1
00:00:00,000 --> 00:00:02,757
excitement here has just
been incredible to watch. Uh,

2
00:00:02,840 --> 00:00:05,837
and it's not just ChatGPT, Codex
is the fastest growing Codex market in the

3
00:00:05,920 --> 00:00:11,077
world now, too. Um, but the energy here
and what people are building is, uh,

4
00:00:11,160 --> 00:00:14,397
is remarkable. >> Okay.
And in terms of, uh, the the

5
00:00:14,480 --> 00:00:18,117
potential of this market, I know you've
got the the Go plan here, uh, a low-cost

6
00:00:18,200 --> 00:00:21,077
plan. How do you plan to sort
of monetize people in this market as well

7
00:00:21,160 --> 00:00:22,957
as you continue to expand here?

8
00:00:23,040 --> 00:00:27,317
>> Uh, Go is one thing that we're we
really designed with India in mind.

9
00:00:27,400 --> 00:00:29,197
It's a market that clearly needed it.

10
00:00:29,280 --> 00:00:33,997
Um, and the enterprise adoption, I think,
will be another very strong angle.

11
00:00:34,080 --> 00:00:36,757
>> In terms of the infrastructure front,
you have this partnership with Tata as

12
00:00:36,840 --> 00:00:40,557
well, uh, and the data center.
Also wanted to get your take on what the

13
00:00:40,640 --> 00:00:44,077
plans are for Stargate in the India
market. Um, when is that happening? I

14
00:00:44,160 --> 00:00:46,877
think we had the IT minister this week
speaking and saying it's happening. Uh,

15
00:00:46,960 --> 00:00:47,997
can you give us some updates on that?

16
00:00:48,080 --> 00:00:53,037
>> We are working, uh, as hard and as
fast as we can to build sufficient AI

17
00:00:53,120 --> 00:00:55,477
infrastructure in India
to serve the market here.

18
00:00:55,560 --> 00:01:00,197
Um, it's going pretty well. I don't know
the exact timelines on every project, but,

19
00:01:00,280 --> 00:01:02,597
uh, the country here
is motivated to build fast.

20
00:01:02,680 --> 00:01:04,317
>> Do you think we'll
get something this year?

21
00:01:04,400 --> 00:01:07,917
>> I hope so. >> Hm. Sam,
on on terms of AI competition,

22
00:01:08,000 --> 00:01:11,357
we know it's intense. There's a there's
a ton of players and and there's a lot

23
00:01:11,440 --> 00:01:14,637
of competition on the frontier at this
point. Last year you issued this this

24
00:01:14,720 --> 00:01:18,317
code red, uh, around the competition.
Do you feel you're in a much better place

25
00:01:18,400 --> 00:01:20,877
now than you were, say,
last back in the last year?

26
00:01:20,960 --> 00:01:24,437
>> Um, yeah, ChatGPT has been growing very

27
00:01:24,520 --> 00:01:28,357
strongly, uh, and I feel
great about the current

28
00:01:28,440 --> 00:01:31,477
models, the upcoming models. It's been
very fun to see the momentum with Codex

29
00:01:31,560 --> 00:01:35,037
as a new product. Like, ChatGPT
is great, uh, but, you know, always

30
00:01:35,120 --> 00:01:39,757
expected that to do well. The strength,
uh, of Codex winning the coding market,

31
00:01:39,840 --> 00:01:41,677
uh, has been a new
and really fun thing to watch.

32
00:01:41,760 --> 00:01:44,477
>> And and in the enterprise specifically,
I mean, you go up there against

33
00:01:44,560 --> 00:01:47,917
Anthropic, for example. Um,
the competition Are you managing to gain

34
00:01:48,000 --> 00:01:50,317
customers? Co-work, there
was a big deal made out of that.

35
00:01:50,400 --> 00:01:54,917
>> Yeah. >> And the impact it's having.
Um, >> Um, I think most of the enterprise

36
00:01:55,000 --> 00:01:58,477
momentum right now is still, honestly,
around coding, but I I hope and expect

37
00:01:58,560 --> 00:02:01,997
that over this coming year we'll start to
see it for knowledge work more generally.

38
00:02:02,080 --> 00:02:06,597
>> When you start to think about revenue
as well, Sam, ads, this is a new area

39
00:02:06,680 --> 00:02:10,277
you've only just started them.
What has been the initial feedback so far?

40
00:02:10,360 --> 00:02:13,597
>> So it's super early and I always
hesitate to talk when something's this

41
00:02:13,680 --> 00:02:18,117
early, but the anecdotally
the feedback has been pretty

42
00:02:18,200 --> 00:02:22,597
positive. I think we still have some
work to do to figure out the exact ad

43
00:02:22,680 --> 00:02:25,197
format um that's going
to work the best, but

44
00:02:25,280 --> 00:02:30,240
people have are like, oh, this is like,
you know, a cool thing. The

45
00:02:30,680 --> 00:02:34,677
The ads that I have personally liked the
most in recent years from tech have been

46
00:02:34,760 --> 00:02:39,357
sort of Instagram style ads um
where you discover something new that

47
00:02:39,440 --> 00:02:42,077
you might really like and otherwise
wouldn't have known about. I think we've

48
00:02:42,160 --> 00:02:45,037
got a real opportunity to push
in that direction with ads in ChatGPT.

49
00:02:45,120 --> 00:02:47,557
>> When you say there's some stuff to
figure out, what are the things that you

50
00:02:47,640 --> 00:02:50,477
hope to make improvements on? >> Oh,
we're still just like testing very

51
00:02:50,560 --> 00:02:53,917
different formats, you know,
where and how should an ad show up uh

52
00:02:54,000 --> 00:02:56,237
like that kind of tactical stuff.

53
00:02:56,320 --> 00:02:59,637
>> It is it something for I know it's
super early stage, but how much are you

54
00:02:59,720 --> 00:03:02,557
thinking about taking this to
other markets outside the US?

55
00:03:02,640 --> 00:03:06,477
>> Eventually we want to, but I
don't know any kind of timeline there.

56
00:03:06,560 --> 00:03:10,477
>> Uh another big potential new revenue
driver in the future for you, Sam, is

57
00:03:10,560 --> 00:03:13,357
around hardware. I know you've got Jony
Ive on the team. Is there anything you

58
00:03:13,440 --> 00:03:16,157
can tease out at this point
around what you're working on?

59
00:03:16,240 --> 00:03:17,957
>> One of the powerful things about AI is

60
00:03:18,040 --> 00:03:21,997
how much context it can handle and

61
00:03:22,080 --> 00:03:26,517
an AI that and we see a little
bit of this with uh Open uh

62
00:03:26,600 --> 00:03:30,557
an AI that can get to know more and more
about you can give you better and better

63
00:03:30,640 --> 00:03:34,557
products and services and we'd like
to build a hardware that can do that very

64
00:03:34,640 --> 00:03:36,517
well in the virtual
and the physical world.

65
00:03:36,600 --> 00:03:40,997
>> Not a smartphone? >> [laughter]
>> Um the other part of the hardware

66
00:03:41,080 --> 00:03:44,997
equation as well is is semiconductors,
an incredibly important part of the data

67
00:03:45,080 --> 00:03:47,437
center, incredibly important part
of what you do in terms of training and

68
00:03:47,520 --> 00:03:50,957
inference, etc.
Uh semiconductors is an area I believe

69
00:03:51,040 --> 00:03:54,637
you are investing in and trying
to figure out your own kind of product.

70
00:03:54,720 --> 00:03:57,917
What does that look like now? Is this
going to be something you can train on

71
00:03:58,000 --> 00:04:01,277
as well as inference? >> Um

72
00:04:01,360 --> 00:04:03,277
we we will have a rich

73
00:04:03,360 --> 00:04:06,557
portfolio of semiconductor approaches.

74
00:04:06,640 --> 00:04:10,277
We will partner with uh
you know, the sort of

75
00:04:10,360 --> 00:04:13,997
giants of the field like Nvidia.
We'll partner with newer startups in areas

76
00:04:14,080 --> 00:04:15,957
like fast inference like Cerebras. We'll

77
00:04:16,040 --> 00:04:19,517
do our own inference chip as well. Um we

78
00:04:19,600 --> 00:04:24,277
we expect this to have this
to need more of a portfolio of

79
00:04:24,360 --> 00:04:26,877
solutions than maybe we initially thought,
but we're very excited with our

80
00:04:26,960 --> 00:04:29,557
own work here. >> So, we're
not talking just one kind of

81
00:04:29,640 --> 00:04:32,797
processor like a T like Google's gone
with the TPU. We're talking potentially

82
00:04:32,880 --> 00:04:35,797
a multiple portfolio
of semiconductors out of Open AI.

83
00:04:35,880 --> 00:04:38,757
>> Yeah, I mean Google has the TPU. They
also buy Nvidia chips and I believe some

84
00:04:38,840 --> 00:04:41,117
others too and I think you
should expect the same from us.

85
00:04:41,200 --> 00:04:45,717
>> Um financing has been a big focus.
I know you're currently undergoing

86
00:04:45,800 --> 00:04:48,477
potentially more than a hundred billion
dollar round at this point. Are you

87
00:04:48,560 --> 00:04:52,357
close to closing that? >> Uh if
I were as you know, like legally I

88
00:04:52,440 --> 00:04:53,997
would not be allowed to talk about that.

89
00:04:54,080 --> 00:04:56,557
>> Uh just in terms of financing
though, there has been a lot

90
00:04:56,640 --> 00:05:00,717
of concern about the way artificial
intelligence has been being financed.

91
00:05:00,800 --> 00:05:04,477
Terms like circular financing have been
thrown around. Vendor financing as well

92
00:05:04,560 --> 00:05:09,517
giving the fact that you buy chips from
Nvidia invested in you as well. Um there

93
00:05:09,600 --> 00:05:12,877
is also concern that if there is a
failure at some point in this chain,

94
00:05:12,960 --> 00:05:15,477
this could have knock-on effects
on companies like Open AI.

95
00:05:15,560 --> 00:05:17,757
Are you concerned about that? >> I
don't agree with the first part of the

96
00:05:17,840 --> 00:05:22,477
concern. I do agree with the second part.
Um new revenue has got to come into this

97
00:05:22,560 --> 00:05:24,837
system. Like people can invest in each
other and buy each other's stuff

98
00:05:24,920 --> 00:05:28,317
whatever, but if there's not new revenue
coming in, that's when you get into

99
00:05:28,400 --> 00:05:32,757
there's like nothing here. But, revenue
is growing so fast not just for us, but

100
00:05:32,840 --> 00:05:36,437
the whole ecosystem that there there's
like a massive amount of revenue coming

101
00:05:36,520 --> 00:05:39,157
in to support all of this infrastructure.
So, I don't worry about

102
00:05:39,240 --> 00:05:43,637
that part. Um in terms of any one
layer of the stack failing would

103
00:05:43,720 --> 00:05:46,517
cause some real knock-on effects
and it's sort of like a very tight supply

104
00:05:46,600 --> 00:05:48,477
chain right now. I do worry about that.

105
00:05:48,560 --> 00:05:50,197
>> In terms of not
being able to get access

106
00:05:50,280 --> 00:05:52,637
to certain technologies, like chips?

107
00:05:52,720 --> 00:05:53,997
>> If yeah, something like that.

108
00:05:54,080 --> 00:05:58,797
>> Yeah. Uh the other part of the concern,
some have criticized the valuation

109
00:05:58,880 --> 00:06:02,117
OpenAI has got and said, "Well, you know,
there's intense competition. We've

110
00:06:02,200 --> 00:06:06,437
seen things like the deep seek moment
can can disrupt markets uh as well." Um

111
00:06:06,520 --> 00:06:10,637
your valuation,
do you think it's justified?

112
00:06:10,720 --> 00:06:13,717
>> I think so.
The revenue that that's up to other

113
00:06:13,800 --> 00:06:15,837
that's like up to a market
to decide that. I think so.

114
00:06:15,920 --> 00:06:19,477
>> to profitability eventually,
what does that look like for OpenAI?

115
00:06:19,560 --> 00:06:21,477
>> Uh

116
00:06:21,560 --> 00:06:26,797
we are growing at a extremely fast rate
uh right now. I think as long as we can

117
00:06:26,880 --> 00:06:29,437
have reasonable unit economics,
we should focus on continuing to grow

118
00:06:29,520 --> 00:06:33,917
faster and faster. Um and we'll
get profitable when we think

119
00:06:34,000 --> 00:06:35,557
we when we think it makes sense.

120
00:06:35,640 --> 00:06:37,597
>> Sam, just a very quick final one China.

121
00:06:37,680 --> 00:06:42,037
Um we had an analyst on CNBC the other
day saying that China techs tech shock

122
00:06:42,120 --> 00:06:45,517
is just getting started that in 5 to 10
years time, most of the world could be

123
00:06:45,600 --> 00:06:49,157
running on the China tech stack.
Demis Hassabis was speaking to me very

124
00:06:49,240 --> 00:06:52,917
recently saying China is just months
behind in terms of its ability to bring

125
00:06:53,000 --> 00:06:56,077
out models. Just in the past 2 weeks,
we've seen a ton of interesting models

126
00:06:56,160 --> 00:07:00,277
out of China on on video,
on physical AI. Um

127
00:07:00,360 --> 00:07:03,357
are Chinese companies' progress being
underestimated right now? Do you feel

128
00:07:03,440 --> 00:07:05,597
that they're that close as Demis says?

129
00:07:05,680 --> 00:07:10,597
>> I don't know about being
underestimated, but certainly the progress

130
00:07:10,680 --> 00:07:12,157
of Chinese tech companies across the

131
00:07:12,240 --> 00:07:14,517
entire stack and also not just in AI,

132
00:07:14,600 --> 00:07:19,237
but in many fields is remarkable. Um

133
00:07:19,320 --> 00:07:22,117
the reason I'm pushing back
on underestimated is it feels like every

134
00:07:22,200 --> 00:07:24,397
conversation I have is like, "Oh,
China's beating us. What do we do about

135
00:07:24,480 --> 00:07:27,717
it?" So, I think people are aware
of what's happening there, but uh

136
00:07:27,800 --> 00:07:31,117
yes, Chinese progress is amazingly fast.

137
00:07:31,200 --> 00:07:34,237
>> Are they are they near the frontier?
Are they are they near you?

138
00:07:34,320 --> 00:07:37,757
>> Uh in some areas, yes, and in some
areas, no. It's not quite a unilateral

139
00:07:37,840 --> 00:07:40,480
thing like that.