1
00:00:00,320 --> 00:00:03,637
Let's get started with the show.
Our first guest tonight is someone we've

2
00:00:03,720 --> 00:00:06,997
been very excited we've been trying
to get them on the show for many years. He

3
00:00:07,080 --> 00:00:09,117
finally agreed and he's here tonight.

4
00:00:09,200 --> 00:00:11,357
Everyone please give a warm welcome to

5
00:00:11,440 --> 00:00:14,327
Satya Nadella, CEO of Microsoft.

6
00:00:14,410 --> 00:00:16,430
>> [cheering and applause]

7
00:00:18,400 --> 00:00:22,980
>> Hello. >> [applause]

8
00:00:24,800 --> 00:00:27,277
>> Now Satya, I have to start by making a

9
00:00:27,360 --> 00:00:29,717
confession which is that I have not been

10
00:00:29,800 --> 00:00:32,957
a regular user of a Microsoft AI product

11
00:00:33,040 --> 00:00:38,757
since 2023 when when one of them
tried to break up my marriage.

12
00:00:38,840 --> 00:00:41,717
Um Unsuccessfully I should add. My wife IS

13
00:00:41,800 --> 00:00:46,560
RIGHT THERE. >> [applause]

14
00:00:48,280 --> 00:00:50,917
>> THOSE WERE THE DAYS WHEN
WE DIDN'T [laughter] have

15
00:00:51,000 --> 00:00:54,797
Think about a world without guardrails.

16
00:00:54,880 --> 00:00:58,797
>> But so much has been happening
at Microsoft with AI since then. So just

17
00:00:58,880 --> 00:01:01,757
catch us up like what what for people
who may not have tuned in a little

18
00:01:01,840 --> 00:01:03,317
while, what have you
guys been up to in AI?

19
00:01:03,400 --> 00:01:06,397
>> Yeah, I mean, look um

20
00:01:06,480 --> 00:01:10,077
the fundamental thing that I feel we're

21
00:01:10,160 --> 00:01:15,717
about to move from um is not
talking about AI as a one

22
00:01:15,800 --> 00:01:16,957
thing

23
00:01:17,040 --> 00:01:19,877
to sort of having even a mental picture

24
00:01:19,960 --> 00:01:23,117
of what is an ecosystem

25
00:01:23,200 --> 00:01:27,277
that is sort of driven by AI, right?
So today if you think about it even since

26
00:01:27,360 --> 00:01:30,997
when you first used Sydney to now,
it has been about frontier model you sort

27
00:01:31,080 --> 00:01:33,477
of talked about Fable, what have you.

28
00:01:33,560 --> 00:01:38,717
But if we are ever going to transition
to an economy that is driven by AI, it

29
00:01:38,800 --> 00:01:42,397
can't be about one model, it can't
be about three firms, it has to be

30
00:01:42,480 --> 00:01:44,717
something that's broadly felt where the

31
00:01:44,800 --> 00:01:47,837
economy is at the frontier, not a firm

32
00:01:47,920 --> 00:01:50,437
or a model is at the frontier. So

33
00:01:50,520 --> 00:01:55,437
Microsoft is a platform company, to me
that's what we are at up to, right? So

34
00:01:55,520 --> 00:01:59,477
to me we had our developer conference
last week. It was all about, "Hey, can

35
00:01:59,560 --> 00:02:04,197
we build a platform and the tools where
every enterprise in every country can

36
00:02:04,280 --> 00:02:08,237
operate at the frontier?" To me, that's
the question. To be saying, "Hey, my

37
00:02:08,320 --> 00:02:10,437
model does this, but the economy is

38
00:02:10,520 --> 00:02:13,437
growing at 2%." means this is not going

39
00:02:13,520 --> 00:02:17,597
to end well. Unless we sort of really
get to a place where the economy is

40
00:02:17,680 --> 00:02:21,957
inflecting in terms of its economic
growth and it's broad-spread because the

41
00:02:22,040 --> 00:02:25,517
frontier benefits are. That's what
happened in electricity and every other

42
00:02:25,600 --> 00:02:27,677
technology which was a
general-purpose technology.

43
00:02:27,760 --> 00:02:29,877
>> Let me ask about one of the platform

44
00:02:29,960 --> 00:02:34,837
shifts that you signaled at Build last
week when you announced Project Solara.

45
00:02:34,920 --> 00:02:36,877
You said it would be agent-first

46
00:02:36,960 --> 00:02:41,957
hardware, kind of next-generation
set of devices that would have agents

47
00:02:42,040 --> 00:02:45,037
running them. Tell us a little bit more
about that what that looks like. Like,

48
00:02:45,120 --> 00:02:47,957
can you give us an example of maybe
what Project Solara looks like?

49
00:02:48,040 --> 00:02:51,797
>> uh, Casey, we did at Build which were
interesting, right? One was we took the

50
00:02:51,880 --> 00:02:54,197
PC itself. In fact, Jensen had done it

51
00:02:54,280 --> 00:02:59,037
the previous, uh, night at Comdex, uh,
at, uh, Computex where he talked about,

52
00:02:59,120 --> 00:03:02,237
you know, if you the the picture I one
of the my favorite pictures was Jensen

53
00:03:02,320 --> 00:03:07,077
with the desktops, the laptops, and
because the and he obviously had the RTX

54
00:03:07,160 --> 00:03:11,437
chip, which is a new SOC, uh,
with essentially a petaflop of

55
00:03:11,520 --> 00:03:13,717
compute, right? Right on your PC. So,

56
00:03:13,800 --> 00:03:17,357
that was about new functionality

57
00:03:17,440 --> 00:03:19,557
coming to the old form factor, right?

58
00:03:19,640 --> 00:03:23,237
So, think about the what I describe
as unmetered intelligence, right? So, the

59
00:03:23,320 --> 00:03:27,317
fact that you can have a Windows computer
that can run a one bill, you know,

60
00:03:27,400 --> 00:03:31,957
a trillion parameter model locally, uh,
I think it's going to be very needed,

61
00:03:32,040 --> 00:03:34,997
uh, if you're going
to ever have token, I mean,

62
00:03:35,080 --> 00:03:38,037
agents running 24/7. But then, the

63
00:03:38,120 --> 00:03:40,997
question it sets up is in a world where

64
00:03:41,080 --> 00:03:43,477
we have these models

65
00:03:43,560 --> 00:03:45,477
and these agents that are long-running,

66
00:03:45,560 --> 00:03:48,637
can I have like a badge?

67
00:03:48,720 --> 00:03:51,437
Uh, right? If I'm I'm in a hospital and

68
00:03:51,520 --> 00:03:54,557
I'm walking to from station to station.

69
00:03:54,640 --> 00:03:58,037
Can I just take out this device, which

70
00:03:58,120 --> 00:04:00,877
can scan, which can input, which can

71
00:04:00,960 --> 00:04:03,717
take my speech output, and turn it into

72
00:04:03,800 --> 00:04:07,277
a prompt? Um that is what I think an

73
00:04:07,360 --> 00:04:09,157
agent-first device looks like, right?

74
00:04:09,240 --> 00:04:13,437
So, it's sort of really a new
The centrality of the phone, I think, is

75
00:04:13,520 --> 00:04:17,597
still going to be there for a lot
of apps that we use. But in agent world,

76
00:04:17,680 --> 00:04:19,877
you kind of have that ambient

77
00:04:19,960 --> 00:04:22,797
intelligence that is like a sense of

78
00:04:22,880 --> 00:04:26,037
feel that then works with your models.

79
00:04:26,120 --> 00:04:28,717
And so, that's what our goal is, to

80
00:04:28,800 --> 00:04:31,677
invent the new form factors that are not

81
00:04:31,760 --> 00:04:36,077
beholden to the old form
factors for this functionality.

82
00:04:36,160 --> 00:04:39,317
>> It's super interesting, and I want
to hear a lot more about it. We have many

83
00:04:39,400 --> 00:04:43,397
more questions about AI, but I wanted
to ask about one more device that has been

84
00:04:43,480 --> 00:04:45,917
in the news today, which is the Xbox.

85
00:04:46,000 --> 00:04:49,317
Um the leaders of the Xbox division put
out a memo today saying that we should

86
00:04:49,400 --> 00:04:52,357
expect a hard reset of Xbox coming soon.

87
00:04:52,440 --> 00:04:56,077
They said that there are massive
increases in the prices of components,

88
00:04:56,160 --> 00:05:00,597
and that Xbox might need a new business
model. So, as a big gamer who's enjoyed

89
00:05:00,680 --> 00:05:04,717
many happy hours on Xbox, Satya, I have
to ask, what is your strategy for Xbox?

90
00:05:04,800 --> 00:05:07,077
>> Yeah, so look, in fact, we're you know,

91
00:05:07,160 --> 00:05:10,277
this is the 25th year of Xbox.

92
00:05:10,360 --> 00:05:14,597
Um and we're very
thrilled about the progress

93
00:05:14,680 --> 00:05:17,917
we have made. I mean, gaming, in an
interesting way, at Microsoft is older

94
00:05:18,000 --> 00:05:21,037
than even Windows and Office,
right? The first

95
00:05:21,120 --> 00:05:23,957
app we built was the Flight Simulator.

96
00:05:24,040 --> 00:05:26,957
And so, it's got a long heritage. Xbox

97
00:05:27,040 --> 00:05:30,037
itself has been there for 24 5 years.

98
00:05:30,120 --> 00:05:33,237
The challenge now for us is to think

99
00:05:33,320 --> 00:05:36,837
about how do you innovate both in

100
00:05:36,920 --> 00:05:40,077
hardware as well as in the games going

101
00:05:40,160 --> 00:05:44,717
forward in a world in an economically
viable way, right? I think one of the

102
00:05:44,800 --> 00:05:49,157
things that Asha, who's just taken over
Xbox, put out is that we've invested a

103
00:05:49,240 --> 00:05:51,637
lot. No one can accuse Microsoft of not

104
00:05:51,720 --> 00:05:53,397
having invested

105
00:05:53,480 --> 00:05:56,517
for the last 25 years. And now we have

106
00:05:56,600 --> 00:05:59,877
to turn this into a sustainable business

107
00:05:59,960 --> 00:06:02,477
that delivers what is fundamentally the

108
00:06:02,560 --> 00:06:05,037
one of the best sources of entertainment

109
00:06:05,120 --> 00:06:09,797
still. The challenge we have
is we're not the been monetizing that

110
00:06:09,880 --> 00:06:13,037
entertainment. In fact,
if anything we've been subsidizing that

111
00:06:13,120 --> 00:06:17,357
entertainment. In fact, like there's
more monetization of Xbox games

112
00:06:17,440 --> 00:06:21,517
happening on YouTube than at Microsoft.

113
00:06:21,600 --> 00:06:25,037
And so that doesn't mean we go do things
that are unnatural. We want us to do

114
00:06:25,120 --> 00:06:29,437
what is really our job, which is to build
great games, build great hardware,

115
00:06:29,520 --> 00:06:33,117
but we've got to do it in an economically
sustainable way. So I think

116
00:06:33,200 --> 00:06:37,957
Asha is really 100 days in and she put
out a post saying in the next 100 days

117
00:06:38,040 --> 00:06:41,917
she's going to take a fresh look
and make sure we deliver on what our fans

118
00:06:42,000 --> 00:06:45,117
expect of us both on the hardware
side or on the publishing side.

119
00:06:45,200 --> 00:06:47,717
>> I mean, can you give us just any
more detail? Like like when I hear that I

120
00:06:47,800 --> 00:06:51,077
think, okay, so maybe like the Xbox gets
way more expensive, the games get way

121
00:06:51,160 --> 00:06:54,757
more expensive. Like is there any sort
of like carrot you can offer the gamers?

122
00:06:54,840 --> 00:06:57,597
>> Like I I think we have to find ways to

123
00:06:57,680 --> 00:07:00,797
deliver the games at which

124
00:07:00,880 --> 00:07:02,677
it's economically relevant for the

125
00:07:02,760 --> 00:07:06,157
customer and for us. So today there's a

126
00:07:06,240 --> 00:07:08,957
issue, in fact, unfortunately because of

127
00:07:09,040 --> 00:07:13,517
what's happening with the cloud and AI,
the prices have gone up, right? It's

128
00:07:13,600 --> 00:07:17,717
happening with PCs, it's happening
with phones, Xbox is impacted

129
00:07:17,800 --> 00:07:20,357
as well. So the scarcity of the

130
00:07:20,440 --> 00:07:23,317
semiconductor supply and memory in

131
00:07:23,400 --> 00:07:25,117
particular

132
00:07:25,200 --> 00:07:30,997
having a massive impact on consumer
electronics all up. That's a temporal

133
00:07:31,080 --> 00:07:34,357
thing. That I think
we'll get through. It is

134
00:07:34,440 --> 00:07:38,397
not going to be a permanent. But there
is a permanent thing, which is what's

135
00:07:38,480 --> 00:07:43,677
the Xbox model going forward. And that's
where if you think about it, like PCs

136
00:07:43,760 --> 00:07:45,837
and consoles both have their place.

137
00:07:45,920 --> 00:07:49,757
Obviously, mobile has people play
and elsewhere. And so we have to now bring

138
00:07:49,840 --> 00:07:53,277
it all together while staying
true to what we've always done.

139
00:07:53,360 --> 00:07:57,437
>> Satya, I want to take you back
in a time machine. The year is 2023.

140
00:07:57,520 --> 00:07:59,877
The board of OpenAI has just fired Sam

141
00:07:59,960 --> 00:08:01,917
Altman, one of Microsoft's biggest

142
00:08:02,000 --> 00:08:05,077
partners. You and your team spent a

143
00:08:05,160 --> 00:08:07,637
harried weekend trying to pull together

144
00:08:07,720 --> 00:08:10,037
an entirely new division of Microsoft

145
00:08:10,120 --> 00:08:13,117
Microsoft Advanced AI Research to sort

146
00:08:13,200 --> 00:08:15,637
of catch the employees that are making a

147
00:08:15,720 --> 00:08:20,237
mass exodus from OpenAI. You're ordering
laptops. You're opening up an office in

148
00:08:20,320 --> 00:08:23,437
San Francisco so that all these people
have a place to go work. Uh the company

149
00:08:23,520 --> 00:08:26,517
looks like it's on the verge of collapse.
Uh and then nothing happens.

150
00:08:26,600 --> 00:08:30,757
Then then Sam gets rehired and OpenAI
stands back up on its feet. Um and I

151
00:08:30,840 --> 00:08:34,357
want to present you as part of this time
machine experiment with a piece of rare

152
00:08:34,440 --> 00:08:36,517
merchandise, which I recently acquired

153
00:08:36,600 --> 00:08:38,957
from an OpenAI employee, which is a

154
00:08:39,040 --> 00:08:42,157
Microsoft Advanced AI Research

155
00:08:42,240 --> 00:08:45,997
sweatshirt. >> That is awesome.

156
00:08:46,080 --> 00:08:48,837
>> To be clear,
this division never existed

157
00:08:48,920 --> 00:08:52,597
and I was told by the person who made
this that they had to sort of fudge a

158
00:08:52,680 --> 00:08:56,757
little bit on the the sort of copying
and printing shop application, which

159
00:08:56,840 --> 00:08:59,277
made them prove that they were
a Microsoft employee >> [laughter]

160
00:08:59,360 --> 00:09:01,557
>> to to get this.
But this is for you if you ever want to

161
00:09:01,640 --> 00:09:03,197
>> Thank you. Thank you so much.

162
00:09:03,280 --> 00:09:06,917
>> down memory lane. >> Oh, man.
That weekend, I remember that

163
00:09:07,000 --> 00:09:10,317
forever. And thank you for this. But all
I remember, quite frankly, the weekend

164
00:09:10,400 --> 00:09:13,877
is India getting thrashed by Australia
in cricket. That was the >> [laughter]

165
00:09:13,960 --> 00:09:18,197
>> the more tragic thing. >> So,

166
00:09:18,280 --> 00:09:21,997
if that had happened, in the world where
all of these OpenAI employees end up

167
00:09:22,080 --> 00:09:26,797
working at Microsoft in a new Advanced
AI Research division run by Sam Altman

168
00:09:26,880 --> 00:09:30,197
and Greg Brockman,
in that world is Microsoft better or

169
00:09:30,280 --> 00:09:32,877
worse off with AI than it is today in

170
00:09:32,960 --> 00:09:34,077
this world?

171
00:09:34,160 --> 00:09:37,957
>> Look, we're thrilled that Greg and Sam

172
00:09:38,040 --> 00:09:40,197
made it back to OpenAI and they are

173
00:09:40,280 --> 00:09:42,277
where they are and they're now, as you

174
00:09:42,360 --> 00:09:45,597
said, our father S1 or what have you and

175
00:09:45,680 --> 00:09:49,797
look, it's fascinating,
right? When we initially

176
00:09:49,880 --> 00:09:53,757
um took the bet on OpenAI, it was a not,

177
00:09:53,840 --> 00:09:55,077
you know, it was a

178
00:09:55,160 --> 00:09:58,797
a research lab, a nonprofit entity that

179
00:09:58,880 --> 00:10:01,317
had created a for-profit

180
00:10:01,400 --> 00:10:03,957
uh unit and said, "Hey, they went and

181
00:10:04,040 --> 00:10:08,957
shopped around and said, who can back
our crazy idea that intelligence is log

182
00:10:09,040 --> 00:10:11,997
of compute?" Uh and quite
frankly, there were lots of

183
00:10:12,080 --> 00:10:14,437
people that uh or at Microsoft at that

184
00:10:14,520 --> 00:10:16,997
time who thought this is nuts.

185
00:10:17,080 --> 00:10:21,197
Um but, you know, we said, "I think
this is a worthwhile thing to back." And

186
00:10:21,280 --> 00:10:24,837
quite frankly, we changed uh I think and
they changed through their work and the

187
00:10:24,920 --> 00:10:29,277
OpenAI through their work, the world
and here we are in 2026 and we're thrilled

188
00:10:29,360 --> 00:10:32,357
about it. >> You guys
recently renegotiated your deal

189
00:10:32,440 --> 00:10:37,517
um between OpenAI and Microsoft. And uh
I understand what OpenAI got out of that

190
00:10:37,600 --> 00:10:41,397
deal, they got the ability to work with
multiple cloud providers, to be a little

191
00:10:41,480 --> 00:10:44,677
bit more open about how they
commercialize their technology. What did

192
00:10:44,760 --> 00:10:47,197
Microsoft get out
of the revision of that deal?

193
00:10:47,280 --> 00:10:49,277
>> I mean, we have a lot of, you know,

194
00:10:49,360 --> 00:10:51,077
interests in OpenAI. We're obviously on

195
00:10:51,160 --> 00:10:54,957
their cap table and uh we're a They're a

196
00:10:55,040 --> 00:10:58,117
customer of ours, a large one. We are uh

197
00:10:58,200 --> 00:11:00,637
They're a source of IP for us uh all the

198
00:11:00,720 --> 00:11:05,437
way to 32 and uh and so all
of the things and and and at

199
00:11:05,520 --> 00:11:09,557
the same time we have the ability
and the flexibility to reuse the IP, build

200
00:11:09,640 --> 00:11:14,077
our own IP. We just last week launched
MAI models, uh which have hill-climbed

201
00:11:14,160 --> 00:11:17,877
from the ground up. We published,
in fact, the paper, which I think should

202
00:11:17,960 --> 00:11:22,157
help people even get the capability we
have, especially if you take those two

203
00:11:22,240 --> 00:11:24,637
thoughts, right, which is uh here is,

204
00:11:24,720 --> 00:11:26,677
you know, intelligence is log of compute

205
00:11:26,760 --> 00:11:29,517
and here is a pure lineage model from

206
00:11:29,600 --> 00:11:32,037
Microsoft that climbed all the way,

207
00:11:32,120 --> 00:11:34,437
means that we now have the ability to

208
00:11:34,520 --> 00:11:39,157
keep going and to us and we would
and and our infrastructure. I must mention

209
00:11:39,240 --> 00:11:43,037
that that we wouldn't have been
where we are with even Azure

210
00:11:43,120 --> 00:11:46,757
but for that close partnership with
OpenAI. So, we have the compute, we have

211
00:11:46,840 --> 00:11:49,317
now the model and we have
still the partnership.

212
00:11:49,400 --> 00:11:52,037
>> Let's talk about those models. Is your

213
00:11:52,120 --> 00:11:56,677
goal to make a frontier best model
in the world? And if so, what's the

214
00:11:56,760 --> 00:11:59,837
strategy for overtaking
a ChatGPT, a Gemini, a Claude?

215
00:11:59,920 --> 00:12:02,517
>> Yeah, so I think the way I would say is

216
00:12:02,600 --> 00:12:07,597
our real goal is to get
everyone across the

217
00:12:07,680 --> 00:12:11,477
ecosystem to the frontier. So, we are
going to take a slightly different take

218
00:12:11,560 --> 00:12:14,397
in saying for example,
if you think about what's

219
00:12:14,480 --> 00:12:17,597
how does one build a frontier model? Um

220
00:12:17,680 --> 00:12:20,797
you hill climb, URL

221
00:12:20,880 --> 00:12:25,557
um and then you need data, right? So,
at this point we have saturated the data

222
00:12:25,640 --> 00:12:29,877
and so that means you're basically
hovering the data from every place,

223
00:12:29,960 --> 00:12:33,357
right? So, the question is what if you
turn that around and said, "No, there's

224
00:12:33,440 --> 00:12:36,077
a base model that has reasoning, that

225
00:12:36,160 --> 00:12:38,997
has the agent loop, but you can bring it

226
00:12:39,080 --> 00:12:41,677
into your R&D." Every company, right? If

227
00:12:41,760 --> 00:12:43,597
you if the if the comp- the future of

228
00:12:43,680 --> 00:12:46,397
the firm is human capital and token

229
00:12:46,480 --> 00:12:49,637
capital. I want every balance sheet,

230
00:12:49,720 --> 00:12:53,077
every income statement in every company

231
00:12:53,160 --> 00:12:55,837
to have both. And that's our goal with

232
00:12:55,920 --> 00:12:57,837
our frontier model. Our model should be

233
00:12:57,920 --> 00:13:00,437
the best model that they can use as a

234
00:13:00,520 --> 00:13:05,357
base and keep even the weights,
uh definitely the harness and the

235
00:13:05,440 --> 00:13:08,797
context which is theirs and they can
replace our model with anything else.

236
00:13:08,880 --> 00:13:13,077
So, that to me is more of a vision that I
think is what Micro- I always ask the

237
00:13:13,160 --> 00:13:15,797
question, why does Microsoft

238
00:13:15,880 --> 00:13:18,157
uh or why does the world need Microsoft?

239
00:13:18,240 --> 00:13:22,197
And if we are successful, can the world
around us be successful? This I believe

240
00:13:22,280 --> 00:13:24,597
is a more sustainable way to go at it.

241
00:13:24,680 --> 00:13:27,957
>> We have to ask a question about the AI
backlash that we're seeing around the

242
00:13:28,040 --> 00:13:30,117
country. Graduation speakers are getting

243
00:13:30,200 --> 00:13:33,237
booted. AI is polling terribly. Uh lots

244
00:13:33,320 --> 00:13:38,157
of people upset about data centers. What
role does Microsoft have in that image

245
00:13:38,240 --> 00:13:41,917
or helping to solve that?
And how do you think the industry can

246
00:13:42,000 --> 00:13:46,537
find a path forward that involves,
I don't know, being a little more popular?

247
00:13:46,620 --> 00:13:50,877
>> [laughter] >> I think we
can start, but frankly, by

248
00:13:50,960 --> 00:13:53,237
painting a picture

249
00:13:53,320 --> 00:13:57,277
and delivering the results on why

250
00:13:57,360 --> 00:14:01,917
there are more than, you know, everyone
is a stakeholder, right? You can't go

251
00:14:02,000 --> 00:14:03,477
out there and say, "I have this

252
00:14:03,560 --> 00:14:06,357
unbelievable technology. Uh except

253
00:14:06,440 --> 00:14:10,597
you're not going to have a job and um in
fact, we're going to take all your water

254
00:14:10,680 --> 00:14:12,917
and all your energy and, you know, uh

255
00:14:13,000 --> 00:14:16,437
good luck." I mean, that cannot be uh an

256
00:14:16,520 --> 00:14:20,757
an And then the no wonder there's so
much anxiety, right? You talked about

257
00:14:20,840 --> 00:14:24,317
the students or you can talk about
a community. And so therefore, you got to

258
00:14:24,400 --> 00:14:26,037
do the hard work at this point. It is

259
00:14:26,120 --> 00:14:28,117
what it is, right? So, you can't deny

260
00:14:28,200 --> 00:14:31,397
that the perception is terrible. And so,

261
00:14:31,480 --> 00:14:36,117
I feel at Microsoft, we want to whether
it's like take the the data centers. Uh

262
00:14:36,200 --> 00:14:37,957
it's fascinating. We've been operating

263
00:14:38,040 --> 00:14:41,397
in Quincy, Washington for 20 plus years.

264
00:14:41,480 --> 00:14:45,517
And I just, you know,
uh we celebrated our 20th year.

265
00:14:45,600 --> 00:14:49,917
You know, if I look at what all has
happened in Quincy in the 20 years,

266
00:14:50,000 --> 00:14:52,477
their tax base has gone up.

267
00:14:52,560 --> 00:14:55,557
Their taxes locally have gone down. Uh

268
00:14:55,640 --> 00:14:58,397
they have more employment locally.

269
00:14:58,480 --> 00:14:59,517
>> Because of the data center.

270
00:14:59,600 --> 00:15:02,877
>> Because of the data center. In fact,
the data center is it's kind of like

271
00:15:02,960 --> 00:15:07,717
basically it's become a data center town.
Uh and we when we we we did this,

272
00:15:07,800 --> 00:15:11,917
you know, cookout and people came
and they said they celebrate the

273
00:15:12,000 --> 00:15:14,717
rejuvenation of Quincy, Washington

274
00:15:14,800 --> 00:15:19,117
because of our presence for the last
20 years. That long That's the first

275
00:15:19,200 --> 00:15:23,397
longitudinal thing of 20 years that I've
seen. And that's what the communities

276
00:15:23,480 --> 00:15:26,597
where data centers are. They They
definitely can't increase price on

277
00:15:26,680 --> 00:15:30,677
energy. They have to be, you know,
very, you know, in fact, they they should

278
00:15:30,760 --> 00:15:35,597
replenish all the water they use uh
and create economic opportunity. So,

279
00:15:35,680 --> 00:15:40,557
that's on the data center side. But,
across the economy, if there are small

280
00:15:40,640 --> 00:15:44,357
businesses feeling like, "Wow,
AI is making me more productive." If every

281
00:15:44,440 --> 00:15:49,197
large multinational is able to say, "Oh,
I'm building that token capital and the

282
00:15:49,280 --> 00:15:53,877
human capital." Right? Because
the the the big question is employment.

283
00:15:53,960 --> 00:15:58,437
Everybody thinks that all jobs are going
away. But, if you sort of And I'm not

284
00:15:58,520 --> 00:16:01,837
saying there won't be real displacement,
the workflow doesn't change. But, take

285
00:16:01,920 --> 00:16:07,117
software development. Of course,
you know, software development is now

286
00:16:07,200 --> 00:16:10,277
all agentic. Except,
if you think about even the

287
00:16:10,360 --> 00:16:14,077
evolution of the GitHub app, right?
You know, when I had a hundred CLIs, what

288
00:16:14,160 --> 00:16:18,757
did I need? I need a new IDE. It's
called an ADE, right? I'm back at again,

289
00:16:18,840 --> 00:16:20,437
some piece of software that helps me

290
00:16:20,520 --> 00:16:23,197
manage all of this complexity. So, I

291
00:16:23,280 --> 00:16:25,837
think we have to think about new work

292
00:16:25,920 --> 00:16:30,837
that gets done, which will
be meta-cognition, meta-work,

293
00:16:30,920 --> 00:16:33,997
that is going to have wages.
And we have to be concrete about that.

294
00:16:34,080 --> 00:16:38,397
>> But, but help us understand like
what your own view is of the potential

295
00:16:38,480 --> 00:16:41,477
disruption, right? Like there because
I've I've been talking to so many

296
00:16:41,560 --> 00:16:44,157
economists, like tech leaders like
yourself about this over the past couple

297
00:16:44,240 --> 00:16:47,357
of months, and truly opinion is all over
the place. And I talked to some folks

298
00:16:47,440 --> 00:16:50,717
who say, "Yes, you better believe I'm
hiring fewer people next year because of

299
00:16:50,800 --> 00:16:54,317
AI." And then I talked to other
people that say, "I I can't get enough

300
00:16:54,400 --> 00:16:57,917
engineers. I want more than I have."
So, where are you on that? In two years

301
00:16:58,000 --> 00:16:59,437
you're going to have
more engineers or fewer?

302
00:16:59,520 --> 00:17:01,917
>> Yeah, so I think the

303
00:17:02,000 --> 00:17:06,837
if in the early '80s, if someone had
come to us and said, "Hey, we're going

304
00:17:06,920 --> 00:17:09,637
to have three and a half billion people
in the world who are all going to be

305
00:17:09,720 --> 00:17:12,557
typists." We would have
said, "Why does the world

306
00:17:12,640 --> 00:17:14,957
need three and a half billion typists?"

307
00:17:15,040 --> 00:17:18,597
Except, we do. We all get up in the
morning and type, but we're doing {quote

308
00:17:18,680 --> 00:17:20,877
unquote} information work, knowledge

309
00:17:20,960 --> 00:17:25,237
work, and so on. So, that type of change

310
00:17:25,320 --> 00:17:28,877
is going to have to happen, and that and

311
00:17:28,960 --> 00:17:33,397
each of them will have a name that
and it'll have a wage support. That

312
00:17:33,480 --> 00:17:35,877
reinvention, right? So, the software

313
00:17:35,960 --> 00:17:38,437
developer of the past to the software

314
00:17:38,520 --> 00:17:40,717
developer of the future may have the

315
00:17:40,800 --> 00:17:44,157
similar sort of skills, but the work

316
00:17:44,240 --> 00:17:48,797
they do is different. Because they're
managing a group of 100 agents, 1,000

317
00:17:48,880 --> 00:17:52,477
agents. In fact, there's a beautiful
term one of my colleagues has, which is

318
00:17:52,560 --> 00:17:55,997
just like in software development
we already had this concept of test

319
00:17:56,080 --> 00:18:00,797
coverage. One of the new things that we
are learning is what I'll call cognitive

320
00:18:00,880 --> 00:18:03,957
coverage, right? So, what does
a software developer do? I might have I

321
00:18:04,040 --> 00:18:05,877
have a repo full of code that was

322
00:18:05,960 --> 00:18:08,917
written by agents. I am

323
00:18:09,000 --> 00:18:13,517
cognitively understanding what happened,
and I now need tools for cognitive

324
00:18:13,600 --> 00:18:17,077
coverage on what there has been. That's
a job, I think, of a software developer.

325
00:18:17,160 --> 00:18:19,277
And in order to do that, you got to go
to school, you got to learn computer

326
00:18:19,360 --> 00:18:20,997
science, and have cognitive coverage.

327
00:18:21,080 --> 00:18:24,157
And so, this reinvention of work, the

328
00:18:24,240 --> 00:18:27,157
work artifact, the workflow, right?

329
00:18:27,240 --> 00:18:31,237
Because software went from input
to output. That's a format change, an

330
00:18:31,320 --> 00:18:35,557
artifact change, the workflow has
changed, and the work changes with it.

331
00:18:35,640 --> 00:18:38,797
>> I feel like what people really want
to hear is some combination of like, your

332
00:18:38,880 --> 00:18:42,077
job isn't going to change that much,
or if it does change, you're going to get

333
00:18:42,160 --> 00:18:44,917
paid more. Do you think that either of
those things will be true for most people?

334
00:18:45,000 --> 00:18:47,037
>> That's the thing. The wages, you know,

335
00:18:47,120 --> 00:18:49,877
have always been about what is it that

336
00:18:49,960 --> 00:18:53,237
we as a society value, right? Um I mean,

337
00:18:53,320 --> 00:18:56,317
we have grown in the last 200 years, 250

338
00:18:56,400 --> 00:18:58,677
years, was about a particular form of

339
00:18:58,760 --> 00:19:01,797
expertise and accrual of knowledge. So,

340
00:19:01,880 --> 00:19:03,957
when you have abundance of some form of

341
00:19:04,040 --> 00:19:07,397
expertise, what is that human ability to

342
00:19:07,480 --> 00:19:10,877
now build a new expertise that is not

343
00:19:10,960 --> 00:19:15,837
trainable? Right? In fact, there was a
nice blog I read this morning from Sarah

344
00:19:15,920 --> 00:19:18,077
Go, which was I thought was an
interesting one where she sort of

345
00:19:18,160 --> 00:19:22,437
mentions, "Hey, what is the untrainable
part?" And that applies to organizations

346
00:19:22,520 --> 00:19:24,637
and I think us as well. And it's

347
00:19:24,720 --> 00:19:27,957
We as humans have agency, ambition,

348
00:19:28,040 --> 00:19:30,837
that should not be counted out. And if

349
00:19:30,920 --> 00:19:35,157
you look at even what is human capital
today, the human cap in spite of all the

350
00:19:35,240 --> 00:19:37,557
digital systems we have in in at our

351
00:19:37,640 --> 00:19:41,157
disposal, we do the glue work. We will

352
00:19:41,240 --> 00:19:44,237
discover the new glue work that happens

353
00:19:44,320 --> 00:19:48,397
in with all this automation. And that I
think is the process of change. One idea

354
00:19:48,480 --> 00:19:52,917
that's been floated recently,
including by reportedly Sam Altman and and

355
00:19:53,000 --> 00:19:56,517
President Trump has also weighed
in on it, is the idea of having the US

356
00:19:56,600 --> 00:20:02,117
government take direct investment
stakes in frontier AI companies.

357
00:20:02,200 --> 00:20:05,517
Do you think that's a good idea? What
percent of Microsoft would you like the

358
00:20:05,600 --> 00:20:11,027
US government to own? MSFT, you can trade.

359
00:20:11,110 --> 00:20:13,757
>> [laughter] >> Do you think
that is a way for the the

360
00:20:13,840 --> 00:20:16,317
gains of AI to be more broadly felt? You

361
00:20:16,400 --> 00:20:18,517
know, look, this is all very new, right?

362
00:20:18,600 --> 00:20:21,757
I mean, the idea that there may be

363
00:20:21,840 --> 00:20:23,637
the United States, whether it has a

364
00:20:23,720 --> 00:20:25,877
sovereign fund and the sovereign fund

365
00:20:25,960 --> 00:20:29,357
has equity stakes and that somehow

366
00:20:29,440 --> 00:20:33,397
is part of what is considered
this the wealth of

367
00:20:33,480 --> 00:20:37,237
the citizens of this country, I think
it's a novel idea. Other countries have

368
00:20:37,320 --> 00:20:40,637
done it. I think Alaska
has some form of it

369
00:20:40,720 --> 00:20:43,437
in the state because of the oil wealth.

370
00:20:43,520 --> 00:20:47,717
So, I'm not sort of opposed
to innovative ideas like this, but at the

371
00:20:47,800 --> 00:20:51,197
end of the day, I mean, there's
this entire movement of if only we had

372
00:20:51,280 --> 00:20:56,557
invested some portion of our social
security in S&P 500, we would have a

373
00:20:56,640 --> 00:20:59,837
surplus or what have you. And so, to the
degree to which some of these ideas can

374
00:20:59,920 --> 00:21:02,597
be played out and and they succeed, I

375
00:21:02,680 --> 00:21:04,797
think we'll all benefit from it.

376
00:21:04,880 --> 00:21:07,957
>> You um Dwarkesh Patel in February 2025

377
00:21:08,040 --> 00:21:10,277
that your benchmark for achieving AGI

378
00:21:10,360 --> 00:21:14,037
was 10% GDP growth. Seems like we're not

379
00:21:14,120 --> 00:21:17,637
super close to hitting 10% yet. Um I'm

380
00:21:17,720 --> 00:21:22,997
curious like how how you view that kind
of statement you made a year ago now and

381
00:21:23,080 --> 00:21:26,957
do you see any sort of recent acceleration
that makes you think it's more possible?

382
00:21:27,040 --> 00:21:29,597
>> In fact, it's one of the

383
00:21:29,680 --> 00:21:32,477
things that I think a lot about is the

384
00:21:32,560 --> 00:21:35,317
difficulty of even a very powerful

385
00:21:35,400 --> 00:21:40,077
general purpose technology and its
diffusion and the amount of change

386
00:21:40,160 --> 00:21:43,877
management that is required, right?
I mean, the systems, right? For example,

387
00:21:43,960 --> 00:21:47,477
one of the challenges right now that
we're going to face in the next year, 2

388
00:21:47,560 --> 00:21:50,477
years is this, you know, the economics

389
00:21:50,560 --> 00:21:55,037
of tokens. Uh for example,
the hard truth is that the

390
00:21:55,120 --> 00:21:57,277
marginal cost of productivity

391
00:21:57,360 --> 00:22:00,117
improvement has to match the marginal

392
00:22:00,200 --> 00:22:02,957
cost of the token. That's a management

393
00:22:03,040 --> 00:22:07,157
discipline, right? So, you can't just
say, "Hey, I love token maxing because

394
00:22:07,240 --> 00:22:11,357
it's sort of money in my bank."
That business has to

395
00:22:11,440 --> 00:22:15,397
benefit from it. And that is what
is going to really drive it. So, in fact,

396
00:22:15,480 --> 00:22:17,317
it's fascinating. The equation for the

397
00:22:17,400 --> 00:22:19,397
10% growth would be when you have a

398
00:22:19,480 --> 00:22:23,397
perfect match between the marginal cost

399
00:22:23,480 --> 00:22:27,197
of the token to the marginal
value and it's priced,

400
00:22:27,280 --> 00:22:32,157
right? So, that means it's the best way
to get at it. If that happens, 10% is

401
00:22:32,240 --> 00:22:35,757
definitely going to happen. But not
definitely what's happening right now

402
00:22:35,840 --> 00:22:39,117
where everybody goes and white codes and
token maxes, that's not a way to achieve

403
00:22:39,200 --> 00:22:42,437
10% growth. >> How much
sort of token maxing has been

404
00:22:42,520 --> 00:22:44,957
going on at Microsoft? And then you Yeah.

405
00:22:45,040 --> 00:22:48,637
>> [laughter] >> And I'm
out of it. Like get me

406
00:22:48,720 --> 00:22:52,717
And what I mean by that is um
you know, look, I want people to

407
00:22:52,800 --> 00:22:56,397
obviously and myself,
I'm like a token maxer, too. So,

408
00:22:56,480 --> 00:23:00,397
it's addictive. It's kind of like, "Hey,
I love this thing." So, So you have to

409
00:23:00,480 --> 00:23:03,317
step back when the novelty wears off to

410
00:23:03,400 --> 00:23:05,237
say what is it that I'm trying to

411
00:23:05,320 --> 00:23:07,677
create? Uh in fact the thing that I love

412
00:23:07,760 --> 00:23:11,157
now in in Copilot now is our auto mode.

413
00:23:11,240 --> 00:23:15,157
And so we now have a very good we have
an economic model that's feeding it as

414
00:23:15,240 --> 00:23:18,117
well. Basically I say don't use frontier

415
00:23:18,200 --> 00:23:20,557
models for non-frontier problems, right?

416
00:23:20,640 --> 00:23:25,197
Please let's kind of match these things
such that you get the outputs, you get

417
00:23:25,280 --> 00:23:29,317
the economics, uh and it's
not it can't be a race to

418
00:23:29,400 --> 00:23:32,317
just doing things
that just don't add value.

419
00:23:32,400 --> 00:23:35,797
>> Give us a flavor of like Satya's token
maxing. Like what are some of your your

420
00:23:35,880 --> 00:23:39,117
big token projects lately? >> Yeah,
I mean like the one thing that I

421
00:23:39,200 --> 00:23:44,517
recently um built was you know

422
00:23:44,600 --> 00:23:49,197
I always sort of felt that

423
00:23:49,280 --> 00:23:51,357
I want a repo

424
00:23:51,440 --> 00:23:54,877
that is in sync all the time

425
00:23:54,960 --> 00:23:58,557
with discussions that are
happening out of band

426
00:23:58,640 --> 00:24:02,517
uh that I'm not in. Right?
Think about even that concept.

427
00:24:02,600 --> 00:24:06,277
See I like that thing because it sort of
is sort of not possible today, right? So

428
00:24:06,360 --> 00:24:09,397
you'll have essentially the ability now

429
00:24:09,480 --> 00:24:11,477
to have an agent that literally is

430
00:24:11,560 --> 00:24:13,997
looking at all the work discussions that

431
00:24:14,080 --> 00:24:17,397
may be related to your repo and creating

432
00:24:17,480 --> 00:24:21,437
the plan and executing the plan.
And it's just in fact all I did was I put

433
00:24:21,520 --> 00:24:24,077
work IQ which is the database underneath

434
00:24:24,160 --> 00:24:28,837
uh all of Microsoft 365
and as an MCP server and to my

435
00:24:28,920 --> 00:24:32,837
coding agent and I said keep watching
that and every time people discuss this

436
00:24:32,920 --> 00:24:36,677
repo, please change my repo. And it
keeps working. And so this is like the

437
00:24:36,760 --> 00:24:39,437
best way to keep your come basically a

438
00:24:39,520 --> 00:24:44,037
model you know in sync with all
requirements that ever come up.

439
00:24:44,120 --> 00:24:47,877
>> You've been thinking a lot about the
political economy of AI. Obviously, you

440
00:24:47,960 --> 00:24:50,957
know, the things you're talking about
about diffusion and adoption and GDP

441
00:24:51,040 --> 00:24:54,237
growth are all part of that. What do
you think the people in San Francisco

442
00:24:54,320 --> 00:24:58,637
leading the AI companies here get
wrong about the political economy of AI?

443
00:24:58,720 --> 00:25:01,837
>> I I mean I wouldn't say they're wrong
about the political economy of AI, but I

444
00:25:01,920 --> 00:25:05,717
do think when I look back
at the you know there's

445
00:25:05,800 --> 00:25:09,677
a very cool book I read I
think in December. It's I

446
00:25:09,760 --> 00:25:13,077
I may have the name wrong, but it's
written by Joel Mokyr and a couple of a

447
00:25:13,160 --> 00:25:16,797
co-authors on it's I think it's called
Parallel Paths to Prosperity. They

448
00:25:16,880 --> 00:25:21,597
describe a little bit of how the last
thousand years in the you know the West

449
00:25:21,680 --> 00:25:23,397
grew and what was what was happening in

450
00:25:23,480 --> 00:25:25,957
China. So it is a thousand year history.

451
00:25:26,040 --> 00:25:29,117
Um but the fundamental thing when I take

452
00:25:29,200 --> 00:25:31,677
from that book and and in general when I

453
00:25:31,760 --> 00:25:33,317
read history is

454
00:25:33,400 --> 00:25:36,757
the West in particular got three things

455
00:25:36,840 --> 00:25:40,157
into a virtual cycle, right? They got uh

456
00:25:40,240 --> 00:25:42,637
technological revolutions

457
00:25:42,720 --> 00:25:45,077
mark and markets

458
00:25:45,160 --> 00:25:48,677
and democracy. All both acting as a

459
00:25:48,760 --> 00:25:52,477
check on the other. Uh that's
why there is no such thing as

460
00:25:52,560 --> 00:25:56,237
an economy. It's a political economy.
A democracy controls ultimately what

461
00:25:56,320 --> 00:26:00,597
happens in a market and then technology
sort of tries to disrupt the two and

462
00:26:00,680 --> 00:26:02,597
then you keep sort of the checks and

463
00:26:02,680 --> 00:26:04,557
balances. That's magical. It's one of

464
00:26:04,640 --> 00:26:07,757
the most unbelievable social constructs

465
00:26:07,840 --> 00:26:10,997
ever to emerge in the world, right?

466
00:26:11,080 --> 00:26:13,037
Think about sort of you know that it

467
00:26:13,120 --> 00:26:14,837
became the model.

468
00:26:14,920 --> 00:26:18,637
And we now need that same model

469
00:26:18,720 --> 00:26:23,997
to be redefined for this age, but it
will work because it worked the last

470
00:26:24,080 --> 00:26:28,117
time. And so therefore I think us
reminding ourselves that the balance the

471
00:26:28,200 --> 00:26:31,557
checks and uh that each
one has on the other is

472
00:26:31,640 --> 00:26:35,317
what I think we have a we have to aspire
for whether it's in San Francisco,

473
00:26:35,400 --> 00:26:38,397
whether it's in Washington D.C.
or quite frankly anywhere else.

474
00:26:38,480 --> 00:26:41,757
>> Mhm. Maybe just as a
last question, I'm still

475
00:26:41,840 --> 00:26:43,757
trying to like hone in on what I think

476
00:26:43,840 --> 00:26:46,197
Kevin might call how AGI pilled you are.

477
00:26:46,280 --> 00:26:50,877
Like there's a sense in the in Silicon
Valley that that it really is different

478
00:26:50,960 --> 00:26:54,157
this time and that the jagged frontier
is going to keep advancing forever, and

479
00:26:54,240 --> 00:26:57,077
all of a sudden the little tasks that
AI can automate today it are going to

480
00:26:57,160 --> 00:27:00,317
convert into full jobs.
How much do you buy that story?

481
00:27:00,400 --> 00:27:04,037
>> Look, I I buy that anything

482
00:27:04,120 --> 00:27:06,917
where the loops can be closed.

483
00:27:07,000 --> 00:27:08,917
Right? Like coding.

484
00:27:09,000 --> 00:27:11,917
In fact, AI research

485
00:27:12,000 --> 00:27:15,117
is sort of possible to close. I think we

486
00:27:15,200 --> 00:27:17,877
have now got sufficient um I'd say

487
00:27:17,960 --> 00:27:21,117
evidence uh of that.

488
00:27:21,200 --> 00:27:25,197
But is that enough? I don't think so.

489
00:27:25,280 --> 00:27:28,237
Um in in and when I think about

490
00:27:28,320 --> 00:27:29,837
you know, people talk about how

491
00:27:29,920 --> 00:27:32,837
verifiable is this task. And in the

492
00:27:32,920 --> 00:27:36,917
messy real world of even knowledge work,

493
00:27:37,000 --> 00:27:40,077
uh just say I'm going to look at the

494
00:27:40,160 --> 00:27:43,597
traces of human activity

495
00:27:43,680 --> 00:27:47,357
uh is enough to close the loop.
I don't think so. That I think is the the

496
00:27:47,440 --> 00:27:49,957
challenge, which is when I am in a

497
00:27:50,040 --> 00:27:51,437
meeting,

498
00:27:51,520 --> 00:27:56,237
I say things, I note things, um I may

499
00:27:56,320 --> 00:28:00,117
observe things, but what I do with it is

500
00:28:00,200 --> 00:28:03,077
not a trace today that I can orally my

501
00:28:03,160 --> 00:28:06,597
way in, right? And that to me is where

502
00:28:06,680 --> 00:28:10,237
we're selling short what is um I would

503
00:28:10,320 --> 00:28:13,157
say unverifiable part of the human

504
00:28:13,240 --> 00:28:16,357
capital. And so to me, that's where So,

505
00:28:16,440 --> 00:28:18,877
I believe the advances keep happening.

506
00:28:18,960 --> 00:28:21,597
I still am in the more in the world of

507
00:28:21,680 --> 00:28:24,237
hey, this is platforms tools. Very

508
00:28:24,320 --> 00:28:27,397
powerful, very disruptive. I have a lot

509
00:28:27,480 --> 00:28:29,757
of sort of I'd say, you know, uh

510
00:28:29,840 --> 00:28:31,797
humility to say a lot of things will

511
00:28:31,880 --> 00:28:34,477
change, but at the end of the day, so

512
00:28:34,560 --> 00:28:38,957
was electricity, so were a lot of other,
you know, steam when it first came out

513
00:28:39,040 --> 00:28:41,917
and what have you. And so I'm not sort
of sitting there and thinking this is

514
00:28:42,000 --> 00:28:44,797
the last technology we ever will invent.

515
00:28:44,880 --> 00:28:46,917
I I buy that. I I I kind of feel like

516
00:28:47,000 --> 00:28:48,997
yeah, this is in the pantheon of all

517
00:28:49,080 --> 00:28:54,077
technologies a big step up. Uh but I
do think that it you know, there will be

518
00:28:54,160 --> 00:28:57,317
more to come. >> Very good. >> Well,
Satya Nadella, thanks so much for

519
00:28:57,400 --> 00:29:00,357
joining us. Please give
Satya Nadella a hand.

520
00:29:00,440 --> 00:29:01,477
>> Thank you. >> Thank you so much.

521
00:29:01,560 --> 00:29:04,240
>> Thank you so much. Thank you.

522
00:29:06,250 --> 00:29:08,270
>> [music]

523
00:29:10,840 --> 00:29:12,860
[music]