1
00:00:00,430 --> 00:00:03,417
Super excited to be here
with Dario and Jensen.

2
00:00:03,500 --> 00:00:06,187
What both of you are doing
- NVIDIA at the silicon layer

3
00:00:06,270 --> 00:00:10,557
and Anthropic at the cognition
layer is shaping what every developer

4
00:00:10,640 --> 00:00:14,497
and every organization will be
able to build on going forward.

5
00:00:14,580 --> 00:00:18,507
On our side, we have been scaling
NVIDIA across this fungible Azure fleet

6
00:00:18,590 --> 00:00:20,737
at the speed of light.
Something that Jensen's been

7
00:00:20,820 --> 00:00:25,107
talking to me about for multiple years
now, and we're continuing to set pace,

8
00:00:25,190 --> 00:00:28,617
including with the AI superfactory
we announced just last week.

9
00:00:28,700 --> 00:00:31,477
And we have been deepening
our partnership with Anthropic as well,

10
00:00:31,560 --> 00:00:35,117
incorporating their models
across the entire Copilot family.

11
00:00:35,200 --> 00:00:39,357
And today, I'm excited to share
that we are taking another step.

12
00:00:39,440 --> 00:00:41,957
We are increasingly going
to be customers of each other.

13
00:00:42,040 --> 00:00:45,367
We will use Anthropic models.
They will use our infrastructure

14
00:00:45,450 --> 00:00:49,967
and we'll go to market together to help
our customers realize the value of AI.

15
00:00:50,050 --> 00:00:53,867
This announcement
encompasses four key things:

16
00:00:53,950 --> 00:00:58,207
First, customers of Microsoft
Foundry will be able to access

17
00:00:58,290 --> 00:01:02,117
Anthropic's Claude models.
Second, we are continuing access

18
00:01:02,200 --> 00:01:04,777
for Claude across our Copilot family.

19
00:01:04,860 --> 00:01:08,087
Third, Anthropic is committing
to Azure capacity.

20
00:01:08,170 --> 00:01:12,027
And finally, NVIDIA and Anthropic
are also establishing a partnership

21
00:01:12,110 --> 00:01:14,387
to support Anthropic's future growth.

22
00:01:14,470 --> 00:01:17,997
For us, this is all about
deepening our commitment to bringing

23
00:01:18,080 --> 00:01:22,467
the best infrastructure, model choice,
and applications to our customers.

24
00:01:22,550 --> 00:01:26,337
And of course, this all builds
on the partnership we have with OpenAI,

25
00:01:26,420 --> 00:01:29,807
which remains a critical
partner for Microsoft and OpenAI

26
00:01:29,890 --> 00:01:33,147
and our customers and provides
more innovation and choice.

27
00:01:33,230 --> 00:01:36,517
With that, Dario, why don't
you tell us a little more about

28
00:01:36,600 --> 00:01:38,977
what this partnership means to Anthropic?

29
00:01:39,060 --> 00:01:43,257
Yes, in terms of Microsoft,
I think, as you mentioned, Satya,

30
00:01:43,340 --> 00:01:47,527
both of us believe in choice.
And we're excited to bring our models

31
00:01:47,610 --> 00:01:51,397
as a choice to Microsoft Azure.

32
00:01:51,480 --> 00:01:55,397
And Anthropic will be the first
model that is available

33
00:01:55,480 --> 00:01:59,137
on all three of the biggest clouds.

34
00:01:59,220 --> 00:02:06,047
Second, Microsoft has a reputation
as a strong enterprise company,

35
00:02:06,130 --> 00:02:09,847
and Anthropic does as well. So we
have the opportunity to work together,

36
00:02:09,930 --> 00:02:14,487
to go to market together and to provide
intelligence to the world together.

37
00:02:14,570 --> 00:02:18,687
And we're excited to accelerate
the diffusion of this technology

38
00:02:18,770 --> 00:02:22,057
as the technology continues to improve.

39
00:02:22,140 --> 00:02:27,527
And finally, we're very excited
to get additional capacity,

40
00:02:27,610 --> 00:02:32,367
that we can use both to train our models
to support Microsoft first-party products

41
00:02:32,450 --> 00:02:36,507
and to sell together. And that brings
me to the NVIDIA part of it.

42
00:02:36,590 --> 00:02:42,677
We are very excited to add
substantial support and substantial use

43
00:02:42,760 --> 00:02:46,417
of NVIDIA's accelerators
for use by Anthropic.

44
00:02:46,500 --> 00:02:51,057
NVIDIA has led the way in this field
in many ways, has helped make

45
00:02:51,140 --> 00:02:55,157
this entire AI boom possible. We think
this is going to be the beginning,

46
00:02:55,240 --> 00:02:57,897
just the beginning
of a very long partnership.

47
00:02:57,980 --> 00:03:02,667
We are excited to work together
to co-optimize models together,

48
00:03:02,750 --> 00:03:06,267
starting with Blackwell
and then moving on to Vera Rubin.

49
00:03:06,350 --> 00:03:09,877
We're excited to announce
up to a gigawatt of capacity.

50
00:03:09,960 --> 00:03:12,847
And that's just for now.
That's just where we're starting.

51
00:03:12,930 --> 00:03:17,177
And so, we're very excited to continue
that, to continue the co-optimization

52
00:03:17,260 --> 00:03:21,247
to further build out NVIDIA's
already incredible ecosystem.

53
00:03:21,330 --> 00:03:23,357
That's fantastic. Well said, Dario.

54
00:03:23,440 --> 00:03:26,227
And one of our core beliefs
is that you can't make progress

55
00:03:26,310 --> 00:03:28,597
just in one layer of the stack.

56
00:03:28,680 --> 00:03:33,927
You have to advance every layer - silicon,
systems, models, applications,

57
00:03:34,010 --> 00:03:38,297
while optimizing effectively for all
the things that customers care about,

58
00:03:38,380 --> 00:03:43,937
COGS, latency, performance. And one of the
things that we are also establishing today

59
00:03:44,020 --> 00:03:47,707
is this new partnership, as you described,
between NVIDIA and Anthropic.

60
00:03:47,790 --> 00:03:52,160
So Jensen, maybe you should talk a little
bit about what you all are excited about.

61
00:03:52,730 --> 00:03:57,357
Thanks, Satya. You know, NVIDIA's
DNA is to build the most advanced

62
00:03:57,440 --> 00:04:03,197
computing systems in the world and to
accelerate the most challenging workloads

63
00:04:03,280 --> 00:04:07,067
in the world on the most
important platforms in the world.

64
00:04:07,150 --> 00:04:12,467
And this conference call right
here embodies that very thing.

65
00:04:12,550 --> 00:04:14,837
This is a dream come true for us.

66
00:04:14,920 --> 00:04:20,007
We've admired the work of Anthropic
and Dario for a long time,

67
00:04:20,090 --> 00:04:23,617
and this is the first time
we are going to deeply partner

68
00:04:23,700 --> 00:04:27,187
with Anthropic to accelerate Claude.

69
00:04:27,270 --> 00:04:29,787
I can't wait to go accelerate Claude.

70
00:04:29,870 --> 00:04:35,827
The work that Anthropic has done,
the seminal work in AI safety,

71
00:04:35,910 --> 00:04:42,127
the advances of Claude code,
the engineers of NVIDIA love Claude code.

72
00:04:42,210 --> 00:04:46,267
The fact that you can go and literally
refactor your code for you,

73
00:04:46,350 --> 00:04:48,367
I mean, it's pretty an amazing thing.

74
00:04:48,450 --> 00:04:54,407
And the work on MCP,
the Model Context Protocol,

75
00:04:54,490 --> 00:04:58,400
has completely revolutionized
the agentic AI landscape.

76
00:04:58,670 --> 00:05:03,587
And so the contributions of Anthropic,
the advanced research that is done there,

77
00:05:03,670 --> 00:05:09,257
the incredible researchers,
the incredible infrastructure team

78
00:05:09,340 --> 00:05:12,297
that works at Anthropic, that makes
it possible for you to scale up

79
00:05:12,380 --> 00:05:15,627
to what you have already done,
is really quite phenomenal.

80
00:05:15,710 --> 00:05:20,037
And now your business is on a rocket ship.

81
00:05:20,120 --> 00:05:23,437
It's just scaling so incredibly.

82
00:05:23,520 --> 00:05:29,277
And so, I can't wait to go accelerate
Claude and Grace Blackwell with NVLink.

83
00:05:29,360 --> 00:05:33,017
I'm really, really hoping
for an order of magnitude speed up,

84
00:05:33,100 --> 00:05:37,957
and that's going to help you scale
even faster, drive down token economics,

85
00:05:38,040 --> 00:05:42,297
and really make it possible
for us to spread AI everywhere.

86
00:05:42,380 --> 00:05:45,227
And so I'm really, really
super excited about that.

87
00:05:45,310 --> 00:05:49,597
Now, the work that we've done
with Microsoft over the years, Satya,

88
00:05:49,680 --> 00:05:54,507
it's broad and deep. I mean,
it's incredible all the things that we do,

89
00:05:54,590 --> 00:05:59,347
the work that we've done to left,
shift, shift left, all of our engineering

90
00:05:59,430 --> 00:06:04,677
so that at the moment we have new
technology, it appears on Azure.

91
00:06:04,760 --> 00:06:11,187
Notice the scale that we've now already
achieved with Grace Blackwell 200 and 300,

92
00:06:11,270 --> 00:06:15,927
the number of systems that are
already out there helping researchers

93
00:06:16,010 --> 00:06:20,697
pioneer the next frontier in AI,
really fantastic work there.

94
00:06:20,780 --> 00:06:25,607
But we do everything from data processing,
to search, to image recognition,

95
00:06:25,690 --> 00:06:29,907
to fraud detection, to all
kinds of stuff that we're doing,

96
00:06:29,990 --> 00:06:35,307
from classical computing, to classical ML,

97
00:06:35,390 --> 00:06:38,017
to generative AI, to agentic AI.

98
00:06:38,100 --> 00:06:42,117
The work that we're doing just
spans the entire range of technology.

99
00:06:42,200 --> 00:06:47,327
But what's really incredible is,
of course, Microsoft has the world's best

100
00:06:47,410 --> 00:06:51,927
enterprise go-to-market.
This is the next giant frontier -

101
00:06:52,010 --> 00:06:55,667
enterprise and industrial AI.
That's where, as you know,

102
00:06:55,750 --> 00:06:59,497
the vast majority of world's
economics economy is.

103
00:06:59,580 --> 00:07:04,907
And in order for us to get to every single
enterprise, the enterprise go-to-market

104
00:07:04,990 --> 00:07:09,747
takes decades to build up.
It's not one of those things,

105
00:07:09,830 --> 00:07:12,617
that just because you put it
on the cloud, you're going to be able

106
00:07:12,700 --> 00:07:14,347
to serve the world's enterprise.

107
00:07:14,430 --> 00:07:16,587
The enterprise go-to-market
is very complicated.

108
00:07:16,670 --> 00:07:22,127
And this is where the two of us have such
great harmony because NVIDIA's computing

109
00:07:22,210 --> 00:07:27,427
is in every enterprise. And we're in every
enterprise in every single country.

110
00:07:27,510 --> 00:07:31,997
Now, this partnership of the three
of us will be able to bring AI,

111
00:07:32,080 --> 00:07:35,607
bring Claude to every enterprise,
to every industry around the world.

112
00:07:35,690 --> 00:07:37,807
And so this is a really exciting time.

113
00:07:37,890 --> 00:07:42,307
I'll close with this - as an industry,
we really need to move beyond

114
00:07:42,390 --> 00:07:46,287
any type of zero-sum narrative
or winner-take-all hype.

115
00:07:46,370 --> 00:07:49,687
What's required now
is the hard work of building

116
00:07:49,770 --> 00:07:54,627
broad, durable capabilities together
so that this technology can deliver

117
00:07:54,710 --> 00:08:00,927
real, tangible, local success for every
country, every sector, and every customer.

118
00:08:01,010 --> 00:08:04,137
The opportunity is simply too
big to approach any other way.

119
00:08:04,220 --> 00:08:06,767
Jensen and Dario,
any last thoughts on this?

120
00:08:06,850 --> 00:08:11,777
I'm just excited to, you know, take a
large number of chips and use it to serve

121
00:08:11,860 --> 00:08:15,717
our mutual enterprise customers
together to make the smartest models,

122
00:08:15,800 --> 00:08:18,787
the models that run the fastest,
for the lowest possible cost.

123
00:08:18,870 --> 00:08:25,057
Jensen? You know, I think the world
is just barely realizing where we are

124
00:08:25,140 --> 00:08:29,627
in the AI journey. You know,
we're seeing three scaling laws

125
00:08:29,710 --> 00:08:34,997
happening at the same time. Pre-training
is scaling incredibly well still.

126
00:08:35,080 --> 00:08:39,037
Post-training, the more compute
you give it, the smarter the AI.

127
00:08:39,120 --> 00:08:42,607
And then of course, inference
time scaling, test time scaling,

128
00:08:42,690 --> 00:08:46,337
the more the AI thinks,
the higher the quality of answer.

129
00:08:46,420 --> 00:08:50,747
And so we're now at a point
with AI where it is very clear that

130
00:08:50,830 --> 00:08:55,287
the more compute we give it, the more
cost-effective compute we give it,

131
00:08:55,370 --> 00:08:59,827
the smarter the tokens,
the smarter the AI is going to be.

132
00:08:59,910 --> 00:09:05,427
And the smarter the AI, the more
adoption, both in the new applications

133
00:09:05,510 --> 00:09:08,267
that integrate these AI APIs

134
00:09:08,350 --> 00:09:12,597
and how more frequently you use it.

135
00:09:12,680 --> 00:09:17,707
And so the quality of these AI models now
have really reached an inflection point.

136
00:09:17,790 --> 00:09:24,047
And so we've got these two simultaneous,
exponentially increasing compute demand,

137
00:09:24,130 --> 00:09:27,947
and I guess the thing that's really
great is they're going to need

138
00:09:28,030 --> 00:09:32,917
a lot more Azure compute resources,
and they're going to need a lot more GPUs.

139
00:09:33,000 --> 00:09:38,757
And we're just delighted to partner
with you, Dario, to bring AI to the world.

140
00:09:38,840 --> 00:09:42,297
Absolutely. Thank you
so much Dario and Jensen.

141
00:09:42,380 --> 00:09:45,067
I'm really looking forward to everything
we're going to build together,

142
00:09:45,150 --> 00:09:49,237
and more importantly, how customers
can benefit from all this innovation,

143
00:09:49,320 --> 00:09:52,437
and really thrive with their
business and their outcomes.

144
00:09:52,520 --> 00:09:54,447
So thank you all for joining today.

145
00:09:54,530 --> 00:09:56,000
Thank you.