that infrastructure.

345
00:27:31,768 --> 00:27:35,432
And I'll also say this: it's the infrastructure is required.

346
00:27:35,432 --> 00:27:35,702
Right?

347
00:27:35,702 --> 00:27:37,533
I just for some reason.

348
00:27:37,993 --> 00:27:49,130
The masses just feel like you'll get this outweighed return of value without the
infrastructure afforded to it to actually get the component and compute capabilities that

349
00:27:49,130 --> 00:27:50,852
you're required to actually generate it.

350
00:27:50,852 --> 00:27:56,435
And so I think that's the balance that data centers are a requirement and we gotta figure
out how to do it effectively and efficiently.

351
00:27:56,435 --> 00:28:00,578
But I think that's what you'll see that'll be measured in five years is

352
00:28:00,660 --> 00:28:01,831
What did you get out of it?

353
00:28:01,831 --> 00:28:02,071
Right?

354
00:28:02,071 --> 00:28:03,822
Like what was the value delivered?

355
00:28:03,822 --> 00:28:05,163
What was the capability delivered?

356
00:28:05,163 --> 00:28:14,199
And this is where I mean you can read about like all of your passion projects and I I do
this like Alpha Go from like Goo like DeepMind, some of the work that they've been doing,

357
00:28:14,199 --> 00:28:16,771
like the true intelligence afforded across these things.

358
00:28:16,771 --> 00:28:19,283
It's just there's no vertical that's not impacted.

359
00:28:19,283 --> 00:28:24,456
And quite frankly, some of the work that's coming out is just mind-boggling to me.

360
00:28:24,745 --> 00:28:26,208
Chris, how much do you read?

361
00:28:26,208 --> 00:28:30,695
Because I I just it's I I I just think sometimes does he even sleep?

362
00:28:30,695 --> 00:28:34,484
I mean, are you always just reading and keeping up with the latest?

363
00:28:34,484 --> 00:28:34,874
Yeah.

364
00:28:34,874 --> 00:28:41,227
Well hopefully you can hear like I'm very passionate about like not only the technology
but utilizing and harnessing it in the right way.

365
00:28:41,227 --> 00:28:45,519
But to answer your question, I read more than I probably should.

366
00:28:46,340 --> 00:28:48,841
I try to balance my life out as much as possible.

367
00:28:48,841 --> 00:28:50,892
But um no, I think you absolutely have to read.

368
00:28:50,892 --> 00:28:54,113
I think what's also afforded, like I'll give you a little tin X er, right?

369
00:28:54,113 --> 00:29:03,848
Like I joke, like sometimes I get to sit with engineers and one of the best things I love
to do is in a closed door session, just watch somebody else use their laptop.

370
00:29:03,848 --> 00:29:04,408
It's the most simple.

371
00:29:04,408 --> 00:29:05,068
Simplistic thing.

372
00:29:05,068 --> 00:29:12,974
Like, if you sit next to somebody and you see their skill set or you see how they invoke
something, and you're like, whoa, whoa, whoa, what did you just do?

373
00:29:12,974 --> 00:29:14,775
And like you'll just be talking about it.

374
00:29:14,775 --> 00:29:17,737
The cross-pollinization of best practices are everything.

375
00:29:17,737 --> 00:29:22,800
So when you talk about reading, I mean short form, like even videos.

376
00:29:22,800 --> 00:29:34,058
Like if you curate your bubble correctly, and and I'm not generally talking about like the
TikTok of today, but like there's short form video formats that just give you the so what

377
00:29:34,058 --> 00:29:34,341
and

378
00:29:34,341 --> 00:29:46,105
watching what's happening also just curate your your like what you read because there is
so much misinformation it's so hard to find unbiased kind of a global clean perspective so

379
00:29:46,105 --> 00:29:54,777
the tools you use to collect what you read is another critical thing that'll 10x your life
but consolidate things down but like I also warn you on the other side because I I fell

380
00:29:54,777 --> 00:30:03,130
prey to this as well where you summarize it too much and you farm too much of that
critical thinking to your tools.

381
00:30:03,198 --> 00:30:07,725
that you'll go into kind of your own bubble where you're not metabolizing the actual
value.

382
00:30:07,725 --> 00:30:18,823
But yeah, no, I I love to read because quite frankly, it gives you a lot of introspective
time on how to digest it and how then to apply it to your life and apply it to your work.

383
00:30:18,823 --> 00:30:20,214
Hmm, so true.

384
00:30:20,214 --> 00:30:34,542
So what do you think when we look back, let's say three years from now, I know I'm pushing
you into the future a lot right now, but three years from now and we look back, what if

385
00:30:34,542 --> 00:30:35,903
how do I say this?

386
00:30:36,644 --> 00:30:41,306
And what what if we say we should have seen that coming?

387
00:30:42,487 --> 00:30:44,218
What should we have seen coming?

388
00:30:44,218 --> 00:30:45,044
Does that make sense?

389
00:30:45,044 --> 00:30:45,964
No, no, it absolutely does.

390
00:30:45,964 --> 00:30:46,374
Absolutely does.

391
00:30:46,374 --> 00:30:47,205
And and I love it, right?

392
00:30:47,205 --> 00:30:56,659
Like, I don't know, when you say three years, I'm like, gosh, three years like, even
though as we all mature in age and get older, like time has such a different perspective

393
00:30:56,659 --> 00:31:02,288
to me now with my children getting older than it would it ever had when I was single or I
was younger.

394
00:31:02,288 --> 00:31:05,674
It's just crazy how we're younger and like, my gosh, a week is so long.

395
00:31:05,674 --> 00:31:14,768
But like you say three years and with the landscape we operate in with technology, that a
a news cycle of forty-eight hours, twenty-four hours changes so

396
00:31:15,018 --> 00:31:18,050
And what is capable and what the gravity is.

397
00:31:18,050 --> 00:31:28,117
But I'll tell you, I I think what we will what we will have missed, and what we will not
truly, I think, have an appreciation for is like most people are talking about like the

398
00:31:28,117 --> 00:31:29,368
models were the bottleneck.

399
00:31:29,368 --> 00:31:38,664
Like, my gosh, like what's the next version of this model, frontier model or private
model, and just the incremental capabilities, sometimes monumental capabilities, like in

400
00:31:38,664 --> 00:31:42,677
this week with some of the stuff that came out, with some of the open weights
capabilities.

401
00:31:42,677 --> 00:31:44,961
But I would say electrons and

402
00:31:44,961 --> 00:31:50,902
elect electricians were what really created the bottleneck and

403
00:31:50,911 --> 00:31:57,544
you know, the data called it before, you know, the first gigawatt campus even broke, like
even got launched.

404
00:31:57,544 --> 00:32:02,256
Like and I and I joke a lot and I this one gets me in trouble as well, like the Bragawats.

405
00:32:02,256 --> 00:32:10,019
Like there's so many meetings I go sit in and in public forums and I always love a good
panel, especially when you can kind of like get after each other.

406
00:32:10,019 --> 00:32:18,103
But like there's so many Bragawatts where they don't have the capital, they don't have the
entitlement, they don't have the actual utility or even generation capability that the

407
00:32:18,103 --> 00:32:20,644
these these facilities will never see the light of day.

408
00:32:20,644 --> 00:32:20,824
And

409
00:32:20,824 --> 00:32:23,417
And so they're just straight kind of concepts.

410
00:32:23,417 --> 00:32:30,594
And so I think the electricians and electrons is the thing that we're gonna be looking
back and three like, uh, we should have done more.

411
00:32:30,594 --> 00:32:38,342
And that's why I talk about there's no reason you shouldn't use every tool in your tool
chest to overcome those those kind of challenges.

412
00:32:38,707 --> 00:32:40,880
Chris, it's always a pleasure talking to you.

413
00:32:40,880 --> 00:32:42,031
I always learn so much.

414
00:32:42,031 --> 00:32:48,589
And I want to encourage every single person that is listening right now to go on LinkedIn
and follow Chris Sharp.

415
00:32:48,589 --> 00:32:52,293
You will just become a smarter human.

416
00:32:52,293 --> 00:32:54,596
Chris, I really appreciate you coming on the show again.

417
00:32:54,596 --> 00:32:55,360
Thank you.

418
00:32:55,360 --> 00:32:58,176
Thank you for the opportunity, I appreciate it, and thank you for the kind words.

419
00:32:59,562 --> 00:33:00,302
Thank

420
00:33:02,024 --> 00:33:03,654
All right, that was incredible.

421
00:33:03,654 --> 00:33:03,844
Wow.

422
00:33:03,844 --> 00:33:05,532
my god.