1
00:00:00,080 --> 00:00:01,997
Microsoft has a new version of its

2
00:00:02,080 --> 00:00:05,197
co-pilot AI assistant that merges the

3
00:00:05,280 --> 00:00:09,197
consumer and workplace versions into one
product that's just aimed at corporate

4
00:00:09,280 --> 00:00:13,117
customers. It's an interesting week
as we talked about uh with MetaMuse coming

5
00:00:13,200 --> 00:00:15,677
out for this also news to drop. Joining

6
00:00:15,760 --> 00:00:17,677
us for more Dedra Bosa she's bringing a

7
00:00:17,760 --> 00:00:22,797
new show to Yahoo Finance and just
spoke to Microsoft CEO Satcha Nadella and

8
00:00:22,880 --> 00:00:25,117
she's joining us now. Hey D, good morning.

9
00:00:25,200 --> 00:00:28,557
>> Hey guys. Well, first of all, I'm
thrilled to be here. my first appearance

10
00:00:28,640 --> 00:00:33,117
on Yahoo Finance, but of course longtime
watcher and listeners. So, I was up in

11
00:00:33,200 --> 00:00:37,277
Redwood, Seattle earlier this week
talking to Sati Nadella and the

12
00:00:37,360 --> 00:00:41,197
Microsoft team and I got a preview
of this new co-pilot and Microsoft had to

13
00:00:41,280 --> 00:00:42,397
do something special here, right?

14
00:00:42,480 --> 00:00:45,917
Because even though Sati Nadella says
that the adoption numbers, it's about 30

15
00:00:46,000 --> 00:00:49,437
million customers so far, he's happy
with that. They really want to be the

16
00:00:49,520 --> 00:00:52,237
operating layer for enterprise AI. So,

17
00:00:52,320 --> 00:00:54,637
this new co-pilot has a few different

18
00:00:54,720 --> 00:00:57,757
modes. And the bet is essentially that

19
00:00:57,840 --> 00:01:01,997
companies shouldn't have to watch the
model race, right? It moves too quickly.

20
00:01:02,080 --> 00:01:05,517
Even for someone like me that's living
here in San Francisco, every week

21
00:01:05,600 --> 00:01:08,317
there's another model that's on top
and you got to switch and figure out what

22
00:01:08,400 --> 00:01:12,957
that is. Microsoft is betting, you know,
just come to us, come to Copilot. We'll

23
00:01:13,040 --> 00:01:16,397
figure it out for you. We have access
to all of the models and we'll serve you

24
00:01:16,480 --> 00:01:20,637
the right one for your task. So hard
model for a hard task, you know, more

25
00:01:20,720 --> 00:01:25,917
capable, cheaper model for an easier task.
The new co-pilot has three modes.

26
00:01:26,000 --> 00:01:29,037
Chat, which is like your typical chatbot,
code, which is kind of like a

27
00:01:29,120 --> 00:01:33,357
co-working where you can vibe code
and do other kinds of coding. And then

28
00:01:33,440 --> 00:01:37,837
autopilot, which I really think stole
the show in terms of this new co-pilot.

29
00:01:37,920 --> 00:01:42,317
I'm calling it kind of like Muse
for business because it's very intuitive.

30
00:01:42,400 --> 00:01:44,237
that's already connected to all of your

31
00:01:44,320 --> 00:01:47,437
apps, all of your email, all of your

32
00:01:47,520 --> 00:01:51,917
workflows. It kind of just works like
a personal assistant on your behalf. So, I

33
00:01:52,000 --> 00:01:55,197
asked Ache Nadella about this, kind
of told him that I thought it stole the

34
00:01:55,280 --> 00:01:57,197
show. Here's how he described it.

35
00:01:57,280 --> 00:01:59,197
>> Think about it, right?
We launched uh in

36
00:01:59,280 --> 00:02:02,157
fact co-pilot task um just around the

37
00:02:02,240 --> 00:02:04,397
same time as open clock came out and we

38
00:02:04,480 --> 00:02:08,077
saw a ton of activity uh in the consumer

39
00:02:08,160 --> 00:02:11,117
space. uh but you know autopilots are

40
00:02:11,200 --> 00:02:14,237
fantastic until you have to do something

41
00:02:14,320 --> 00:02:16,877
with the output of an autopilot. So this

42
00:02:16,960 --> 00:02:21,757
is where I don't think of any one form
factor as the be all end all right this

43
00:02:21,840 --> 00:02:25,197
is where like I'll give you another
example right even with co-work and

44
00:02:25,280 --> 00:02:28,477
excel right coowork can create fantastic

45
00:02:28,560 --> 00:02:31,357
spreadsheets but at some point I want to

46
00:02:31,440 --> 00:02:33,277
reason and that's the so there's an

47
00:02:33,360 --> 00:02:36,317
outer loop uh that's kind of co-work

48
00:02:36,400 --> 00:02:40,717
inner loop is the agent loop inside
of Excel same thing with autopilot I think

49
00:02:40,800 --> 00:02:45,277
of autopilot as something that has
an identity it does complete jobs but then

50
00:02:45,360 --> 00:02:47,517
when it hands off or when I am trying to

51
00:02:47,600 --> 00:02:52,397
instruct it, I need my co-pilot uh
and my cowork and so this is where I don't

52
00:02:52,480 --> 00:02:54,237
think of this as any one thing but it's

53
00:02:54,320 --> 00:02:57,277
the composition but to your point having

54
00:02:57,360 --> 00:03:00,717
autopilot is about that next generation

55
00:03:00,800 --> 00:03:03,597
of I would say automation uh in the

56
00:03:03,680 --> 00:03:07,837
enterprise that I think is going to have
very high ROI for uh businesses adopting

57
00:03:07,920 --> 00:03:12,637
it. So this autopilot mode which I think
a lot of enterprises are going to be

58
00:03:12,720 --> 00:03:16,477
interested in that rules out on the
coming weeks and sort of asked Nadella

59
00:03:16,560 --> 00:03:20,397
you know why not give this to people
sooner right away since you know it

60
00:03:20,480 --> 00:03:23,757
looks useful and people are getting used
to sort of these consumer habits having

61
00:03:23,840 --> 00:03:27,357
an agent with Muse and Instinct
and others. Um I heard your guys

62
00:03:27,440 --> 00:03:29,277
conversation just before I came on. You

63
00:03:29,360 --> 00:03:31,517
were talking about Muse and how you

64
00:03:31,600 --> 00:03:34,877
needed to trust a corporation to

65
00:03:34,960 --> 00:03:38,557
actually hand over all of your data
and that is something that Microsoft is

66
00:03:38,640 --> 00:03:40,637
absolutely thinking about really part of

67
00:03:40,720 --> 00:03:43,517
its core you know mission is to make

68
00:03:43,600 --> 00:03:48,237
sure that the trust the permissions the
security is in place here's what Nadella

69
00:03:48,320 --> 00:03:52,557
had to say about that >> trust
is going to be the maximum biggest

70
00:03:52,640 --> 00:03:56,797
issue for us in fact it's you're going
to hit that on the consumer side uh in

71
00:03:56,880 --> 00:03:59,677
fact if I look back at even
our experience experience in the consumer

72
00:03:59,760 --> 00:04:02,957
space. That is it, right? Okay.
Can I really trust it with all of my

73
00:04:03,040 --> 00:04:05,117
credentials when it does autonomous

74
00:04:05,200 --> 00:04:07,917
activity? Uh how do I make sure that

75
00:04:08,000 --> 00:04:12,637
this is something that I can feel that
I'm in control? I can audit. And in the

76
00:04:12,720 --> 00:04:16,077
enterprise, this is everything. That's
why when you even think about copilot,

77
00:04:16,160 --> 00:04:19,517
we don't think about it in in an
isolation like any form factor like

78
00:04:19,600 --> 00:04:23,197
autopilot. Agent 365 is probably
the most important. You know, it's the

79
00:04:23,280 --> 00:04:26,957
product that is getting the fastest
adoption in the enterprise. that ability

80
00:04:27,040 --> 00:04:28,957
to observe everything, govern

81
00:04:29,040 --> 00:04:34,077
everything, set policy, have security,
have the phops. By the way, the other

82
00:04:34,160 --> 00:04:35,757
thing that you'll notice is that this

83
00:04:35,840 --> 00:04:38,557
thing once the subsidies stop for

84
00:04:38,640 --> 00:04:41,117
everybody, uh, this is going to there's

85
00:04:41,200 --> 00:04:45,597
marginal cost to tokens. And so one
of the core business model realizations is

86
00:04:45,680 --> 00:04:47,677
that this is there's a certain amount of

87
00:04:47,760 --> 00:04:49,837
subsidy anyone can do but at the end of

88
00:04:49,920 --> 00:04:54,877
the day uh this all has to be about
the return that the user gets whether it's

89
00:04:54,960 --> 00:04:58,077
in the consumer space
or in the enterprise space.

90
00:04:58,160 --> 00:05:00,877
>> So I thought that was one of the most
interesting parts of the interview. He

91
00:05:00,960 --> 00:05:04,877
says when the subsidies stop and that's
part of the reason that they're not just

92
00:05:04,960 --> 00:05:07,197
rolling autopilot out immediately to

93
00:05:07,280 --> 00:05:12,317
everyone because of course requires
tokens requires compute can be expensive

94
00:05:12,400 --> 00:05:15,597
and Sachi Nadella at the end he says we
got to make sure that our customers have

95
00:05:15,680 --> 00:05:18,797
ROI their return on investment um and

96
00:05:18,880 --> 00:05:23,197
Nadella's always been one of these
CEOs that has to create value I mean of

97
00:05:23,280 --> 00:05:26,637
course they all do but for the enterprise
so it was just interesting to

98
00:05:26,720 --> 00:05:29,837
hear him speak so candidly like that and

99
00:05:29,920 --> 00:05:33,837
I'm sure you as co-pilot the new one
rolls out um we'll hear from different

100
00:05:33,920 --> 00:05:36,797
companies will it be enough to take them

101
00:05:36,880 --> 00:05:39,837
away from you know an anthropic or open

102
00:05:39,920 --> 00:05:41,917
AI contract I mean that is the real

103
00:05:42,000 --> 00:05:46,877
question the battle in AI over
enterprise is extremely fierce and it's

104
00:05:46,960 --> 00:05:49,997
extremely lucrative >> and you're
proious you know China top of

105
00:05:50,080 --> 00:05:52,877
mind obviously with regards to AI
and things like that you trade talks going

106
00:05:52,960 --> 00:05:56,717
on you had the big state dinner with
Nadella in attendance I believe so did

107
00:05:56,800 --> 00:06:00,957
he have anything to say about sort
of AI and proliferation with China.

108
00:06:01,040 --> 00:06:03,837
>> Yeah. And you know it's a huge topic
here especially as we're talking about

109
00:06:03,920 --> 00:06:07,517
safety and alignment and if the US AI

110
00:06:07,600 --> 00:06:11,917
industry slows down is China going to do
that with Xiinping coming over here this

111
00:06:12,000 --> 00:06:16,237
week. Um you know questions is there
room for them to agree on sort of some

112
00:06:16,320 --> 00:06:20,557
basic stuff. So certainly did talk
to Nadella about that and you know he was

113
00:06:20,640 --> 00:06:22,957
optimistic. I asked him you know what

114
00:06:23,040 --> 00:06:25,037
needs to happen and he thinks that it's

115
00:06:25,120 --> 00:06:27,837
in both countries interest to get on the

116
00:06:27,920 --> 00:06:32,077
same page at least when it comes
to notification. So if there is a cyber

117
00:06:32,160 --> 00:06:35,917
security attack, there is sort of a
direct line from president to president

118
00:06:36,000 --> 00:06:39,677
to say this is what's happening similar
to what happens in the military right

119
00:06:39,760 --> 00:06:43,677
because AI is that powerful.
Uh have a listen to what he said.

120
00:06:43,760 --> 00:06:47,997
>> I would sort of encourage uh the US
and China because I think both sides will

121
00:06:48,080 --> 00:06:50,957
care about the same set of things at the
end of the day which is they want to be

122
00:06:51,040 --> 00:06:55,677
able to have this technology actually
help their citizens get value. Uh so

123
00:06:55,760 --> 00:06:59,917
diffusion is sort of one thing that both
sides will care about and then to be

124
00:07:00,000 --> 00:07:03,917
able to do that safely uh is also going
to be important to both sides. So the

125
00:07:04,000 --> 00:07:06,397
more there are norms that they set uh

126
00:07:06,480 --> 00:07:09,277
and the US in particular can lead uh the

127
00:07:09,360 --> 00:07:13,997
better off we are as a world.
>> So here I'll be a little bit skeptical.

128
00:07:14,080 --> 00:07:15,517
Well, I mean, we're a long way from

129
00:07:15,600 --> 00:07:18,797
actions and I think a set of norms, but

130
00:07:18,880 --> 00:07:23,757
if we get sort of notifications,
a direct line of communication out of the

131
00:07:23,840 --> 00:07:27,837
US China summit this week, that's
positive. You guys saw how many of

132
00:07:27,920 --> 00:07:31,757
American business leaders and politicians
and important people were in

133
00:07:31,840 --> 00:07:33,517
the room at last night's state dinner.

134
00:07:33,600 --> 00:07:37,197
Nadella was there. So some progress um

135
00:07:37,280 --> 00:07:39,277
and Nadella seemed you know fairly

136
00:07:39,360 --> 00:07:42,397
optimistic certainly that incentives on

137
00:07:42,480 --> 00:07:46,477
alignment and safety um would be sort
of similar for both countries and there

138
00:07:46,560 --> 00:07:48,637
would be at least some takeaways.

139
00:07:48,720 --> 00:07:51,837
>> Dearra we've seen a growing trend
of states starting to really look at

140
00:07:51,920 --> 00:07:53,837
regulation of the data center buildout.

141
00:07:53,920 --> 00:07:56,957
You look at the one-year moratorium we
have in New York on hypers scale data

142
00:07:57,040 --> 00:08:01,837
centers. You look at the uh pipeline
audit project audits we see in Texas

143
00:08:01,920 --> 00:08:05,357
under Greg Abbott. And then setting
aside the governments, you just have

144
00:08:05,440 --> 00:08:08,797
kind of the plain push back from the
everyday American who says, "I don't

145
00:08:08,880 --> 00:08:12,637
want this data center in my backyard.
I don't want my power bills to go up. I

146
00:08:12,720 --> 00:08:16,157
don't want my water bill to go up.
I don't want the lights." What's your

147
00:08:16,240 --> 00:08:19,757
sense of how Satia and Nadella
is thinking about that and what how that

148
00:08:19,840 --> 00:08:24,397
might impact Microsoft's plans as they
look to fund this build out and push

149
00:08:24,480 --> 00:08:27,917
themselves forward? You know,
I don't remember the gap ever

150
00:08:28,000 --> 00:08:31,117
being larger between how the technology

151
00:08:31,200 --> 00:08:35,677
industry feels about something this big,
the data center buildout that is going

152
00:08:35,760 --> 00:08:39,197
to be the backbone that is going
to power the AI revolution and how sort of

153
00:08:39,280 --> 00:08:43,037
the average American feels. So, when
I asked Nadella about this, he kind of

154
00:08:43,120 --> 00:08:47,357
gave me the answer that you hear from
a lot of the big tech CEOs. you know,

155
00:08:47,440 --> 00:08:50,717
we're providing benefits. We're
investing back in the communities. But

156
00:08:50,800 --> 00:08:54,557
something that Microsoft is doing
differently, and I, you know, I gave him

157
00:08:54,640 --> 00:08:56,557
credit for this, they sort of backed

158
00:08:56,640 --> 00:08:58,877
away from the NDA, the non-disclosure

159
00:08:58,960 --> 00:09:03,997
agreement that tech has really had in
place when they build these data centers

160
00:09:04,080 --> 00:09:08,637
and build in different communities.
And I asked him, has that been effective?

161
00:09:08,720 --> 00:09:11,677
And he says, and will it be the industry
norm? Right? That's what I was curious

162
00:09:11,760 --> 00:09:14,157
about. will Google will um the other

163
00:09:14,240 --> 00:09:18,957
hyperscalers will Amazon have to follow
and you know be more transparent which

164
00:09:19,040 --> 00:09:23,357
is what the communities want and you
said he thought so. So Microsoft is you

165
00:09:23,440 --> 00:09:28,077
know in my opinion kind of leading the
way on this. Will it be enough? That is

166
00:09:28,160 --> 00:09:32,237
another question because you guys
know very well that this is just such a

167
00:09:32,320 --> 00:09:36,477
polarizing issue such a hot button
issue for the communities where these are

168
00:09:36,560 --> 00:09:38,397
being built. >> Yeah.
I mean the NDA it seems like the

169
00:09:38,480 --> 00:09:42,237
NDAs aren't really an option anymore
because the communities will not allow

170
00:09:42,320 --> 00:09:46,077
it effectively because
there's been such opposition.

171
00:09:46,160 --> 00:09:49,837
>> Yeah. And you know something that he
talked about and you know Dina Pal

172
00:09:49,920 --> 00:09:53,437
McCormack at Meta has talked about
and the others as well is really you know

173
00:09:53,520 --> 00:09:55,677
can giving back to these communities

174
00:09:55,760 --> 00:09:57,757
building hospitals building schools

175
00:09:57,840 --> 00:10:01,197
putting money into them. Um it sounds

176
00:10:01,280 --> 00:10:04,157
sounds great in practice. Um but you

177
00:10:04,240 --> 00:10:08,397
know these t these things take time.
you have a buildout, you have, you know,

178
00:10:08,480 --> 00:10:11,117
sound in some cases that affect the

179
00:10:11,200 --> 00:10:12,557
community. So, these are really real

180
00:10:12,640 --> 00:10:16,157
issues and I think that tech as a whole,

181
00:10:16,240 --> 00:10:18,717
the hyperscalers, the neoclouds, they're

182
00:10:18,800 --> 00:10:21,437
having to think about this in new ways

183
00:10:21,520 --> 00:10:23,197
because I'm not sure that anyone

184
00:10:23,280 --> 00:10:26,477
predicted just how controversial this

185
00:10:26,560 --> 00:10:29,037
would be a few years ago. But to be able

186
00:10:29,120 --> 00:10:32,557
to, you know, build more gigawatts of

187
00:10:32,640 --> 00:10:35,597
data of compute in America, they have to

188
00:10:35,680 --> 00:10:40,080
figure this piece out.
It's absolutely essential.